4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License version 2 for more details. A copy is
14 * included in the COPYING file that accompanied this code.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
23 * Copyright (c) 2013, 2016, 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_V1 0xA0629D03
45 #define LFSCK_NAMESPACE_MAGIC_V2 0xA0621A0B
46 #define LFSCK_NAMESPACE_MAGIC_V3 0xA06249FF
48 /* For Lustre-2.x (x <= 6), the namespace LFSCK used LFSCK_NAMESPACE_MAGIC_V1
49 * as the trace file magic. When downgrade to such old release, the old LFSCK
50 * will not recognize the new LFSCK_NAMESPACE_MAGIC_V2 in the new trace file,
51 * then it will reset the whole LFSCK, and will not cause start failure. The
52 * similar case will happen when upgrade from such old release. */
53 #define LFSCK_NAMESPACE_MAGIC LFSCK_NAMESPACE_MAGIC_V3
55 enum lfsck_nameentry_check {
56 LFSCK_NAMEENTRY_DEAD = 1, /* The object has been unlinked. */
57 LFSCK_NAMEENTRY_REMOVED = 2, /* The entry has been removed. */
58 LFSCK_NAMEENTRY_RECREATED = 3, /* The entry has been recreated. */
61 static struct lfsck_namespace_req *
62 lfsck_namespace_assistant_req_init(struct lfsck_instance *lfsck,
63 struct lfsck_assistant_object *lso,
64 struct lu_dirent *ent, __u16 type)
66 struct lfsck_namespace_req *lnr;
69 size = sizeof(*lnr) + (ent->lde_namelen & ~3) + 4;
72 return ERR_PTR(-ENOMEM);
74 INIT_LIST_HEAD(&lnr->lnr_lar.lar_list);
75 lnr->lnr_lar.lar_parent = lfsck_assistant_object_get(lso);
76 lnr->lnr_lmv = lfsck_lmv_get(lfsck->li_lmv);
77 lnr->lnr_fid = ent->lde_fid;
78 lnr->lnr_dir_cookie = ent->lde_hash;
79 lnr->lnr_attr = ent->lde_attrs;
82 lnr->lnr_namelen = ent->lde_namelen;
83 memcpy(lnr->lnr_name, ent->lde_name, ent->lde_namelen);
88 static void lfsck_namespace_assistant_req_fini(const struct lu_env *env,
89 struct lfsck_assistant_req *lar)
91 struct lfsck_namespace_req *lnr =
92 container_of0(lar, struct lfsck_namespace_req, lnr_lar);
94 if (lnr->lnr_lmv != NULL)
95 lfsck_lmv_put(env, lnr->lnr_lmv);
97 lfsck_assistant_object_put(env, lar->lar_parent);
98 OBD_FREE(lnr, lnr->lnr_size);
101 static void lfsck_namespace_le_to_cpu(struct lfsck_namespace *dst,
102 struct lfsck_namespace *src)
104 dst->ln_magic = le32_to_cpu(src->ln_magic);
105 dst->ln_status = le32_to_cpu(src->ln_status);
106 dst->ln_flags = le32_to_cpu(src->ln_flags);
107 dst->ln_success_count = le32_to_cpu(src->ln_success_count);
108 dst->ln_run_time_phase1 = le64_to_cpu(src->ln_run_time_phase1);
109 dst->ln_run_time_phase2 = le64_to_cpu(src->ln_run_time_phase2);
110 dst->ln_time_last_complete = le64_to_cpu(src->ln_time_last_complete);
111 dst->ln_time_latest_start = le64_to_cpu(src->ln_time_latest_start);
112 dst->ln_time_last_checkpoint =
113 le64_to_cpu(src->ln_time_last_checkpoint);
114 lfsck_position_le_to_cpu(&dst->ln_pos_latest_start,
115 &src->ln_pos_latest_start);
116 lfsck_position_le_to_cpu(&dst->ln_pos_last_checkpoint,
117 &src->ln_pos_last_checkpoint);
118 lfsck_position_le_to_cpu(&dst->ln_pos_first_inconsistent,
119 &src->ln_pos_first_inconsistent);
120 dst->ln_items_checked = le64_to_cpu(src->ln_items_checked);
121 dst->ln_items_repaired = le64_to_cpu(src->ln_items_repaired);
122 dst->ln_items_failed = le64_to_cpu(src->ln_items_failed);
123 dst->ln_dirs_checked = le64_to_cpu(src->ln_dirs_checked);
124 dst->ln_objs_checked_phase2 = le64_to_cpu(src->ln_objs_checked_phase2);
125 dst->ln_objs_repaired_phase2 =
126 le64_to_cpu(src->ln_objs_repaired_phase2);
127 dst->ln_objs_failed_phase2 = le64_to_cpu(src->ln_objs_failed_phase2);
128 dst->ln_objs_nlink_repaired = le64_to_cpu(src->ln_objs_nlink_repaired);
129 fid_le_to_cpu(&dst->ln_fid_latest_scanned_phase2,
130 &src->ln_fid_latest_scanned_phase2);
131 dst->ln_dirent_repaired = le64_to_cpu(src->ln_dirent_repaired);
132 dst->ln_linkea_repaired = le64_to_cpu(src->ln_linkea_repaired);
133 dst->ln_mul_linked_checked = le64_to_cpu(src->ln_mul_linked_checked);
134 dst->ln_mul_linked_repaired = le64_to_cpu(src->ln_mul_linked_repaired);
135 dst->ln_unknown_inconsistency =
136 le64_to_cpu(src->ln_unknown_inconsistency);
137 dst->ln_unmatched_pairs_repaired =
138 le64_to_cpu(src->ln_unmatched_pairs_repaired);
139 dst->ln_dangling_repaired = le64_to_cpu(src->ln_dangling_repaired);
140 dst->ln_mul_ref_repaired = le64_to_cpu(src->ln_mul_ref_repaired);
141 dst->ln_bad_type_repaired = le64_to_cpu(src->ln_bad_type_repaired);
142 dst->ln_lost_dirent_repaired =
143 le64_to_cpu(src->ln_lost_dirent_repaired);
144 dst->ln_striped_dirs_scanned =
145 le64_to_cpu(src->ln_striped_dirs_scanned);
146 dst->ln_striped_dirs_repaired =
147 le64_to_cpu(src->ln_striped_dirs_repaired);
148 dst->ln_striped_dirs_failed =
149 le64_to_cpu(src->ln_striped_dirs_failed);
150 dst->ln_striped_dirs_disabled =
151 le64_to_cpu(src->ln_striped_dirs_disabled);
152 dst->ln_striped_dirs_skipped =
153 le64_to_cpu(src->ln_striped_dirs_skipped);
154 dst->ln_striped_shards_scanned =
155 le64_to_cpu(src->ln_striped_shards_scanned);
156 dst->ln_striped_shards_repaired =
157 le64_to_cpu(src->ln_striped_shards_repaired);
158 dst->ln_striped_shards_failed =
159 le64_to_cpu(src->ln_striped_shards_failed);
160 dst->ln_striped_shards_skipped =
161 le64_to_cpu(src->ln_striped_shards_skipped);
162 dst->ln_name_hash_repaired = le64_to_cpu(src->ln_name_hash_repaired);
163 dst->ln_local_lpf_scanned = le64_to_cpu(src->ln_local_lpf_scanned);
164 dst->ln_local_lpf_moved = le64_to_cpu(src->ln_local_lpf_moved);
165 dst->ln_local_lpf_skipped = le64_to_cpu(src->ln_local_lpf_skipped);
166 dst->ln_local_lpf_failed = le64_to_cpu(src->ln_local_lpf_failed);
167 dst->ln_bitmap_size = le32_to_cpu(src->ln_bitmap_size);
168 dst->ln_time_latest_reset = le64_to_cpu(src->ln_time_latest_reset);
169 dst->ln_linkea_overflow_cleared =
170 le64_to_cpu(src->ln_linkea_overflow_cleared);
173 static void lfsck_namespace_cpu_to_le(struct lfsck_namespace *dst,
174 struct lfsck_namespace *src)
176 dst->ln_magic = cpu_to_le32(src->ln_magic);
177 dst->ln_status = cpu_to_le32(src->ln_status);
178 dst->ln_flags = cpu_to_le32(src->ln_flags);
179 dst->ln_success_count = cpu_to_le32(src->ln_success_count);
180 dst->ln_run_time_phase1 = cpu_to_le64(src->ln_run_time_phase1);
181 dst->ln_run_time_phase2 = cpu_to_le64(src->ln_run_time_phase2);
182 dst->ln_time_last_complete = cpu_to_le64(src->ln_time_last_complete);
183 dst->ln_time_latest_start = cpu_to_le64(src->ln_time_latest_start);
184 dst->ln_time_last_checkpoint =
185 cpu_to_le64(src->ln_time_last_checkpoint);
186 lfsck_position_cpu_to_le(&dst->ln_pos_latest_start,
187 &src->ln_pos_latest_start);
188 lfsck_position_cpu_to_le(&dst->ln_pos_last_checkpoint,
189 &src->ln_pos_last_checkpoint);
190 lfsck_position_cpu_to_le(&dst->ln_pos_first_inconsistent,
191 &src->ln_pos_first_inconsistent);
192 dst->ln_items_checked = cpu_to_le64(src->ln_items_checked);
193 dst->ln_items_repaired = cpu_to_le64(src->ln_items_repaired);
194 dst->ln_items_failed = cpu_to_le64(src->ln_items_failed);
195 dst->ln_dirs_checked = cpu_to_le64(src->ln_dirs_checked);
196 dst->ln_objs_checked_phase2 = cpu_to_le64(src->ln_objs_checked_phase2);
197 dst->ln_objs_repaired_phase2 =
198 cpu_to_le64(src->ln_objs_repaired_phase2);
199 dst->ln_objs_failed_phase2 = cpu_to_le64(src->ln_objs_failed_phase2);
200 dst->ln_objs_nlink_repaired = cpu_to_le64(src->ln_objs_nlink_repaired);
201 fid_cpu_to_le(&dst->ln_fid_latest_scanned_phase2,
202 &src->ln_fid_latest_scanned_phase2);
203 dst->ln_dirent_repaired = cpu_to_le64(src->ln_dirent_repaired);
204 dst->ln_linkea_repaired = cpu_to_le64(src->ln_linkea_repaired);
205 dst->ln_mul_linked_checked = cpu_to_le64(src->ln_mul_linked_checked);
206 dst->ln_mul_linked_repaired = cpu_to_le64(src->ln_mul_linked_repaired);
207 dst->ln_unknown_inconsistency =
208 cpu_to_le64(src->ln_unknown_inconsistency);
209 dst->ln_unmatched_pairs_repaired =
210 cpu_to_le64(src->ln_unmatched_pairs_repaired);
211 dst->ln_dangling_repaired = cpu_to_le64(src->ln_dangling_repaired);
212 dst->ln_mul_ref_repaired = cpu_to_le64(src->ln_mul_ref_repaired);
213 dst->ln_bad_type_repaired = cpu_to_le64(src->ln_bad_type_repaired);
214 dst->ln_lost_dirent_repaired =
215 cpu_to_le64(src->ln_lost_dirent_repaired);
216 dst->ln_striped_dirs_scanned =
217 cpu_to_le64(src->ln_striped_dirs_scanned);
218 dst->ln_striped_dirs_repaired =
219 cpu_to_le64(src->ln_striped_dirs_repaired);
220 dst->ln_striped_dirs_failed =
221 cpu_to_le64(src->ln_striped_dirs_failed);
222 dst->ln_striped_dirs_disabled =
223 cpu_to_le64(src->ln_striped_dirs_disabled);
224 dst->ln_striped_dirs_skipped =
225 cpu_to_le64(src->ln_striped_dirs_skipped);
226 dst->ln_striped_shards_scanned =
227 cpu_to_le64(src->ln_striped_shards_scanned);
228 dst->ln_striped_shards_repaired =
229 cpu_to_le64(src->ln_striped_shards_repaired);
230 dst->ln_striped_shards_failed =
231 cpu_to_le64(src->ln_striped_shards_failed);
232 dst->ln_striped_shards_skipped =
233 cpu_to_le64(src->ln_striped_shards_skipped);
234 dst->ln_name_hash_repaired = cpu_to_le64(src->ln_name_hash_repaired);
235 dst->ln_local_lpf_scanned = cpu_to_le64(src->ln_local_lpf_scanned);
236 dst->ln_local_lpf_moved = cpu_to_le64(src->ln_local_lpf_moved);
237 dst->ln_local_lpf_skipped = cpu_to_le64(src->ln_local_lpf_skipped);
238 dst->ln_local_lpf_failed = cpu_to_le64(src->ln_local_lpf_failed);
239 dst->ln_bitmap_size = cpu_to_le32(src->ln_bitmap_size);
240 dst->ln_time_latest_reset = cpu_to_le64(src->ln_time_latest_reset);
241 dst->ln_linkea_overflow_cleared =
242 cpu_to_le64(src->ln_linkea_overflow_cleared);
245 static void lfsck_namespace_record_failure(const struct lu_env *env,
246 struct lfsck_instance *lfsck,
247 struct lfsck_namespace *ns)
249 struct lfsck_position pos;
251 ns->ln_items_failed++;
252 lfsck_pos_fill(env, lfsck, &pos, false);
253 if (lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent) ||
254 lfsck_pos_is_eq(&pos, &ns->ln_pos_first_inconsistent) < 0) {
255 ns->ln_pos_first_inconsistent = pos;
257 CDEBUG(D_LFSCK, "%s: namespace LFSCK hit first non-repaired "
258 "inconsistency at the pos [%llu, "DFID", %#llx]\n",
259 lfsck_lfsck2name(lfsck),
260 ns->ln_pos_first_inconsistent.lp_oit_cookie,
261 PFID(&ns->ln_pos_first_inconsistent.lp_dir_parent),
262 ns->ln_pos_first_inconsistent.lp_dir_cookie);
267 * Load the MDT bitmap from the lfsck_namespace trace file.
269 * \param[in] env pointer to the thread context
270 * \param[in] com pointer to the lfsck component
272 * \retval 0 for success
273 * \retval negative error number on failure or data corruption
275 static int lfsck_namespace_load_bitmap(const struct lu_env *env,
276 struct lfsck_component *com)
278 struct dt_object *obj = com->lc_obj;
279 struct lfsck_assistant_data *lad = com->lc_data;
280 struct lfsck_namespace *ns = com->lc_file_ram;
281 struct cfs_bitmap *bitmap = lad->lad_bitmap;
287 if (com->lc_lfsck->li_mdt_descs.ltd_tgts_bitmap->size >
289 nbits = com->lc_lfsck->li_mdt_descs.ltd_tgts_bitmap->size;
291 nbits = ns->ln_bitmap_size;
293 if (unlikely(nbits < BITS_PER_LONG))
294 nbits = BITS_PER_LONG;
296 if (nbits > bitmap->size) {
297 __u32 new_bits = bitmap->size;
298 struct cfs_bitmap *new_bitmap;
300 while (new_bits < nbits)
303 new_bitmap = CFS_ALLOCATE_BITMAP(new_bits);
304 if (new_bitmap == NULL)
307 lad->lad_bitmap = new_bitmap;
308 CFS_FREE_BITMAP(bitmap);
312 if (ns->ln_bitmap_size == 0) {
313 lad->lad_incomplete = 0;
314 CFS_RESET_BITMAP(bitmap);
319 size = (ns->ln_bitmap_size + 7) >> 3;
320 rc = dt_xattr_get(env, obj,
321 lfsck_buf_get(env, bitmap->data, size),
322 XATTR_NAME_LFSCK_BITMAP);
324 RETURN(rc >= 0 ? -EINVAL : rc);
326 if (cfs_bitmap_check_empty(bitmap))
327 lad->lad_incomplete = 0;
329 lad->lad_incomplete = 1;
335 * Load namespace LFSCK statistics information from the trace file.
337 * \param[in] env pointer to the thread context
338 * \param[in] com pointer to the lfsck component
340 * \retval 0 for success
341 * \retval negative error number on failure
343 static int lfsck_namespace_load(const struct lu_env *env,
344 struct lfsck_component *com)
346 int len = com->lc_file_size;
349 rc = dt_xattr_get(env, com->lc_obj,
350 lfsck_buf_get(env, com->lc_file_disk, len),
351 XATTR_NAME_LFSCK_NAMESPACE);
353 struct lfsck_namespace *ns = com->lc_file_ram;
355 lfsck_namespace_le_to_cpu(ns,
356 (struct lfsck_namespace *)com->lc_file_disk);
357 if (ns->ln_magic != LFSCK_NAMESPACE_MAGIC) {
358 CDEBUG(D_LFSCK, "%s: invalid lfsck_namespace magic "
359 "%#x != %#x\n", lfsck_lfsck2name(com->lc_lfsck),
360 ns->ln_magic, LFSCK_NAMESPACE_MAGIC);
365 } else if (rc != -ENODATA) {
366 CDEBUG(D_LFSCK, "%s: fail to load lfsck_namespace, "
367 "expected = %d: rc = %d\n",
368 lfsck_lfsck2name(com->lc_lfsck), len, rc);
376 static int lfsck_namespace_store(const struct lu_env *env,
377 struct lfsck_component *com)
379 struct dt_object *obj = com->lc_obj;
380 struct lfsck_instance *lfsck = com->lc_lfsck;
381 struct lfsck_namespace *ns = com->lc_file_ram;
382 struct lfsck_assistant_data *lad = com->lc_data;
383 struct dt_device *dev = lfsck_obj2dev(obj);
384 struct cfs_bitmap *bitmap = NULL;
385 struct thandle *handle;
387 int len = com->lc_file_size;
392 bitmap = lad->lad_bitmap;
393 nbits = bitmap->size;
396 LASSERTF((nbits & 7) == 0, "Invalid nbits %u\n", nbits);
399 ns->ln_bitmap_size = nbits;
400 lfsck_namespace_cpu_to_le((struct lfsck_namespace *)com->lc_file_disk,
402 handle = dt_trans_create(env, dev);
404 GOTO(log, rc = PTR_ERR(handle));
406 rc = dt_declare_xattr_set(env, obj,
407 lfsck_buf_get(env, com->lc_file_disk, len),
408 XATTR_NAME_LFSCK_NAMESPACE, 0, handle);
412 if (bitmap != NULL) {
413 rc = dt_declare_xattr_set(env, obj,
414 lfsck_buf_get(env, bitmap->data, nbits >> 3),
415 XATTR_NAME_LFSCK_BITMAP, 0, handle);
420 rc = dt_trans_start_local(env, dev, handle);
424 rc = dt_xattr_set(env, obj,
425 lfsck_buf_get(env, com->lc_file_disk, len),
426 XATTR_NAME_LFSCK_NAMESPACE, 0, handle);
427 if (rc == 0 && bitmap != NULL)
428 rc = dt_xattr_set(env, obj,
429 lfsck_buf_get(env, bitmap->data, nbits >> 3),
430 XATTR_NAME_LFSCK_BITMAP, 0, handle);
435 dt_trans_stop(env, dev, handle);
439 CDEBUG(D_LFSCK, "%s: fail to store lfsck_namespace: rc = %d\n",
440 lfsck_lfsck2name(lfsck), rc);
444 static int lfsck_namespace_init(const struct lu_env *env,
445 struct lfsck_component *com)
447 struct lfsck_namespace *ns = com->lc_file_ram;
450 memset(ns, 0, sizeof(*ns));
451 ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
452 ns->ln_status = LS_INIT;
453 ns->ln_time_latest_reset = ktime_get_real_seconds();
454 down_write(&com->lc_sem);
455 rc = lfsck_namespace_store(env, com);
457 rc = lfsck_load_sub_trace_files(env, com,
458 &dt_lfsck_namespace_features, LFSCK_NAMESPACE, true);
459 up_write(&com->lc_sem);
465 * Update the namespace LFSCK trace file for the given @fid
467 * \param[in] env pointer to the thread context
468 * \param[in] com pointer to the lfsck component
469 * \param[in] fid the fid which flags to be updated in the lfsck
471 * \param[in] add true if add new flags, otherwise remove flags
473 * \retval 0 for success or nothing to be done
474 * \retval negative error number on failure
476 int lfsck_namespace_trace_update(const struct lu_env *env,
477 struct lfsck_component *com,
478 const struct lu_fid *fid,
479 const __u8 flags, bool add)
481 struct lfsck_instance *lfsck = com->lc_lfsck;
482 struct dt_object *obj;
483 struct lu_fid *key = &lfsck_env_info(env)->lti_fid3;
484 struct dt_device *dev;
485 struct thandle *th = NULL;
494 if (unlikely(!fid_is_sane(fid)))
497 idx = lfsck_sub_trace_file_fid2idx(fid);
498 mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
499 obj = com->lc_sub_trace_objs[idx].lsto_obj;
500 if (unlikely(obj == NULL)) {
501 mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
505 lfsck_object_get(obj);
506 dev = lfsck_obj2dev(obj);
507 fid_cpu_to_be(key, fid);
508 rc = dt_lookup(env, obj, (struct dt_rec *)&old,
509 (const struct dt_key *)key);
512 GOTO(unlock, rc = 0);
516 } else if (rc == 0) {
518 if ((old & flags) == flags)
519 GOTO(unlock, rc = 0);
523 if ((old & flags) == 0)
524 GOTO(unlock, rc = 0);
532 th = dt_trans_create(env, dev);
534 GOTO(log, rc = PTR_ERR(th));
537 rc = dt_declare_delete(env, obj,
538 (const struct dt_key *)key, th);
544 rc = dt_declare_insert(env, obj,
545 (const struct dt_rec *)&new,
546 (const struct dt_key *)key, th);
551 rc = dt_trans_start_local(env, dev, th);
556 rc = dt_delete(env, obj, (const struct dt_key *)key, th);
562 rc = dt_insert(env, obj, (const struct dt_rec *)&new,
563 (const struct dt_key *)key, th, 1);
571 if (th != NULL && !IS_ERR(th))
572 dt_trans_stop(env, dev, th);
574 CDEBUG(D_LFSCK, "%s: namespace LFSCK %s flags for "DFID" in the "
575 "trace file, flags %x, old %x, new %x: rc = %d\n",
576 lfsck_lfsck2name(lfsck), add ? "add" : "del", PFID(fid),
577 (__u32)flags, (__u32)old, (__u32)new, rc);
580 mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
581 lfsck_object_put(env, obj);
586 int lfsck_namespace_check_exist(const struct lu_env *env,
587 struct dt_object *dir,
588 struct dt_object *obj, const char *name)
590 struct lu_fid *fid = &lfsck_env_info(env)->lti_fid;
594 if (unlikely(lfsck_is_dead_obj(obj)))
595 RETURN(LFSCK_NAMEENTRY_DEAD);
597 rc = dt_lookup(env, dir, (struct dt_rec *)fid,
598 (const struct dt_key *)name);
600 RETURN(LFSCK_NAMEENTRY_REMOVED);
605 if (!lu_fid_eq(fid, lfsck_dto2fid(obj)))
606 RETURN(LFSCK_NAMEENTRY_RECREATED);
611 static int lfsck_declare_namespace_exec_dir(const struct lu_env *env,
612 struct dt_object *obj,
613 struct thandle *handle)
617 /* For remote updating LINKEA, there may be further LFSCK action
618 * on remote MDT after the updating, so update the LINKEA ASAP. */
619 if (dt_object_remote(obj))
622 /* For destroying all invalid linkEA entries. */
623 rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, handle);
625 /* For insert new linkEA entry. */
626 rc = dt_declare_xattr_set(env, obj,
627 lfsck_buf_get_const(env, NULL, MAX_LINKEA_SIZE),
628 XATTR_NAME_LINK, 0, handle);
632 int __lfsck_links_read(const struct lu_env *env, struct dt_object *obj,
633 struct linkea_data *ldata, bool with_rec)
637 if (ldata->ld_buf->lb_buf == NULL)
640 if (!dt_object_exists(obj))
643 rc = dt_xattr_get(env, obj, ldata->ld_buf, XATTR_NAME_LINK);
645 /* Buf was too small, figure out what we need. */
646 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LINK);
647 if (unlikely(rc == 0))
653 lu_buf_realloc(ldata->ld_buf, rc);
654 if (ldata->ld_buf->lb_buf == NULL)
657 rc = dt_xattr_get(env, obj, ldata->ld_buf, XATTR_NAME_LINK);
660 if (unlikely(rc == 0))
665 rc = linkea_init_with_rec(ldata);
667 rc = linkea_init(ldata);
674 * Remove linkEA for the given object.
676 * The caller should take the ldlm lock before the calling.
678 * \param[in] env pointer to the thread context
679 * \param[in] com pointer to the lfsck component
680 * \param[in] obj pointer to the dt_object to be handled
682 * \retval 0 for repaired cases
683 * \retval negative error number on failure
685 static int lfsck_namespace_links_remove(const struct lu_env *env,
686 struct lfsck_component *com,
687 struct dt_object *obj)
689 struct lfsck_instance *lfsck = com->lc_lfsck;
690 struct dt_device *dev = lfsck_obj2dev(obj);
691 struct thandle *th = NULL;
695 LASSERT(dt_object_remote(obj) == 0);
697 th = dt_trans_create(env, dev);
699 GOTO(log, rc = PTR_ERR(th));
701 rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, th);
705 rc = dt_trans_start_local(env, dev, th);
709 dt_write_lock(env, obj, 0);
710 if (unlikely(lfsck_is_dead_obj(obj)))
711 GOTO(unlock, rc = -ENOENT);
713 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
714 GOTO(unlock, rc = 0);
716 rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, th);
721 dt_write_unlock(env, obj);
724 dt_trans_stop(env, dev, th);
727 CDEBUG(D_LFSCK, "%s: namespace LFSCK remove invalid linkEA "
728 "for the object "DFID": rc = %d\n",
729 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
732 struct lfsck_namespace *ns = com->lc_file_ram;
734 ns->ln_flags |= LF_INCONSISTENT;
740 static int lfsck_links_write(const struct lu_env *env, struct dt_object *obj,
741 struct linkea_data *ldata, struct thandle *handle)
746 lfsck_buf_init(&buf, ldata->ld_buf->lb_buf, ldata->ld_leh->leh_len);
749 rc = dt_xattr_set(env, obj, &buf, XATTR_NAME_LINK, 0, handle);
750 if (unlikely(rc == -ENOSPC)) {
751 rc = linkea_overflow_shrink(ldata);
752 if (likely(rc > 0)) {
761 static int lfsck_namespace_unpack_linkea_entry(struct linkea_data *ldata,
762 struct lu_name *cname,
764 char *buf, const int buflen)
766 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, cname, pfid);
767 if (unlikely(ldata->ld_reclen <= 0 ||
768 ldata->ld_reclen + sizeof(struct link_ea_header) >
769 ldata->ld_leh->leh_len ||
770 cname->ln_namelen <= 0 ||
771 cname->ln_namelen > NAME_MAX ||
772 cname->ln_namelen >= buflen ||
776 /* To guarantee the 'name' is terminated with '0'. */
777 memcpy(buf, cname->ln_name, cname->ln_namelen);
778 buf[cname->ln_namelen] = 0;
779 cname->ln_name = buf;
784 static void lfsck_linkea_del_buf(struct linkea_data *ldata,
785 const struct lu_name *lname)
787 LASSERT(ldata->ld_leh != NULL && ldata->ld_lee != NULL);
789 /* If current record is corrupted, all the subsequent
790 * records will be dropped. */
791 if (unlikely(ldata->ld_reclen <= 0 ||
792 ldata->ld_reclen + sizeof(struct link_ea_header) >
793 ldata->ld_leh->leh_len)) {
794 void *ptr = ldata->ld_lee;
796 ldata->ld_leh->leh_len = sizeof(struct link_ea_header);
797 ldata->ld_leh->leh_reccount = 0;
798 linkea_first_entry(ldata);
799 while (ldata->ld_lee != NULL &&
800 (char *)ldata->ld_lee < (char *)ptr) {
801 int reclen = (ldata->ld_lee->lee_reclen[0] << 8) |
802 ldata->ld_lee->lee_reclen[1];
804 ldata->ld_leh->leh_len += reclen;
805 ldata->ld_leh->leh_reccount++;
806 ldata->ld_lee = (struct link_ea_entry *)
807 ((char *)ldata->ld_lee + reclen);
810 ldata->ld_lee = NULL;
812 linkea_del_buf(ldata, lname);
816 static int lfsck_namespace_filter_linkea_entry(struct linkea_data *ldata,
817 struct lu_name *cname,
821 struct link_ea_entry *oldlee;
825 oldlee = ldata->ld_lee;
826 oldlen = ldata->ld_reclen;
827 linkea_next_entry(ldata);
828 while (ldata->ld_lee != NULL) {
829 ldata->ld_reclen = (ldata->ld_lee->lee_reclen[0] << 8) |
830 ldata->ld_lee->lee_reclen[1];
831 if (unlikely(ldata->ld_reclen == oldlen &&
832 memcmp(ldata->ld_lee, oldlee, oldlen) == 0)) {
837 lfsck_linkea_del_buf(ldata, cname);
839 linkea_next_entry(ldata);
842 ldata->ld_lee = oldlee;
843 ldata->ld_reclen = oldlen;
849 * Insert orphan into .lustre/lost+found/MDTxxxx/ locally.
851 * Add the specified orphan MDT-object to the .lustre/lost+found/MDTxxxx/
852 * with the given type to generate the name, the detailed rules for name
853 * have been described as following.
855 * The function also generates the linkEA corresponding to the name entry
856 * under the .lustre/lost+found/MDTxxxx/ for the orphan MDT-object.
858 * \param[in] env pointer to the thread context
859 * \param[in] com pointer to the lfsck component
860 * \param[in] orphan pointer to the orphan MDT-object
861 * \param[in] infix additional information for the orphan name, such as
862 * the FID for original
863 * \param[in] type the type for describing why the orphan MDT-object is
864 * created. The rules are as following:
866 * type "D": The MDT-object is a directory, it may knows its parent
867 * but because there is no valid linkEA, the LFSCK cannot
868 * know where to put it back to the namespace.
869 * type "O": The MDT-object has no linkEA, and there is no name
870 * entry that references the MDT-object.
872 * type "S": The orphan MDT-object is a shard of a striped directory
874 * \see lfsck_layout_recreate_parent() for more types.
876 * The orphan name will be like:
877 * ${FID}-${infix}-${type}-${conflict_version}
879 * \param[out] count if some others inserted some linkEA entries by race,
880 * then return the linkEA entries count.
882 * \retval positive number for repaired cases
883 * \retval 0 if needs to repair nothing
884 * \retval negative error number on failure
886 static int lfsck_namespace_insert_orphan(const struct lu_env *env,
887 struct lfsck_component *com,
888 struct dt_object *orphan,
889 const char *infix, const char *type,
892 struct lfsck_thread_info *info = lfsck_env_info(env);
893 struct lu_name *cname = &info->lti_name;
894 struct dt_insert_rec *rec = &info->lti_dt_rec;
895 struct lu_attr *la = &info->lti_la2;
896 const struct lu_fid *cfid = lfsck_dto2fid(orphan);
897 const struct lu_fid *pfid;
899 struct lfsck_instance *lfsck = com->lc_lfsck;
900 struct dt_device *dev = lfsck_obj2dev(orphan);
901 struct dt_object *parent;
902 struct thandle *th = NULL;
903 struct lfsck_lock_handle *pllh = &info->lti_llh;
904 struct lustre_handle clh = { 0 };
905 struct linkea_data ldata2 = { NULL };
906 struct lu_buf linkea_buf;
913 cname->ln_name = NULL;
914 if (unlikely(lfsck->li_lpf_obj == NULL))
915 GOTO(log, rc = -ENXIO);
917 parent = lfsck->li_lpf_obj;
918 pfid = lfsck_dto2fid(parent);
922 namelen = snprintf(info->lti_key, NAME_MAX, DFID"%s-%s-%d",
923 PFID(cfid), infix, type, idx++);
924 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
925 (const struct dt_key *)info->lti_key);
926 if (rc != 0 && rc != -ENOENT)
929 if (unlikely(rc == 0 && lu_fid_eq(cfid, &tfid)))
931 } while (rc == 0 && !exist);
933 rc = lfsck_lock(env, lfsck, parent, info->lti_key, pllh,
934 MDS_INODELOCK_UPDATE, LCK_PW);
938 /* Re-check whether the name conflict with othrs after taken
940 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
941 (const struct dt_key *)info->lti_key);
943 if (!lu_fid_eq(cfid, &tfid)) {
950 } else if (rc != -ENOENT) {
956 cname->ln_name = info->lti_key;
957 cname->ln_namelen = namelen;
958 rc = linkea_links_new(&ldata2, &info->lti_linkea_buf2,
963 rc = lfsck_ibits_lock(env, lfsck, orphan, &clh,
964 MDS_INODELOCK_UPDATE | MDS_INODELOCK_LOOKUP |
965 MDS_INODELOCK_XATTR, LCK_EX);
969 lfsck_buf_init(&linkea_buf, ldata2.ld_buf->lb_buf,
970 ldata2.ld_leh->leh_len);
971 th = dt_trans_create(env, dev);
973 GOTO(log, rc = PTR_ERR(th));
975 if (S_ISDIR(lfsck_object_type(orphan))) {
976 rc = dt_declare_delete(env, orphan,
977 (const struct dt_key *)dotdot, th);
981 rec->rec_type = S_IFDIR;
983 rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
984 (const struct dt_key *)dotdot, th);
989 rc = dt_declare_xattr_set(env, orphan, &linkea_buf,
990 XATTR_NAME_LINK, 0, th);
995 rec->rec_type = lfsck_object_type(orphan) & S_IFMT;
997 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
998 (const struct dt_key *)cname->ln_name,
1003 if (S_ISDIR(rec->rec_type)) {
1004 rc = dt_declare_ref_add(env, parent, th);
1010 memset(la, 0, sizeof(*la));
1011 la->la_ctime = ktime_get_real_seconds();
1012 la->la_valid = LA_CTIME;
1013 rc = dt_declare_attr_set(env, orphan, la, th);
1017 rc = dt_trans_start_local(env, dev, th);
1021 dt_write_lock(env, orphan, 0);
1022 rc = lfsck_links_read2_with_rec(env, orphan, &ldata2);
1023 if (likely(rc == -ENODATA || rc == -EINVAL)) {
1024 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1025 GOTO(unlock, rc = 1);
1027 if (S_ISDIR(lfsck_object_type(orphan))) {
1028 rc = dt_delete(env, orphan,
1029 (const struct dt_key *)dotdot, th);
1033 rec->rec_type = S_IFDIR;
1034 rec->rec_fid = pfid;
1035 rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1036 (const struct dt_key *)dotdot, th, 1);
1041 rc = dt_xattr_set(env, orphan, &linkea_buf, XATTR_NAME_LINK, 0,
1044 if (rc == 0 && count != NULL)
1045 *count = ldata2.ld_leh->leh_reccount;
1049 dt_write_unlock(env, orphan);
1051 if (rc == 0 && !exist) {
1052 rec->rec_type = lfsck_object_type(orphan) & S_IFMT;
1053 rec->rec_fid = cfid;
1054 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1055 (const struct dt_key *)cname->ln_name, th, 1);
1056 if (rc == 0 && S_ISDIR(rec->rec_type)) {
1057 dt_write_lock(env, parent, 0);
1058 rc = dt_ref_add(env, parent, th);
1059 dt_write_unlock(env, parent);
1064 rc = dt_attr_set(env, orphan, la, th);
1066 GOTO(stop, rc = (rc == 0 ? 1 : rc));
1069 dt_write_unlock(env, orphan);
1072 dt_trans_stop(env, dev, th);
1075 lfsck_ibits_unlock(&clh, LCK_EX);
1077 CDEBUG(D_LFSCK, "%s: namespace LFSCK insert orphan for the "
1078 "object "DFID", name = %s: rc = %d\n",
1079 lfsck_lfsck2name(lfsck), PFID(cfid),
1080 cname->ln_name != NULL ? cname->ln_name : "<NULL>", rc);
1083 struct lfsck_namespace *ns = com->lc_file_ram;
1085 ns->ln_flags |= LF_INCONSISTENT;
1092 * Add the specified name entry back to namespace.
1094 * If there is a linkEA entry that back references a name entry under
1095 * some parent directory, but such parent directory does not have the
1096 * claimed name entry. On the other hand, the linkEA entries count is
1097 * not larger than the MDT-object's hard link count. Under such case,
1098 * it is quite possible that the name entry is lost. Then the LFSCK
1099 * should add the name entry back to the namespace.
1101 * \param[in] env pointer to the thread context
1102 * \param[in] com pointer to the lfsck component
1103 * \param[in] parent pointer to the directory under which the name entry
1104 * will be inserted into
1105 * \param[in] child pointer to the object referenced by the name entry
1106 * that to be inserted into the parent
1107 * \param[in] name the name for the child in the parent directory
1109 * \retval positive number for repaired cases
1110 * \retval 0 if nothing to be repaired
1111 * \retval negative error number on failure
1113 static int lfsck_namespace_insert_normal(const struct lu_env *env,
1114 struct lfsck_component *com,
1115 struct dt_object *parent,
1116 struct dt_object *child,
1119 struct lfsck_thread_info *info = lfsck_env_info(env);
1120 struct lu_attr *la = &info->lti_la;
1121 struct dt_insert_rec *rec = &info->lti_dt_rec;
1122 struct lfsck_instance *lfsck = com->lc_lfsck;
1123 /* The child and its name may be on different MDTs. */
1124 const struct lu_fid *pfid = lfsck_dto2fid(parent);
1125 const struct lu_fid *cfid = lfsck_dto2fid(child);
1126 struct dt_device *dev = lfsck->li_next;
1127 struct thandle *th = NULL;
1128 struct lfsck_lock_handle *llh = &info->lti_llh;
1132 /* @parent/@child may be based on lfsck->li_bottom,
1133 * but here we need the object based on the lfsck->li_next. */
1135 parent = lfsck_object_locate(dev, parent);
1137 GOTO(log, rc = PTR_ERR(parent));
1139 if (unlikely(!dt_try_as_dir(env, parent)))
1140 GOTO(log, rc = -ENOTDIR);
1142 child = lfsck_object_locate(dev, child);
1144 GOTO(log, rc = PTR_ERR(child));
1146 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1149 rc = lfsck_lock(env, lfsck, parent, name, llh,
1150 MDS_INODELOCK_UPDATE, LCK_PW);
1154 th = dt_trans_create(env, dev);
1156 GOTO(unlock, rc = PTR_ERR(th));
1158 rec->rec_type = lfsck_object_type(child) & S_IFMT;
1159 rec->rec_fid = cfid;
1160 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1161 (const struct dt_key *)name, th);
1165 if (S_ISDIR(rec->rec_type)) {
1166 rc = dt_declare_ref_add(env, parent, th);
1171 memset(la, 0, sizeof(*la));
1172 la->la_ctime = ktime_get_real_seconds();
1173 la->la_valid = LA_CTIME;
1174 rc = dt_declare_attr_set(env, parent, la, th);
1178 rc = dt_declare_attr_set(env, child, la, th);
1182 rc = dt_trans_start_local(env, dev, th);
1186 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1187 (const struct dt_key *)name, th, 1);
1191 if (S_ISDIR(rec->rec_type)) {
1192 dt_write_lock(env, parent, 0);
1193 rc = dt_ref_add(env, parent, th);
1194 dt_write_unlock(env, parent);
1199 la->la_ctime = ktime_get_real_seconds();
1200 rc = dt_attr_set(env, parent, la, th);
1204 rc = dt_attr_set(env, child, la, th);
1206 GOTO(stop, rc = (rc == 0 ? 1 : rc));
1209 dt_trans_stop(env, dev, th);
1215 CDEBUG(D_LFSCK, "%s: namespace LFSCK insert object "DFID" with "
1216 "the name %s and type %o to the parent "DFID": rc = %d\n",
1217 lfsck_lfsck2name(lfsck), PFID(cfid), name,
1218 lfsck_object_type(child) & S_IFMT, PFID(pfid), rc);
1221 struct lfsck_namespace *ns = com->lc_file_ram;
1223 ns->ln_flags |= LF_INCONSISTENT;
1225 ns->ln_lost_dirent_repaired++;
1232 * Create the specified orphan directory.
1234 * For the case that the parent MDT-object stored in some MDT-object's
1235 * linkEA entry is lost, the LFSCK will re-create the parent object as
1236 * an orphan and insert it into .lustre/lost+found/MDTxxxx/ directory
1237 * with the name ${FID}-P-${conflict_version}.
1239 * \param[in] env pointer to the thread context
1240 * \param[in] com pointer to the lfsck component
1241 * \param[in] orphan pointer to the orphan MDT-object to be created
1242 * \param[in] lmv pointer to master LMV EA that will be set to the orphan
1244 * \retval positive number for repaired cases
1245 * \retval negative error number on failure
1247 static int lfsck_namespace_create_orphan_dir(const struct lu_env *env,
1248 struct lfsck_component *com,
1249 struct dt_object *orphan,
1250 struct lmv_mds_md_v1 *lmv)
1252 struct lfsck_thread_info *info = lfsck_env_info(env);
1253 struct lu_attr *la = &info->lti_la;
1254 struct dt_allocation_hint *hint = &info->lti_hint;
1255 struct dt_object_format *dof = &info->lti_dof;
1256 struct lu_name *cname = &info->lti_name2;
1257 struct dt_insert_rec *rec = &info->lti_dt_rec;
1258 struct lmv_mds_md_v1 *lmv2 = &info->lti_lmv2;
1259 const struct lu_fid *cfid = lfsck_dto2fid(orphan);
1261 struct lfsck_instance *lfsck = com->lc_lfsck;
1262 struct lfsck_namespace *ns = com->lc_file_ram;
1263 struct dt_device *dev = lfsck_obj2dev(orphan);
1264 struct dt_object *parent = NULL;
1265 struct thandle *th = NULL;
1266 struct lfsck_lock_handle *llh = &info->lti_llh;
1267 struct linkea_data ldata = { NULL };
1268 struct lu_buf linkea_buf;
1269 struct lu_buf lmv_buf;
1277 LASSERT(!dt_object_exists(orphan));
1279 cname->ln_name = NULL;
1280 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1283 if (dt_object_remote(orphan)) {
1284 LASSERT(lfsck->li_lpf_root_obj != NULL);
1286 idx = lfsck_find_mdt_idx_by_fid(env, lfsck, cfid);
1288 GOTO(log, rc = idx);
1290 snprintf(name, 8, "MDT%04x", idx);
1291 rc = dt_lookup(env, lfsck->li_lpf_root_obj,
1292 (struct dt_rec *)&tfid,
1293 (const struct dt_key *)name);
1295 GOTO(log, rc = (rc == -ENOENT ? -ENXIO : rc));
1297 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
1299 GOTO(log, rc = PTR_ERR(parent));
1301 if (unlikely(!dt_try_as_dir(env, parent)))
1302 GOTO(log, rc = -ENOTDIR);
1304 if (unlikely(lfsck->li_lpf_obj == NULL))
1305 GOTO(log, rc = -ENXIO);
1307 parent = lfsck->li_lpf_obj;
1310 dev = lfsck_find_dev_by_fid(env, lfsck, cfid);
1312 GOTO(log, rc = PTR_ERR(dev));
1318 namelen = snprintf(name, 31, DFID"-P-%d",
1320 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
1321 (const struct dt_key *)name);
1322 if (rc != 0 && rc != -ENOENT)
1326 rc = lfsck_lock(env, lfsck, parent, name, llh,
1327 MDS_INODELOCK_UPDATE, LCK_PW);
1331 /* Re-check whether the name conflict with othrs after taken
1333 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
1334 (const struct dt_key *)name);
1335 if (unlikely(rc == 0)) {
1343 cname->ln_name = name;
1344 cname->ln_namelen = namelen;
1346 memset(la, 0, sizeof(*la));
1347 la->la_mode = S_IFDIR | 0700;
1348 la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
1349 LA_ATIME | LA_MTIME | LA_CTIME;
1351 orphan->do_ops->do_ah_init(env, hint, parent, orphan,
1352 la->la_mode & S_IFMT);
1354 memset(dof, 0, sizeof(*dof));
1355 dof->dof_type = dt_mode_to_dft(S_IFDIR);
1357 rc = linkea_links_new(&ldata, &info->lti_linkea_buf2,
1358 cname, lfsck_dto2fid(parent));
1362 th = dt_trans_create(env, dev);
1364 GOTO(unlock1, rc = PTR_ERR(th));
1366 /* Sync the remote transaction to guarantee that the subsequent
1367 * lock against the @orphan can find the @orphan in time. */
1368 if (dt_object_remote(orphan))
1371 rc = dt_declare_create(env, orphan, la, hint, dof, th);
1375 if (unlikely(!dt_try_as_dir(env, orphan)))
1376 GOTO(stop, rc = -ENOTDIR);
1378 rc = dt_declare_ref_add(env, orphan, th);
1382 rec->rec_type = S_IFDIR;
1383 rec->rec_fid = cfid;
1384 rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
1385 (const struct dt_key *)dot, th);
1389 rec->rec_fid = lfsck_dto2fid(parent);
1390 rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
1391 (const struct dt_key *)dotdot, th);
1396 lmv->lmv_magic = LMV_MAGIC;
1397 lmv->lmv_master_mdt_index = lfsck_dev_idx(lfsck);
1398 lfsck_lmv_header_cpu_to_le(lmv2, lmv);
1399 lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
1400 rc = dt_declare_xattr_set(env, orphan, &lmv_buf,
1401 XATTR_NAME_LMV, 0, th);
1406 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
1407 ldata.ld_leh->leh_len);
1408 rc = dt_declare_xattr_set(env, orphan, &linkea_buf,
1409 XATTR_NAME_LINK, 0, th);
1413 rec->rec_fid = cfid;
1414 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1415 (const struct dt_key *)name, th);
1417 rc = dt_declare_ref_add(env, parent, th);
1422 rc = dt_trans_start_local(env, dev, th);
1426 dt_write_lock(env, orphan, 0);
1427 rc = dt_create(env, orphan, la, hint, dof, th);
1431 rc = dt_ref_add(env, orphan, th);
1435 rec->rec_fid = cfid;
1436 rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1437 (const struct dt_key *)dot, th, 1);
1441 rec->rec_fid = lfsck_dto2fid(parent);
1442 rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1443 (const struct dt_key *)dotdot, th, 1);
1448 rc = dt_xattr_set(env, orphan, &lmv_buf, XATTR_NAME_LMV, 0, th);
1453 rc = dt_xattr_set(env, orphan, &linkea_buf,
1454 XATTR_NAME_LINK, 0, th);
1455 dt_write_unlock(env, orphan);
1459 rec->rec_fid = cfid;
1460 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1461 (const struct dt_key *)name, th, 1);
1463 dt_write_lock(env, parent, 0);
1464 rc = dt_ref_add(env, parent, th);
1465 dt_write_unlock(env, parent);
1468 GOTO(stop, rc = (rc == 0 ? 1 : rc));
1471 dt_write_unlock(env, orphan);
1474 rc1 = dt_trans_stop(env, dev, th);
1475 if (rc1 != 0 && rc > 0)
1482 CDEBUG(D_LFSCK, "%s: namespace LFSCK create orphan dir for "
1483 "the object "DFID", name = %s: rc = %d\n",
1484 lfsck_lfsck2name(lfsck), PFID(cfid),
1485 cname->ln_name != NULL ? cname->ln_name : "<NULL>", rc);
1487 if (parent != NULL && !IS_ERR(parent) && parent != lfsck->li_lpf_obj)
1488 lfsck_object_put(env, parent);
1491 ns->ln_flags |= LF_INCONSISTENT;
1497 * Remove the specified entry from the linkEA.
1499 * Locate the linkEA entry with the given @cname and @pfid, then
1500 * remove this entry or the other entries those are repeated with
1503 * \param[in] env pointer to the thread context
1504 * \param[in] com pointer to the lfsck component
1505 * \param[in] obj pointer to the dt_object to be handled
1506 * \param[in,out]ldata pointer to the buffer that holds the linkEA
1507 * \param[in] cname the name for the child in the parent directory
1508 * \param[in] pfid the parent directory's FID for the linkEA
1509 * \param[in] next if true, then remove the first found linkEA
1510 * entry, and move the ldata->ld_lee to next entry
1512 * \retval positive number for repaired cases
1513 * \retval 0 if nothing to be repaired
1514 * \retval negative error number on failure
1516 static int lfsck_namespace_shrink_linkea(const struct lu_env *env,
1517 struct lfsck_component *com,
1518 struct dt_object *obj,
1519 struct linkea_data *ldata,
1520 struct lu_name *cname,
1521 struct lu_fid *pfid,
1524 struct lfsck_instance *lfsck = com->lc_lfsck;
1525 struct dt_device *dev = lfsck_obj2dev(obj);
1526 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
1527 struct thandle *th = NULL;
1528 struct lustre_handle lh = { 0 };
1529 struct linkea_data ldata_new = { NULL };
1530 struct lu_buf linkea_buf;
1535 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
1536 MDS_INODELOCK_UPDATE | MDS_INODELOCK_XATTR,
1542 lfsck_linkea_del_buf(ldata, cname);
1544 lfsck_namespace_filter_linkea_entry(ldata, cname, pfid,
1546 if (ldata->ld_leh->leh_reccount > 0 ||
1547 unlikely(ldata->ld_leh->leh_overflow_time)) {
1548 lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
1549 ldata->ld_leh->leh_len);
1550 buflen = linkea_buf.lb_len;
1554 th = dt_trans_create(env, dev);
1556 GOTO(unlock1, rc = PTR_ERR(th));
1559 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
1560 XATTR_NAME_LINK, 0, th);
1562 rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, th);
1566 rc = dt_trans_start_local(env, dev, th);
1570 dt_write_lock(env, obj, 0);
1571 if (unlikely(lfsck_is_dead_obj(obj)))
1572 GOTO(unlock2, rc = -ENOENT);
1574 rc = lfsck_links_read2_with_rec(env, obj, &ldata_new);
1576 GOTO(unlock2, rc = (rc == -ENODATA ? 0 : rc));
1578 /* The specified linkEA entry has been removed by race. */
1579 rc = linkea_links_find(&ldata_new, cname, pfid);
1581 GOTO(unlock2, rc = 0);
1583 if (bk->lb_param & LPF_DRYRUN)
1584 GOTO(unlock2, rc = 1);
1587 lfsck_linkea_del_buf(&ldata_new, cname);
1589 lfsck_namespace_filter_linkea_entry(&ldata_new, cname, pfid,
1592 if (buflen < ldata_new.ld_leh->leh_len) {
1593 dt_write_unlock(env, obj);
1594 dt_trans_stop(env, dev, th);
1595 lfsck_buf_init(&linkea_buf, ldata_new.ld_buf->lb_buf,
1596 ldata_new.ld_leh->leh_len);
1600 if (ldata_new.ld_leh->leh_reccount > 0 ||
1601 unlikely(ldata->ld_leh->leh_overflow_time))
1602 rc = lfsck_links_write(env, obj, &ldata_new, th);
1604 rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, th);
1606 GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
1609 dt_write_unlock(env, obj);
1612 dt_trans_stop(env, dev, th);
1615 lfsck_ibits_unlock(&lh, LCK_EX);
1618 CDEBUG(D_LFSCK, "%s: namespace LFSCK remove %s linkEA entry "
1619 "for the object: "DFID", parent "DFID", name %.*s\n",
1620 lfsck_lfsck2name(lfsck), next ? "invalid" : "redundant",
1621 PFID(lfsck_dto2fid(obj)), PFID(pfid), cname->ln_namelen,
1625 struct lfsck_namespace *ns = com->lc_file_ram;
1627 ns->ln_flags |= LF_INCONSISTENT;
1634 * Conditionally remove the specified entry from the linkEA.
1636 * Take the parent lock firstly, then check whether the specified
1637 * name entry exists or not: if yes, do nothing; otherwise, call
1638 * lfsck_namespace_shrink_linkea() to remove the linkea entry.
1640 * \param[in] env pointer to the thread context
1641 * \param[in] com pointer to the lfsck component
1642 * \param[in] parent pointer to the parent directory
1643 * \param[in] child pointer to the child object that holds the linkEA
1644 * \param[in,out]ldata pointer to the buffer that holds the linkEA
1645 * \param[in] cname the name for the child in the parent directory
1646 * \param[in] pfid the parent directory's FID for the linkEA
1648 * \retval positive number for repaired cases
1649 * \retval 0 if nothing to be repaired
1650 * \retval negative error number on failure
1652 static int lfsck_namespace_shrink_linkea_cond(const struct lu_env *env,
1653 struct lfsck_component *com,
1654 struct dt_object *parent,
1655 struct dt_object *child,
1656 struct linkea_data *ldata,
1657 struct lu_name *cname,
1658 struct lu_fid *pfid)
1660 struct lfsck_thread_info *info = lfsck_env_info(env);
1661 struct lu_fid *cfid = &info->lti_fid3;
1662 struct lfsck_lock_handle *llh = &info->lti_llh;
1666 rc = lfsck_lock(env, com->lc_lfsck, parent, cname->ln_name, llh,
1667 MDS_INODELOCK_UPDATE, LCK_PR);
1671 dt_read_lock(env, parent, 0);
1672 if (unlikely(lfsck_is_dead_obj(parent))) {
1673 dt_read_unlock(env, parent);
1675 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
1681 rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
1682 (const struct dt_key *)cname->ln_name);
1683 dt_read_unlock(env, parent);
1685 /* It is safe to release the ldlm lock, because when the logic come
1686 * here, we have got all the needed information above whether the
1687 * linkEA entry is valid or not. It is not important that others
1688 * may add new linkEA entry after the ldlm lock released. If other
1689 * has removed the specified linkEA entry by race, then it is OK,
1690 * because the subsequent lfsck_namespace_shrink_linkea() can handle
1693 if (rc == -ENOENT) {
1694 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
1703 /* The LFSCK just found some internal status of cross-MDTs
1704 * create operation. That is normal. */
1705 if (lu_fid_eq(cfid, lfsck_dto2fid(child))) {
1706 linkea_next_entry(ldata);
1711 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata, cname,
1718 * Conditionally replace name entry in the parent.
1720 * As required, the LFSCK may re-create the lost MDT-object for dangling
1721 * name entry, but such repairing may be wrong because of bad FID in the
1722 * name entry. As the LFSCK processing, the real MDT-object may be found,
1723 * then the LFSCK should check whether the former re-created MDT-object
1724 * has been modified or not, if not, then destroy it and update the name
1725 * entry in the parent to reference the real MDT-object.
1727 * \param[in] env pointer to the thread context
1728 * \param[in] com pointer to the lfsck component
1729 * \param[in] parent pointer to the parent directory
1730 * \param[in] child pointer to the MDT-object that may be the real
1731 * MDT-object corresponding to the name entry in parent
1732 * \param[in] cfid the current FID in the name entry
1733 * \param[in] cname contains the name of the child in the parent directory
1735 * \retval positive number for repaired cases
1736 * \retval 0 if nothing to be repaired
1737 * \retval negative error number on failure
1739 static int lfsck_namespace_replace_cond(const struct lu_env *env,
1740 struct lfsck_component *com,
1741 struct dt_object *parent,
1742 struct dt_object *child,
1743 const struct lu_fid *cfid,
1744 const struct lu_name *cname)
1746 struct lfsck_thread_info *info = lfsck_env_info(env);
1747 struct lu_attr *la = &info->lti_la;
1748 struct dt_insert_rec *rec = &info->lti_dt_rec;
1750 struct lfsck_instance *lfsck = com->lc_lfsck;
1751 /* The child and its name may be on different MDTs. */
1752 struct dt_device *dev = lfsck->li_next;
1753 const char *name = cname->ln_name;
1754 const struct lu_fid *pfid = lfsck_dto2fid(parent);
1755 struct dt_object *cobj = NULL;
1756 struct lfsck_lock_handle *pllh = &info->lti_llh;
1757 struct lustre_handle clh = { 0 };
1758 struct linkea_data ldata = { NULL };
1759 struct thandle *th = NULL;
1764 /* @parent/@child may be based on lfsck->li_bottom,
1765 * but here we need the object based on the lfsck->li_next. */
1767 parent = lfsck_object_locate(dev, parent);
1769 GOTO(log, rc = PTR_ERR(parent));
1771 if (unlikely(!dt_try_as_dir(env, parent)))
1772 GOTO(log, rc = -ENOTDIR);
1774 rc = lfsck_lock(env, lfsck, parent, name, pllh,
1775 MDS_INODELOCK_UPDATE, LCK_PW);
1779 if (!fid_is_sane(cfid)) {
1784 cobj = lfsck_object_find_by_dev(env, dev, cfid);
1787 if (rc == -ENOENT) {
1795 if (!dt_object_exists(cobj)) {
1800 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
1801 (const struct dt_key *)name);
1802 if (rc == -ENOENT) {
1810 /* Someone changed the name entry, cannot replace it. */
1811 if (!lu_fid_eq(cfid, &tfid))
1814 /* lock the object to be destroyed. */
1815 rc = lfsck_ibits_lock(env, lfsck, cobj, &clh,
1816 MDS_INODELOCK_UPDATE |
1817 MDS_INODELOCK_UPDATE | MDS_INODELOCK_XATTR,
1822 if (unlikely(lfsck_is_dead_obj(cobj))) {
1827 rc = dt_attr_get(env, cobj, la);
1831 /* The object has been modified by other(s), or it is not created by
1832 * LFSCK, the two cases are indistinguishable. So cannot replace it. */
1833 if (la->la_ctime != 0)
1836 if (S_ISREG(la->la_mode)) {
1837 rc = dt_xattr_get(env, cobj, &LU_BUF_NULL, XATTR_NAME_LOV);
1838 /* If someone has created related OST-object(s),
1840 if ((rc > 0) || (rc < 0 && rc != -ENODATA))
1841 GOTO(log, rc = (rc > 0 ? 0 : rc));
1845 dt_read_lock(env, child, 0);
1846 rc = lfsck_links_read2_with_rec(env, child, &ldata);
1847 dt_read_unlock(env, child);
1849 /* Someone changed the child, no need to replace. */
1856 rc = linkea_links_find(&ldata, cname, pfid);
1857 /* Someone moved the child, no need to replace. */
1861 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1864 th = dt_trans_create(env, dev);
1866 GOTO(log, rc = PTR_ERR(th));
1869 rc = dt_declare_destroy(env, cobj, th);
1874 rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
1878 rec->rec_type = S_IFDIR;
1879 rec->rec_fid = lfsck_dto2fid(child);
1880 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1881 (const struct dt_key *)name, th);
1885 rc = dt_trans_start_local(env, dev, th);
1890 rc = dt_destroy(env, cobj, th);
1895 /* The old name entry maybe not exist. */
1896 rc = dt_delete(env, parent, (const struct dt_key *)name, th);
1897 if (rc != 0 && rc != -ENOENT)
1900 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1901 (const struct dt_key *)name, th, 1);
1903 GOTO(stop, rc = (rc == 0 ? 1 : rc));
1906 dt_trans_stop(env, dev, th);
1909 lfsck_ibits_unlock(&clh, LCK_EX);
1912 if (cobj != NULL && !IS_ERR(cobj))
1913 lfsck_object_put(env, cobj);
1915 CDEBUG(D_LFSCK, "%s: namespace LFSCK conditionally destroy the "
1916 "object "DFID" because of conflict with the object "DFID
1917 " under the parent "DFID" with name %s: rc = %d\n",
1918 lfsck_lfsck2name(lfsck), PFID(cfid),
1919 PFID(lfsck_dto2fid(child)), PFID(pfid), name, rc);
1925 * Overwrite the linkEA for the object with the given ldata.
1927 * The caller should take the ldlm lock before the calling.
1929 * \param[in] env pointer to the thread context
1930 * \param[in] com pointer to the lfsck component
1931 * \param[in] obj pointer to the dt_object to be handled
1932 * \param[in] ldata pointer to the new linkEA data
1934 * \retval positive number for repaired cases
1935 * \retval 0 if nothing to be repaired
1936 * \retval negative error number on failure
1938 int lfsck_namespace_rebuild_linkea(const struct lu_env *env,
1939 struct lfsck_component *com,
1940 struct dt_object *obj,
1941 struct linkea_data *ldata)
1943 struct lfsck_instance *lfsck = com->lc_lfsck;
1944 struct dt_device *dev = lfsck_obj2dev(obj);
1945 struct thandle *th = NULL;
1946 struct lu_buf linkea_buf;
1950 th = dt_trans_create(env, dev);
1952 GOTO(log, rc = PTR_ERR(th));
1954 lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
1955 ldata->ld_leh->leh_len);
1956 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
1957 XATTR_NAME_LINK, 0, th);
1961 rc = dt_trans_start_local(env, dev, th);
1965 dt_write_lock(env, obj, 0);
1966 if (unlikely(lfsck_is_dead_obj(obj)))
1967 GOTO(unlock, rc = 0);
1969 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1970 GOTO(unlock, rc = 1);
1972 rc = dt_xattr_set(env, obj, &linkea_buf,
1973 XATTR_NAME_LINK, 0, th);
1975 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
1978 dt_write_unlock(env, obj);
1981 dt_trans_stop(env, dev, th);
1984 CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild linkEA for the "
1985 "object "DFID": rc = %d\n",
1986 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
1989 struct lfsck_namespace *ns = com->lc_file_ram;
1991 ns->ln_flags |= LF_INCONSISTENT;
1998 * Repair invalid name entry.
2000 * If the name entry contains invalid information, such as bad file type
2001 * or (and) corrupted object FID, then either remove the name entry or
2002 * udpate the name entry with the given (right) information.
2004 * \param[in] env pointer to the thread context
2005 * \param[in] com pointer to the lfsck component
2006 * \param[in] parent pointer to the parent directory
2007 * \param[in] child pointer to the object referenced by the name entry
2008 * \param[in] name the old name of the child under the parent directory
2009 * \param[in] name2 the new name of the child under the parent directory
2010 * \param[in] type the type claimed by the name entry
2011 * \param[in] update update the name entry if true; otherwise, remove it
2012 * \param[in] dec decrease the parent nlink count if true
2014 * \retval positive number for repaired successfully
2015 * \retval 0 if nothing to be repaired
2016 * \retval negative error number on failure
2018 int lfsck_namespace_repair_dirent(const struct lu_env *env,
2019 struct lfsck_component *com,
2020 struct dt_object *parent,
2021 struct dt_object *child,
2022 const char *name, const char *name2,
2023 __u16 type, bool update, bool dec)
2025 struct lfsck_thread_info *info = lfsck_env_info(env);
2026 struct dt_insert_rec *rec = &info->lti_dt_rec;
2027 const struct lu_fid *pfid = lfsck_dto2fid(parent);
2028 const struct lu_fid *cfid = lfsck_dto2fid(child);
2030 struct lfsck_instance *lfsck = com->lc_lfsck;
2031 struct dt_device *dev = lfsck->li_next;
2032 struct thandle *th = NULL;
2033 struct lfsck_lock_handle *llh = &info->lti_llh;
2034 struct lustre_handle lh = { 0 };
2038 parent = lfsck_object_locate(dev, parent);
2040 GOTO(log, rc = PTR_ERR(parent));
2042 if (unlikely(!dt_try_as_dir(env, parent)))
2043 GOTO(log, rc = -ENOTDIR);
2045 if (!update || strcmp(name, name2) == 0)
2046 rc = lfsck_lock(env, lfsck, parent, name, llh,
2047 MDS_INODELOCK_UPDATE, LCK_PW);
2049 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
2050 MDS_INODELOCK_UPDATE, LCK_PW);
2054 th = dt_trans_create(env, dev);
2056 GOTO(unlock1, rc = PTR_ERR(th));
2058 rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
2063 rec->rec_type = lfsck_object_type(child) & S_IFMT;
2064 rec->rec_fid = cfid;
2065 rc = dt_declare_insert(env, parent,
2066 (const struct dt_rec *)rec,
2067 (const struct dt_key *)name2, th);
2072 if (dec && S_ISDIR(type)) {
2073 rc = dt_declare_ref_del(env, parent, th);
2078 rc = dt_trans_start_local(env, dev, th);
2083 dt_write_lock(env, parent, 0);
2084 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
2085 (const struct dt_key *)name);
2086 /* Someone has removed the bad name entry by race. */
2088 GOTO(unlock2, rc = 0);
2093 /* Someone has removed the bad name entry and reused it for other
2094 * object by race. */
2095 if (!lu_fid_eq(&tfid, cfid))
2096 GOTO(unlock2, rc = 0);
2098 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2099 GOTO(unlock2, rc = 1);
2101 rc = dt_delete(env, parent, (const struct dt_key *)name, th);
2106 rc = dt_insert(env, parent,
2107 (const struct dt_rec *)rec,
2108 (const struct dt_key *)name2, th, 1);
2113 if (dec && S_ISDIR(type)) {
2114 rc = dt_ref_del(env, parent, th);
2119 GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
2122 dt_write_unlock(env, parent);
2125 dt_trans_stop(env, dev, th);
2127 /* We are not sure whether the child will become orphan or not.
2128 * Record it in the LFSCK trace file for further checking in
2129 * the second-stage scanning. */
2130 if (!update && !dec && rc == 0)
2131 lfsck_namespace_trace_update(env, com, cfid,
2132 LNTF_CHECK_LINKEA, true);
2135 /* It is harmless even if unlock the unused lock_handle */
2136 lfsck_ibits_unlock(&lh, LCK_PW);
2140 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found bad name "
2141 "entry for: parent "DFID", child "DFID", name %s, type "
2142 "in name entry %o, type claimed by child %o. repair it "
2143 "by %s with new name2 %s: rc = %d\n",
2144 lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
2145 name, type, update ? lfsck_object_type(child) : 0,
2146 update ? "updating" : "removing", name2, rc);
2149 struct lfsck_namespace *ns = com->lc_file_ram;
2151 ns->ln_flags |= LF_INCONSISTENT;
2158 * Update the ".." name entry for the given object.
2160 * The object's ".." is corrupted, this function will update the ".." name
2161 * entry with the given pfid, and the linkEA with the given ldata.
2163 * The caller should take the ldlm lock before the calling.
2165 * \param[in] env pointer to the thread context
2166 * \param[in] com pointer to the lfsck component
2167 * \param[in] obj pointer to the dt_object to be handled
2168 * \param[in] pfid the new fid for the object's ".." name entry
2169 * \param[in] cname the name for the @obj in the parent directory
2171 * \retval positive number for repaired cases
2172 * \retval 0 if nothing to be repaired
2173 * \retval negative error number on failure
2175 static int lfsck_namespace_repair_unmatched_pairs(const struct lu_env *env,
2176 struct lfsck_component *com,
2177 struct dt_object *obj,
2178 const struct lu_fid *pfid,
2179 struct lu_name *cname)
2181 struct lfsck_thread_info *info = lfsck_env_info(env);
2182 struct dt_insert_rec *rec = &info->lti_dt_rec;
2183 struct lfsck_instance *lfsck = com->lc_lfsck;
2184 struct dt_device *dev = lfsck_obj2dev(obj);
2185 struct thandle *th = NULL;
2186 struct linkea_data ldata = { NULL };
2187 struct lu_buf linkea_buf;
2191 LASSERT(!dt_object_remote(obj));
2192 LASSERT(S_ISDIR(lfsck_object_type(obj)));
2194 rc = linkea_links_new(&ldata, &info->lti_big_buf, cname, pfid);
2198 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
2199 ldata.ld_leh->leh_len);
2201 th = dt_trans_create(env, dev);
2203 GOTO(log, rc = PTR_ERR(th));
2205 rc = dt_declare_delete(env, obj, (const struct dt_key *)dotdot, th);
2209 rec->rec_type = S_IFDIR;
2210 rec->rec_fid = pfid;
2211 rc = dt_declare_insert(env, obj, (const struct dt_rec *)rec,
2212 (const struct dt_key *)dotdot, th);
2216 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
2217 XATTR_NAME_LINK, 0, th);
2221 rc = dt_trans_start_local(env, dev, th);
2225 dt_write_lock(env, obj, 0);
2226 if (unlikely(lfsck_is_dead_obj(obj)))
2227 GOTO(unlock, rc = 0);
2229 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2230 GOTO(unlock, rc = 1);
2232 /* The old ".." name entry maybe not exist. */
2233 dt_delete(env, obj, (const struct dt_key *)dotdot, th);
2235 rc = dt_insert(env, obj, (const struct dt_rec *)rec,
2236 (const struct dt_key *)dotdot, th, 1);
2240 rc = lfsck_links_write(env, obj, &ldata, th);
2242 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2245 dt_write_unlock(env, obj);
2248 dt_trans_stop(env, dev, th);
2251 CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild dotdot name entry for "
2252 "the object "DFID", new parent "DFID": rc = %d\n",
2253 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)),
2257 struct lfsck_namespace *ns = com->lc_file_ram;
2259 ns->ln_flags |= LF_INCONSISTENT;
2266 * Handle orphan @obj during Double Scan Directory.
2268 * Remove the @obj's current (invalid) linkEA entries, and insert
2269 * it in the directory .lustre/lost+found/MDTxxxx/ with the name:
2270 * ${FID}-${PFID}-D-${conflict_version}
2272 * The caller should take the ldlm lock before the calling.
2274 * \param[in] env pointer to the thread context
2275 * \param[in] com pointer to the lfsck component
2276 * \param[in] obj pointer to the orphan object to be handled
2277 * \param[in] pfid the new fid for the object's ".." name entry
2278 * \param[in,out] lh ldlm lock handler for the given @obj
2279 * \param[out] type to tell the caller what the inconsistency is
2281 * \retval positive number for repaired cases
2282 * \retval 0 if nothing to be repaired
2283 * \retval negative error number on failure
2286 lfsck_namespace_dsd_orphan(const struct lu_env *env,
2287 struct lfsck_component *com,
2288 struct dt_object *obj,
2289 const struct lu_fid *pfid,
2290 struct lustre_handle *lh,
2291 enum lfsck_namespace_inconsistency_type *type)
2293 struct lfsck_thread_info *info = lfsck_env_info(env);
2294 struct lfsck_namespace *ns = com->lc_file_ram;
2298 /* Remove the unrecognized linkEA. */
2299 rc = lfsck_namespace_links_remove(env, com, obj);
2300 lfsck_ibits_unlock(lh, LCK_EX);
2301 if (rc < 0 && rc != -ENODATA)
2304 *type = LNIT_MUL_REF;
2306 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
2307 * ever tried to verify some remote MDT-object that resides on this
2308 * MDT, but this MDT failed to respond such request. So means there
2309 * may be some remote name entry on other MDT that references this
2310 * object with another name, so we cannot know whether this linkEA
2311 * is valid or not. So keep it there and maybe resolved when next
2313 if (ns->ln_flags & LF_INCOMPLETE)
2316 /* The unique linkEA is invalid, even if the ".." name entry may be
2317 * valid, we still cannot know via which name entry this directory
2318 * will be referenced. Then handle it as pure orphan. */
2319 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2320 "-"DFID, PFID(pfid));
2321 rc = lfsck_namespace_insert_orphan(env, com, obj,
2322 info->lti_tmpbuf, "D", NULL);
2328 * Double Scan Directory object for single linkEA entry case.
2330 * The given @child has unique linkEA entry. If the linkEA entry is valid,
2331 * then check whether the name is in the namespace or not, if not, add the
2332 * missing name entry back to namespace. If the linkEA entry is invalid,
2333 * then remove it and insert the @child in the .lustre/lost+found/MDTxxxx/
2336 * \param[in] env pointer to the thread context
2337 * \param[in] com pointer to the lfsck component
2338 * \param[in] child pointer to the directory to be double scanned
2339 * \param[in] pfid the FID corresponding to the ".." entry
2340 * \param[in] ldata pointer to the linkEA data for the given @child
2341 * \param[in,out] lh ldlm lock handler for the given @child
2342 * \param[out] type to tell the caller what the inconsistency is
2343 * \param[in] retry if found inconsistency, but the caller does not hold
2344 * ldlm lock on the @child, then set @retry as true
2345 * \param[in] unknown set if does not know how to repair the inconsistency
2347 * \retval positive number for repaired cases
2348 * \retval 0 if nothing to be repaired
2349 * \retval negative error number on failure
2352 lfsck_namespace_dsd_single(const struct lu_env *env,
2353 struct lfsck_component *com,
2354 struct dt_object *child,
2355 const struct lu_fid *pfid,
2356 struct linkea_data *ldata,
2357 struct lustre_handle *lh,
2358 enum lfsck_namespace_inconsistency_type *type,
2359 bool *retry, bool *unknown)
2361 struct lfsck_thread_info *info = lfsck_env_info(env);
2362 struct lu_name *cname = &info->lti_name;
2363 const struct lu_fid *cfid = lfsck_dto2fid(child);
2365 struct lfsck_namespace *ns = com->lc_file_ram;
2366 struct lfsck_instance *lfsck = com->lc_lfsck;
2367 struct dt_object *parent = NULL;
2368 struct lmv_mds_md_v1 *lmv;
2372 rc = lfsck_namespace_unpack_linkea_entry(ldata, cname, &tfid,
2374 sizeof(info->lti_key));
2375 /* The unique linkEA entry with bad parent will be handled as orphan. */
2377 if (!lustre_handle_is_used(lh) && retry != NULL)
2380 rc = lfsck_namespace_dsd_orphan(env, com, child,
2386 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
2388 GOTO(out, rc = PTR_ERR(parent));
2390 /* We trust the unique linkEA entry in spite of whether it matches the
2391 * ".." name entry or not. Because even if the linkEA entry is wrong
2392 * and the ".." name entry is right, we still cannot know via which
2393 * name entry the child will be referenced, since all known entries
2394 * have been verified during the first-stage scanning. */
2395 if (!dt_object_exists(parent)) {
2396 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2397 * has ever tried to verify some remote MDT-object that resides
2398 * on this MDT, but this MDT failed to respond such request. So
2399 * means there may be some remote name entry on other MDT that
2400 * references this object with another name, so we cannot know
2401 * whether this linkEA is valid or not. So keep it there and
2402 * maybe resolved when next LFSCK run. */
2403 if (ns->ln_flags & LF_INCOMPLETE)
2406 if (!lustre_handle_is_used(lh) && retry != NULL) {
2412 lfsck_ibits_unlock(lh, LCK_EX);
2415 lmv = &info->lti_lmv;
2416 rc = lfsck_read_stripe_lmv(env, child, lmv);
2417 if (rc != 0 && rc != -ENODATA)
2420 if (rc == -ENODATA || lmv->lmv_magic != LMV_MAGIC_STRIPE) {
2422 } else if (lfsck_shard_name_to_index(env,
2423 cname->ln_name, cname->ln_namelen,
2424 S_IFDIR, cfid) < 0) {
2425 /* It is an invalid name entry, we
2426 * cannot trust the parent also. */
2427 rc = lfsck_namespace_shrink_linkea(env, com, child,
2428 ldata, cname, &tfid, true);
2432 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2433 "-"DFID, PFID(pfid));
2434 rc = lfsck_namespace_insert_orphan(env, com, child,
2435 info->lti_tmpbuf, "S", NULL);
2440 /* Create the lost parent as an orphan. */
2441 rc = lfsck_namespace_create_orphan_dir(env, com, parent, lmv);
2443 /* Add the missing name entry to the parent. */
2444 rc = lfsck_namespace_insert_normal(env, com, parent,
2445 child, cname->ln_name);
2446 if (unlikely(rc == -EEXIST)) {
2447 /* Unfortunately, someone reused the name
2448 * under the parent by race. So we have
2449 * to remove the linkEA entry from
2450 * current child object. It means that the
2451 * LFSCK cannot recover the system
2452 * totally back to its original status,
2453 * but it is necessary to make the
2454 * current system to be consistent. */
2455 rc = lfsck_namespace_shrink_linkea(env,
2457 cname, &tfid, true);
2459 snprintf(info->lti_tmpbuf,
2460 sizeof(info->lti_tmpbuf),
2461 "-"DFID, PFID(pfid));
2462 rc = lfsck_namespace_insert_orphan(env,
2463 com, child, info->lti_tmpbuf,
2470 } /* !dt_object_exists(parent) */
2472 /* The unique linkEA entry with bad parent will be handled as orphan. */
2473 if (unlikely(!dt_try_as_dir(env, parent))) {
2474 if (!lustre_handle_is_used(lh) && retry != NULL)
2477 rc = lfsck_namespace_dsd_orphan(env, com, child,
2483 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
2484 (const struct dt_key *)cname->ln_name);
2485 if (rc == -ENOENT) {
2486 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2487 * has ever tried to verify some remote MDT-object that resides
2488 * on this MDT, but this MDT failed to respond such request. So
2489 * means there may be some remote name entry on other MDT that
2490 * references this object with another name, so we cannot know
2491 * whether this linkEA is valid or not. So keep it there and
2492 * maybe resolved when next LFSCK run. */
2493 if (ns->ln_flags & LF_INCOMPLETE)
2496 if (!lustre_handle_is_used(lh) && retry != NULL) {
2502 lfsck_ibits_unlock(lh, LCK_EX);
2503 rc = lfsck_namespace_check_name(env, parent, child, cname);
2510 /* It is an invalid name entry, drop it. */
2511 if (unlikely(rc > 0)) {
2512 rc = lfsck_namespace_shrink_linkea(env, com, child,
2513 ldata, cname, &tfid, true);
2515 snprintf(info->lti_tmpbuf,
2516 sizeof(info->lti_tmpbuf),
2517 "-"DFID, PFID(pfid));
2518 rc = lfsck_namespace_insert_orphan(env, com,
2519 child, info->lti_tmpbuf, "D", NULL);
2525 /* Add the missing name entry back to the namespace. */
2526 rc = lfsck_namespace_insert_normal(env, com, parent, child,
2528 if (unlikely(rc == -ESTALE))
2529 /* It may happen when the remote object has been
2530 * removed, but the local MDT is not aware of that. */
2533 if (unlikely(rc == -EEXIST)) {
2534 /* Unfortunately, someone reused the name under the
2535 * parent by race. So we have to remove the linkEA
2536 * entry from current child object. It means that the
2537 * LFSCK cannot recover the system totally back to
2538 * its original status, but it is necessary to make
2539 * the current system to be consistent.
2541 * It also may be because of the LFSCK found some
2542 * internal status of create operation. Under such
2543 * case, nothing to be done. */
2544 rc = lfsck_namespace_shrink_linkea_cond(env, com,
2545 parent, child, ldata, cname, &tfid);
2547 snprintf(info->lti_tmpbuf,
2548 sizeof(info->lti_tmpbuf),
2549 "-"DFID, PFID(pfid));
2550 rc = lfsck_namespace_insert_orphan(env, com,
2551 child, info->lti_tmpbuf, "D", NULL);
2556 } /* rc == -ENOENT */
2561 if (!lu_fid_eq(&tfid, cfid)) {
2562 if (!lustre_handle_is_used(lh) && retry != NULL) {
2568 lfsck_ibits_unlock(lh, LCK_EX);
2569 /* The name entry references another MDT-object that
2570 * may be created by the LFSCK for repairing dangling
2571 * name entry. Try to replace it. */
2572 rc = lfsck_namespace_replace_cond(env, com, parent, child,
2575 rc = lfsck_namespace_dsd_orphan(env, com, child,
2581 /* Zero FID may because the remote directroy object has invalid linkEA,
2582 * or lost linkEA. Under such case, the LFSCK on this MDT does not know
2583 * how to repair the inconsistency, but the namespace LFSCK on the MDT
2584 * where its name entry resides may has more information (name, FID) to
2585 * repair such inconsistency. So here, keep the inconsistency to avoid
2586 * some imporper repairing. */
2587 if (fid_is_zero(pfid)) {
2594 /* The ".." name entry is wrong, update it. */
2595 if (!lu_fid_eq(pfid, lfsck_dto2fid(parent))) {
2596 if (!lustre_handle_is_used(lh) && retry != NULL) {
2602 *type = LNIT_UNMATCHED_PAIRS;
2603 rc = lfsck_namespace_repair_unmatched_pairs(env, com, child,
2604 lfsck_dto2fid(parent), cname);
2610 if (parent != NULL && !IS_ERR(parent))
2611 lfsck_object_put(env, parent);
2617 * Double Scan Directory object for multiple linkEA entries case.
2619 * The given @child has multiple linkEA entries. There is at most one linkEA
2620 * entry will be valid, all the others will be removed. Firstly, the function
2621 * will try to find out the linkEA entry for which the name entry exists under
2622 * the given parent (@pfid). If there is no linkEA entry that matches the given
2623 * ".." name entry, then tries to find out the first linkEA entry that both the
2624 * parent and the name entry exist to rebuild a new ".." name entry.
2626 * \param[in] env pointer to the thread context
2627 * \param[in] com pointer to the lfsck component
2628 * \param[in] child pointer to the directory to be double scanned
2629 * \param[in] pfid the FID corresponding to the ".." entry
2630 * \param[in] ldata pointer to the linkEA data for the given @child
2631 * \param[in,out] lh ldlm lock handler for the given @child
2632 * \param[out] type to tell the caller what the inconsistency is
2633 * \param[in] lpf true if the ".." entry is under lost+found/MDTxxxx/
2634 * \param[in] unknown set if does not know how to repair the inconsistency
2636 * \retval positive number for repaired cases
2637 * \retval 0 if nothing to be repaired
2638 * \retval negative error number on failure
2641 lfsck_namespace_dsd_multiple(const struct lu_env *env,
2642 struct lfsck_component *com,
2643 struct dt_object *child,
2644 const struct lu_fid *pfid,
2645 struct linkea_data *ldata,
2646 struct lustre_handle *lh,
2647 enum lfsck_namespace_inconsistency_type *type,
2648 bool lpf, bool *unknown)
2650 struct lfsck_thread_info *info = lfsck_env_info(env);
2651 struct lu_name *cname = &info->lti_name;
2652 const struct lu_fid *cfid = lfsck_dto2fid(child);
2653 struct lu_fid *pfid2 = &info->lti_fid3;
2655 struct lfsck_namespace *ns = com->lc_file_ram;
2656 struct lfsck_instance *lfsck = com->lc_lfsck;
2657 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
2658 struct dt_object *parent = NULL;
2659 struct linkea_data ldata_new = { NULL };
2660 int dirent_count = 0;
2666 while (ldata->ld_lee != NULL) {
2667 rc = lfsck_namespace_unpack_linkea_entry(ldata, cname, &tfid,
2669 sizeof(info->lti_key));
2670 /* Drop invalid linkEA entry. */
2672 lfsck_linkea_del_buf(ldata, cname);
2676 /* Drop repeated linkEA entries. */
2677 lfsck_namespace_filter_linkea_entry(ldata, cname, &tfid, true);
2679 /* If current dotdot is the .lustre/lost+found/MDTxxxx/,
2680 * then it is possible that: the directry object has ever
2681 * been lost, but its name entry was there. In the former
2682 * LFSCK run, during the first-stage scanning, the LFSCK
2683 * found the dangling name entry, but it did not recreate
2684 * the lost object, and when moved to the second-stage
2685 * scanning, some children objects of the lost directory
2686 * object were found, then the LFSCK recreated such lost
2687 * directory object as an orphan.
2689 * When the LFSCK runs again, if the dangling name is still
2690 * there, the LFSCK should move the orphan directory object
2691 * back to the normal namespace. */
2692 if (!lpf && !fid_is_zero(pfid) &&
2693 !lu_fid_eq(pfid, &tfid) && once) {
2694 linkea_next_entry(ldata);
2698 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
2700 RETURN(PTR_ERR(parent));
2702 if (!dt_object_exists(parent)) {
2703 lfsck_object_put(env, parent);
2704 if (ldata->ld_leh->leh_reccount > 1) {
2705 /* If it is NOT the last linkEA entry, then
2706 * there is still other chance to make the
2707 * child to be visible via other parent, then
2708 * remove this linkEA entry. */
2709 lfsck_linkea_del_buf(ldata, cname);
2716 /* The linkEA entry with bad parent will be removed. */
2717 if (unlikely(!dt_try_as_dir(env, parent))) {
2718 lfsck_object_put(env, parent);
2719 lfsck_linkea_del_buf(ldata, cname);
2723 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
2724 (const struct dt_key *)cname->ln_name);
2725 *pfid2 = *lfsck_dto2fid(parent);
2726 if (rc == -ENOENT) {
2727 lfsck_object_put(env, parent);
2728 linkea_next_entry(ldata);
2733 lfsck_object_put(env, parent);
2738 if (lu_fid_eq(&tfid, cfid)) {
2739 lfsck_object_put(env, parent);
2740 /* If the parent (that is declared via linkEA entry)
2741 * directory contains the specified child, but such
2742 * parent does not match the dotdot name entry, then
2743 * trust the linkEA. */
2744 if (!fid_is_zero(pfid) && !lu_fid_eq(pfid, pfid2)) {
2745 *type = LNIT_UNMATCHED_PAIRS;
2746 rc = lfsck_namespace_repair_unmatched_pairs(env,
2747 com, child, pfid2, cname);
2753 /* It is the most common case that we find the
2754 * name entry corresponding to the linkEA entry
2755 * that matches the ".." name entry. */
2756 rc = linkea_links_new(&ldata_new, &info->lti_big_buf,
2761 rc = lfsck_namespace_rebuild_linkea(env, com, child,
2766 lfsck_linkea_del_buf(ldata, cname);
2767 linkea_first_entry(ldata);
2768 /* There may be some invalid dangling name entries under
2769 * other parent directories, remove all of them. */
2770 while (ldata->ld_lee != NULL) {
2771 rc = lfsck_namespace_unpack_linkea_entry(ldata,
2772 cname, &tfid, info->lti_key,
2773 sizeof(info->lti_key));
2777 parent = lfsck_object_find_bottom(env, lfsck,
2779 if (IS_ERR(parent)) {
2780 rc = PTR_ERR(parent);
2781 if (rc != -ENOENT &&
2782 bk->lb_param & LPF_FAILOUT)
2788 if (!dt_object_exists(parent)) {
2789 lfsck_object_put(env, parent);
2793 rc = lfsck_namespace_repair_dirent(env, com,
2794 parent, child, cname->ln_name,
2795 cname->ln_name, S_IFDIR, false, true);
2796 lfsck_object_put(env, parent);
2798 if (bk->lb_param & LPF_FAILOUT)
2807 lfsck_linkea_del_buf(ldata, cname);
2810 ns->ln_dirent_repaired += dirent_count;
2813 } /* lu_fid_eq(&tfid, lfsck_dto2fid(child)) */
2815 lfsck_ibits_unlock(lh, LCK_EX);
2816 /* The name entry references another MDT-object that may be
2817 * created by the LFSCK for repairing dangling name entry.
2818 * Try to replace it. */
2819 rc = lfsck_namespace_replace_cond(env, com, parent, child,
2821 lfsck_object_put(env, parent);
2828 lfsck_linkea_del_buf(ldata, cname);
2829 } /* while (ldata->ld_lee != NULL) */
2831 /* If there is still linkEA overflow, return. */
2832 if (unlikely(ldata->ld_leh->leh_overflow_time))
2835 linkea_first_entry(ldata);
2836 if (ldata->ld_leh->leh_reccount == 1) {
2837 rc = lfsck_namespace_dsd_single(env, com, child, pfid, ldata,
2838 lh, type, NULL, unknown);
2843 /* All linkEA entries are invalid and removed, then handle the @child
2845 if (ldata->ld_leh->leh_reccount == 0) {
2846 rc = lfsck_namespace_dsd_orphan(env, com, child, pfid, lh,
2852 /* If the dangling name entry for the orphan directory object has
2853 * been remvoed, then just check whether the directory object is
2854 * still under the .lustre/lost+found/MDTxxxx/ or not. */
2860 /* There is no linkEA entry that matches the ".." name entry. Find
2861 * the first linkEA entry that both parent and name entry exist to
2862 * rebuild a new ".." name entry. */
2872 * Repair the object's nlink attribute.
2874 * If all the known name entries have been verified, then the object's hard
2875 * link attribute should match the object's linkEA entries count unless the
2876 * object's has too many hard link to be recorded in the linkEA. Such cases
2877 * should have been marked in the LFSCK trace file. Otherwise, trust the
2878 * linkEA to update the object's nlink attribute.
2880 * \param[in] env pointer to the thread context
2881 * \param[in] com pointer to the lfsck component
2882 * \param[in] obj pointer to the dt_object to be handled
2883 * \param[in,out] la pointer to buffer to object's attribute before
2884 * and after the repairing
2886 * \retval positive number for repaired cases
2887 * \retval 0 if nothing to be repaired
2888 * \retval negative error number on failure
2890 static int lfsck_namespace_repair_nlink(const struct lu_env *env,
2891 struct lfsck_component *com,
2892 struct dt_object *obj,
2895 struct lfsck_namespace *ns = com->lc_file_ram;
2896 struct lfsck_instance *lfsck = com->lc_lfsck;
2897 struct dt_device *dev = lfsck_obj2dev(obj);
2898 const struct lu_fid *cfid = lfsck_dto2fid(obj);
2899 struct thandle *th = NULL;
2900 struct linkea_data ldata = { NULL };
2901 struct lustre_handle lh = { 0 };
2902 __u32 old = la->la_nlink;
2906 LASSERT(!dt_object_remote(obj));
2908 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
2909 MDS_INODELOCK_UPDATE, LCK_PW);
2913 th = dt_trans_create(env, dev);
2915 GOTO(log, rc = PTR_ERR(th));
2917 la->la_valid = LA_NLINK;
2918 rc = dt_declare_attr_set(env, obj, la, th);
2922 rc = dt_trans_start_local(env, dev, th);
2926 dt_write_lock(env, obj, 0);
2927 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
2928 * ever tried to verify some remote MDT-object that resides on this
2929 * MDT, but this MDT failed to respond such request. So means there
2930 * may be some remote name entry on other MDT that references this
2931 * object with another name, so we cannot know whether this linkEA
2932 * is valid or not. So keep it there and maybe resolved when next
2934 if (ns->ln_flags & LF_INCOMPLETE)
2935 GOTO(unlock, rc = 0);
2937 rc = dt_attr_get(env, obj, la);
2939 GOTO(unlock, rc = (rc == -ENOENT ? 0 : rc));
2941 rc = lfsck_links_read2_with_rec(env, obj, &ldata);
2943 GOTO(unlock, rc = (rc == -ENODATA ? 0 : rc));
2945 /* XXX: Currently, we only update the nlink attribute if the known
2946 * linkEA entries is larger than the nlink attribute. That is
2948 if (la->la_nlink >= ldata.ld_leh->leh_reccount ||
2949 unlikely(la->la_nlink == 0 ||
2950 ldata.ld_leh->leh_overflow_time))
2951 GOTO(unlock, rc = 0);
2953 la->la_nlink = ldata.ld_leh->leh_reccount;
2954 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2955 GOTO(unlock, rc = 1);
2957 rc = dt_attr_set(env, obj, la, th);
2959 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2962 dt_write_unlock(env, obj);
2965 dt_trans_stop(env, dev, th);
2968 lfsck_ibits_unlock(&lh, LCK_PW);
2969 CDEBUG(D_LFSCK, "%s: namespace LFSCK repaired the object "DFID"'s "
2970 "nlink count from %u to %u: rc = %d\n",
2971 lfsck_lfsck2name(lfsck), PFID(cfid), old, la->la_nlink, rc);
2974 ns->ln_flags |= LF_INCONSISTENT;
2980 * Double scan the directory object for namespace LFSCK.
2982 * This function will verify the <parent, child> pairs in the namespace tree:
2983 * the parent references the child via some name entry that should be in the
2984 * child's linkEA entry, the child should back references the parent via its
2987 * The LFSCK will scan every linkEA entry in turn until find out the first
2988 * matched pairs. If found, then all other linkEA entries will be dropped.
2989 * If all the linkEA entries cannot match the ".." name entry, then there
2990 * are serveral possible cases:
2992 * 1) If there is only one linkEA entry, then trust it as long as the PFID
2993 * in the linkEA entry is valid.
2995 * 2) If there are multiple linkEA entries, then try to find the linkEA
2996 * that matches the ".." name entry. If found, then all other entries
2997 * are invalid; otherwise, it is quite possible that the ".." name entry
2998 * is corrupted. Under such case, the LFSCK will rebuild the ".." name
2999 * entry according to the first valid linkEA entry (both the parent and
3000 * the name entry should exist).
3002 * 3) If the directory object has no (valid) linkEA entry, then the
3003 * directory object will be handled as pure orphan and inserted
3004 * in the .lustre/lost+found/MDTxxxx/ with the name:
3005 * ${self_FID}-${PFID}-D-${conflict_version}
3007 * \param[in] env pointer to the thread context
3008 * \param[in] com pointer to the lfsck component
3009 * \param[in] child pointer to the directory object to be handled
3010 * \param[in] flags to indicate the specical checking on the @child
3012 * \retval positive number for repaired cases
3013 * \retval 0 if nothing to be repaired
3014 * \retval negative error number on failure
3016 static int lfsck_namespace_double_scan_dir(const struct lu_env *env,
3017 struct lfsck_component *com,
3018 struct dt_object *child, __u8 flags)
3020 struct lfsck_thread_info *info = lfsck_env_info(env);
3021 const struct lu_fid *cfid = lfsck_dto2fid(child);
3022 struct lu_fid *pfid = &info->lti_fid2;
3023 struct lfsck_namespace *ns = com->lc_file_ram;
3024 struct lfsck_instance *lfsck = com->lc_lfsck;
3025 struct lustre_handle lh = { 0 };
3026 struct linkea_data ldata = { NULL };
3027 bool unknown = false;
3030 enum lfsck_namespace_inconsistency_type type = LNIT_BAD_LINKEA;
3034 LASSERT(!dt_object_remote(child));
3036 if (flags & LNTF_UNCERTAIN_LMV) {
3037 if (flags & LNTF_RECHECK_NAME_HASH) {
3038 rc = lfsck_namespace_scan_shard(env, com, child);
3042 ns->ln_striped_shards_scanned++;
3044 ns->ln_striped_shards_skipped++;
3048 flags &= ~(LNTF_RECHECK_NAME_HASH | LNTF_UNCERTAIN_LMV);
3052 if (flags & (LNTF_CHECK_LINKEA | LNTF_CHECK_PARENT) &&
3053 !(lfsck->li_bookmark_ram.lb_param & LPF_ALL_TGT)) {
3054 CDEBUG(D_LFSCK, "%s: some MDT(s) maybe NOT take part in the"
3055 "the namespace LFSCK, then the LFSCK cannot guarantee"
3056 "all the name entries have been verified in first-stage"
3057 "scanning. So have to skip orphan related handling for"
3058 "the directory object "DFID" with remote name entry\n",
3059 lfsck_lfsck2name(lfsck), PFID(cfid));
3064 if (unlikely(!dt_try_as_dir(env, child)))
3065 GOTO(out, rc = -ENOTDIR);
3067 /* We only take ldlm lock on the @child when required. When the
3068 * logic comes here for the first time, it is always false. */
3072 rc = lfsck_ibits_lock(env, lfsck, child, &lh,
3073 MDS_INODELOCK_UPDATE |
3074 MDS_INODELOCK_XATTR, LCK_EX);
3079 dt_read_lock(env, child, 0);
3080 if (unlikely(lfsck_is_dead_obj(child))) {
3081 dt_read_unlock(env, child);
3086 rc = dt_lookup(env, child, (struct dt_rec *)pfid,
3087 (const struct dt_key *)dotdot);
3089 if (rc != -ENOENT && rc != -ENODATA && rc != -EINVAL) {
3090 dt_read_unlock(env, child);
3095 if (!lustre_handle_is_used(&lh)) {
3096 dt_read_unlock(env, child);
3101 } else if (lfsck->li_lpf_obj != NULL &&
3102 lu_fid_eq(pfid, lfsck_dto2fid(lfsck->li_lpf_obj))) {
3104 } else if (unlikely(!fid_is_sane(pfid))) {
3108 rc = lfsck_links_read(env, child, &ldata);
3109 dt_read_unlock(env, child);
3111 if (rc != -ENODATA && rc != -EINVAL)
3114 if (!lustre_handle_is_used(&lh))
3117 if (rc == -EINVAL && !fid_is_zero(pfid)) {
3118 /* Remove the corrupted linkEA. */
3119 rc = lfsck_namespace_links_remove(env, com, child);
3121 /* Here, because of the crashed linkEA, we
3122 * cannot know whether there is some parent
3123 * that references the child directory via
3124 * some name entry or not. So keep it there,
3125 * when the LFSCK run next time, if there is
3126 * some parent that references this object,
3127 * then the LFSCK can rebuild the linkEA;
3128 * otherwise, this object will be handled
3129 * as orphan as above. */
3132 /* 1. If we have neither ".." nor linkEA,
3133 * then it is an orphan.
3135 * 2. If we only have the ".." name entry,
3136 * but no parent references this child
3137 * directory, then handle it as orphan. */
3138 lfsck_ibits_unlock(&lh, LCK_EX);
3139 type = LNIT_MUL_REF;
3141 /* If the LFSCK is marked as LF_INCOMPLETE,
3142 * then means some MDT has ever tried to
3143 * verify some remote MDT-object that resides
3144 * on this MDT, but this MDT failed to respond
3145 * such request. So means there may be some
3146 * remote name entry on other MDT that
3147 * references this object with another name,
3148 * so we cannot know whether this linkEA is
3149 * valid or not. So keep it there and maybe
3150 * resolved when next LFSCK run. */
3151 if (ns->ln_flags & LF_INCOMPLETE)
3154 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
3155 "-"DFID, PFID(pfid));
3156 rc = lfsck_namespace_insert_orphan(env, com, child,
3157 info->lti_tmpbuf, "D", NULL);
3163 linkea_first_entry(&ldata);
3164 /* This is the most common case: the object has unique linkEA entry. */
3165 if (ldata.ld_leh->leh_reccount == 1) {
3166 rc = lfsck_namespace_dsd_single(env, com, child, pfid, &ldata,
3167 &lh, &type, &retry, &unknown);
3169 LASSERT(!lustre_handle_is_used(&lh));
3178 if (!lustre_handle_is_used(&lh))
3181 if (unlikely(ldata.ld_leh->leh_reccount == 0)) {
3182 rc = lfsck_namespace_dsd_orphan(env, com, child, pfid, &lh,
3188 /* When we come here, the cases usually like that:
3189 * 1) The directory object has a corrupted linkEA entry. During the
3190 * first-stage scanning, the LFSCK cannot know such corruption,
3191 * then it appends the right linkEA entry according to the found
3192 * name entry after the bad one.
3194 * 2) The directory object has a right linkEA entry. During the
3195 * first-stage scanning, the LFSCK finds some bad name entry,
3196 * but the LFSCK cannot aware that at that time, then it adds
3197 * the bad linkEA entry for further processing. */
3198 rc = lfsck_namespace_dsd_multiple(env, com, child, pfid, &ldata,
3199 &lh, &type, lpf, &unknown);
3204 lfsck_ibits_unlock(&lh, LCK_EX);
3207 case LNIT_BAD_LINKEA:
3208 ns->ln_linkea_repaired++;
3210 case LNIT_UNMATCHED_PAIRS:
3211 ns->ln_unmatched_pairs_repaired++;
3214 ns->ln_mul_ref_repaired++;
3222 ns->ln_unknown_inconsistency++;
3228 lfsck_namespace_linkea_stale_overflow(struct linkea_data *ldata,
3229 struct lfsck_namespace *ns)
3231 /* Both the leh_overflow_time and ln_time_latest_reset are
3232 * local time based, so need NOT to care about clock drift
3233 * among the servers. */
3234 return ldata->ld_leh->leh_overflow_time &&
3235 ldata->ld_leh->leh_overflow_time < ns->ln_time_latest_reset;
3239 * Clear the object's linkEA overflow timestamp.
3241 * If the MDT-object has too many hard links as to the linkEA cannot hold
3242 * all of them, then overflow timestamp will be set in the linkEA header.
3243 * If some hard links are removed after that, then it is possible to hold
3244 * other missed linkEA entries. If the namespace LFSCK have added all the
3245 * related linkEA entries, then it will remove the overflow timestamp.
3247 * \param[in] env pointer to the thread context
3248 * \param[in] com pointer to the lfsck component
3249 * \param[in] ldata pointer to the linkEA data for the given @obj
3250 * \param[in] obj pointer to the dt_object to be handled
3252 * \retval positive number for repaired cases
3253 * \retval 0 if nothing to be repaired
3254 * \retval negative error number on failure
3256 static int lfsck_namespace_linkea_clear_overflow(const struct lu_env *env,
3257 struct lfsck_component *com,
3258 struct linkea_data *ldata,
3259 struct dt_object *obj)
3261 struct lfsck_namespace *ns = com->lc_file_ram;
3262 struct lfsck_instance *lfsck = com->lc_lfsck;
3263 struct dt_device *dev = lfsck_obj2dev(obj);
3264 struct thandle *th = NULL;
3265 struct lustre_handle lh = { 0 };
3266 struct lu_buf linkea_buf;
3270 LASSERT(!dt_object_remote(obj));
3272 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
3273 MDS_INODELOCK_UPDATE, LCK_PW);
3277 th = dt_trans_create(env, dev);
3279 GOTO(log, rc = PTR_ERR(th));
3281 rc = dt_declare_xattr_set(env, obj,
3282 lfsck_buf_get_const(env, NULL, MAX_LINKEA_SIZE),
3283 XATTR_NAME_LINK, 0, th);
3287 rc = dt_trans_start_local(env, dev, th);
3291 dt_write_lock(env, obj, 0);
3292 rc = lfsck_links_read(env, obj, ldata);
3296 if (unlikely(!lfsck_namespace_linkea_stale_overflow(ldata, ns)))
3297 GOTO(unlock, rc = 0);
3299 ldata->ld_leh->leh_overflow_time = 0;
3300 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
3301 GOTO(unlock, rc = 1);
3303 /* If all known entries are in the linkEA, then the 'leh_reccount'
3304 * should NOT be zero. */
3305 LASSERT(ldata->ld_leh->leh_reccount > 0);
3307 lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
3308 ldata->ld_leh->leh_len);
3309 rc = dt_xattr_set(env, obj, &linkea_buf, XATTR_NAME_LINK, 0, th);
3310 if (unlikely(rc == -ENOSPC))
3318 dt_write_unlock(env, obj);
3321 dt_trans_stop(env, dev, th);
3324 lfsck_ibits_unlock(&lh, LCK_PW);
3325 CDEBUG(D_LFSCK, "%s: clear linkea overflow timestamp for the object "
3327 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
3333 * Double scan the MDT-object for namespace LFSCK.
3335 * If the MDT-object contains invalid or repeated linkEA entries, then drop
3336 * those entries from the linkEA; if the linkEA becomes empty or the object
3337 * has no linkEA, then it is an orphan and will be added into the directory
3338 * .lustre/lost+found/MDTxxxx/; if the remote parent is lost, then recreate
3339 * the remote parent; if the name entry corresponding to some linkEA entry
3340 * is lost, then add the name entry back to the namespace.
3342 * \param[in] env pointer to the thread context
3343 * \param[in] com pointer to the lfsck component
3344 * \param[in] child pointer to the dt_object to be handled
3345 * \param[in] flags some hints to indicate how the @child should be handled
3347 * \retval positive number for repaired cases
3348 * \retval 0 if nothing to be repaired
3349 * \retval negative error number on failure
3351 static int lfsck_namespace_double_scan_one(const struct lu_env *env,
3352 struct lfsck_component *com,
3353 struct dt_object *child, __u8 flags)
3355 struct lfsck_thread_info *info = lfsck_env_info(env);
3356 struct lu_attr *la = &info->lti_la;
3357 struct lu_name *cname = &info->lti_name;
3358 struct lu_fid *pfid = &info->lti_fid;
3359 struct lu_fid *cfid = &info->lti_fid2;
3360 struct lfsck_instance *lfsck = com->lc_lfsck;
3361 struct lfsck_namespace *ns = com->lc_file_ram;
3362 struct dt_object *parent = NULL;
3363 struct linkea_data ldata = { NULL };
3364 bool repaired = false;
3369 dt_read_lock(env, child, 0);
3370 if (unlikely(lfsck_is_dead_obj(child))) {
3371 dt_read_unlock(env, child);
3376 if (S_ISDIR(lfsck_object_type(child))) {
3377 dt_read_unlock(env, child);
3378 rc = lfsck_namespace_double_scan_dir(env, com, child, flags);
3383 rc = lfsck_links_read(env, child, &ldata);
3384 dt_read_unlock(env, child);
3386 if (rc == -EINVAL) {
3387 struct lustre_handle lh = { 0 };
3389 rc = lfsck_ibits_lock(env, com->lc_lfsck, child, &lh,
3390 MDS_INODELOCK_UPDATE |
3391 MDS_INODELOCK_XATTR, LCK_EX);
3393 rc = lfsck_namespace_links_remove(env, com, child);
3394 lfsck_ibits_unlock(&lh, LCK_EX);
3403 if (!(ns->ln_flags & LF_INCOMPLETE) &&
3404 unlikely(lfsck_namespace_linkea_stale_overflow(&ldata, ns))) {
3405 rc = lfsck_namespace_linkea_clear_overflow(env, com, &ldata,
3411 ns->ln_linkea_overflow_cleared++;
3414 linkea_first_entry(&ldata);
3415 while (ldata.ld_lee != NULL) {
3416 rc = lfsck_namespace_unpack_linkea_entry(&ldata, cname, pfid,
3418 sizeof(info->lti_key));
3419 /* Invalid PFID in the linkEA entry. */
3421 rc = lfsck_namespace_shrink_linkea(env, com, child,
3422 &ldata, cname, pfid, true);
3432 rc = lfsck_namespace_filter_linkea_entry(&ldata, cname, pfid,
3434 /* Found repeated linkEA entries */
3436 rc = lfsck_namespace_shrink_linkea(env, com, child,
3437 &ldata, cname, pfid, false);
3449 parent = lfsck_object_find_bottom(env, lfsck, pfid);
3451 GOTO(out, rc = PTR_ERR(parent));
3453 if (!dt_object_exists(parent)) {
3456 if (ldata.ld_leh->leh_reccount > 1) {
3457 /* If it is NOT the last linkEA entry, then
3458 * there is still other chance to make the
3459 * child to be visible via other parent, then
3460 * remove this linkEA entry. */
3461 rc = lfsck_namespace_shrink_linkea(env, com,
3462 child, &ldata, cname, pfid, true);
3464 /* If the LFSCK is marked as LF_INCOMPLETE,
3465 * then means some MDT has ever tried to
3466 * verify some remote MDT-object that resides
3467 * on this MDT, but this MDT failed to respond
3468 * such request. So means there may be some
3469 * remote name entry on other MDT that
3470 * references this object with another name,
3471 * so we cannot know whether this linkEA is
3472 * valid or not. So keep it there and maybe
3473 * resolved when next LFSCK run. */
3474 if (ns->ln_flags & LF_INCOMPLETE) {
3475 lfsck_object_put(env, parent);
3480 /* Create the lost parent as an orphan. */
3481 rc = lfsck_namespace_create_orphan_dir(env, com,
3484 lfsck_object_put(env, parent);
3492 /* Add the missing name entry to the parent. */
3493 rc = lfsck_namespace_insert_normal(env, com,
3494 parent, child, cname->ln_name);
3495 if (unlikely(rc == -EEXIST))
3496 /* Unfortunately, someone reused the
3497 * name under the parent by race. So we
3498 * have to remove the linkEA entry from
3499 * current child object. It means that
3500 * the LFSCK cannot recover the system
3501 * totally back to its original status,
3502 * but it is necessary to make the
3503 * current system to be consistent. */
3504 rc = lfsck_namespace_shrink_linkea(env,
3508 linkea_next_entry(&ldata);
3511 lfsck_object_put(env, parent);
3519 } /* !dt_object_exists(parent) */
3521 /* The linkEA entry with bad parent will be removed. */
3522 if (unlikely(!dt_try_as_dir(env, parent))) {
3523 lfsck_object_put(env, parent);
3524 rc = lfsck_namespace_shrink_linkea(env, com, child,
3525 &ldata, cname, pfid, true);
3535 rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
3536 (const struct dt_key *)cname->ln_name);
3537 if (rc != 0 && rc != -ENOENT) {
3538 lfsck_object_put(env, parent);
3544 if (lu_fid_eq(cfid, lfsck_dto2fid(child))) {
3545 /* It is the most common case that we
3546 * find the name entry corresponding
3547 * to the linkEA entry. */
3548 lfsck_object_put(env, parent);
3549 linkea_next_entry(&ldata);
3551 /* The name entry references another
3552 * MDT-object that may be created by
3553 * the LFSCK for repairing dangling
3554 * name entry. Try to replace it. */
3555 rc = lfsck_namespace_replace_cond(env, com,
3556 parent, child, cfid, cname);
3557 lfsck_object_put(env, parent);
3563 linkea_next_entry(&ldata);
3565 rc = lfsck_namespace_shrink_linkea(env,
3579 /* The following handles -ENOENT case */
3581 rc = dt_attr_get(env, child, la);
3585 /* If there is no name entry in the parent dir and the object
3586 * link count is fewer than the linkea entries count, then the
3587 * linkea entry should be removed. */
3588 if (ldata.ld_leh->leh_reccount > la->la_nlink) {
3589 rc = lfsck_namespace_shrink_linkea_cond(env, com,
3590 parent, child, &ldata, cname, pfid);
3591 lfsck_object_put(env, parent);
3601 /* If the LFSCK is marked as LF_INCOMPLETE, then means some
3602 * MDT has ever tried to verify some remote MDT-object that
3603 * resides on this MDT, but this MDT failed to respond such
3604 * request. So means there may be some remote name entry on
3605 * other MDT that references this object with another name,
3606 * so we cannot know whether this linkEA is valid or not.
3607 * So keep it there and maybe resolved when next LFSCK run. */
3608 if (ns->ln_flags & LF_INCOMPLETE) {
3609 lfsck_object_put(env, parent);
3614 rc = lfsck_namespace_check_name(env, parent, child, cname);
3619 lfsck_object_put(env, parent);
3624 /* It is an invalid name entry, drop it. */
3625 if (unlikely(rc > 0)) {
3626 lfsck_object_put(env, parent);
3627 rc = lfsck_namespace_shrink_linkea(env, com, child,
3628 &ldata, cname, pfid, true);
3638 /* Add the missing name entry back to the namespace. */
3639 rc = lfsck_namespace_insert_normal(env, com, parent, child,
3641 if (unlikely(rc == -ESTALE))
3642 /* It may happen when the remote object has been
3643 * removed, but the local MDT is not aware of that. */
3646 if (unlikely(rc == -EEXIST))
3647 /* Unfortunately, someone reused the name under the
3648 * parent by race. So we have to remove the linkEA
3649 * entry from current child object. It means that the
3650 * LFSCK cannot recover the system totally back to
3651 * its original status, but it is necessary to make
3652 * the current system to be consistent.
3654 * It also may be because of the LFSCK found some
3655 * internal status of create operation. Under such
3656 * case, nothing to be done. */
3657 rc = lfsck_namespace_shrink_linkea_cond(env, com,
3658 parent, child, &ldata, cname, pfid);
3660 linkea_next_entry(&ldata);
3662 lfsck_object_put(env, parent);
3673 if (rc < 0 && rc != -ENODATA)
3676 if (rc == 0 && ldata.ld_leh != NULL)
3677 count = ldata.ld_leh->leh_reccount;
3680 /* If the LFSCK is marked as LF_INCOMPLETE, then means some
3681 * MDT has ever tried to verify some remote MDT-object that
3682 * resides on this MDT, but this MDT failed to respond such
3683 * request. So means there may be some remote name entry on
3684 * other MDT that references this object with another name,
3685 * so we cannot know whether this linkEA is valid or not.
3686 * So keep it there and maybe resolved when next LFSCK run. */
3687 if (!(ns->ln_flags & LF_INCOMPLETE) &&
3688 (ldata.ld_leh == NULL ||
3689 !ldata.ld_leh->leh_overflow_time)) {
3690 /* If the child becomes orphan, then insert it into
3691 * the global .lustre/lost+found/MDTxxxx directory. */
3692 rc = lfsck_namespace_insert_orphan(env, com, child,
3698 ns->ln_mul_ref_repaired++;
3703 rc = dt_attr_get(env, child, la);
3707 if (la->la_nlink != 0 && la->la_nlink != count) {
3708 if (unlikely(!S_ISREG(lfsck_object_type(child)) &&
3709 !S_ISLNK(lfsck_object_type(child)))) {
3710 CDEBUG(D_LFSCK, "%s: namespace LFSCK finds "
3711 "the object "DFID"'s nlink count %d "
3712 "does not match linkEA count %d, "
3713 "type %o, skip it.\n",
3714 lfsck_lfsck2name(lfsck),
3715 PFID(lfsck_dto2fid(child)),
3716 la->la_nlink, count,
3717 lfsck_object_type(child));
3718 } else if (la->la_nlink < count &&
3719 likely(!ldata.ld_leh->leh_overflow_time)) {
3720 rc = lfsck_namespace_repair_nlink(env, com,
3723 ns->ln_objs_nlink_repaired++;
3731 if (la->la_nlink > 1)
3732 ns->ln_mul_linked_repaired++;
3741 static void lfsck_namespace_dump_statistics(struct seq_file *m,
3742 struct lfsck_namespace *ns,
3743 __u64 checked_phase1,
3744 __u64 checked_phase2,
3745 time64_t time_phase1,
3746 time64_t time_phase2, bool dryrun)
3748 const char *postfix = dryrun ? "inconsistent" : "repaired";
3750 seq_printf(m, "checked_phase1: %llu\n"
3751 "checked_phase2: %llu\n"
3754 "failed_phase1: %llu\n"
3755 "failed_phase2: %llu\n"
3756 "directories: %llu\n"
3760 "multiple_linked_checked: %llu\n"
3761 "multiple_linked_%s: %llu\n"
3762 "unknown_inconsistency: %llu\n"
3763 "unmatched_pairs_%s: %llu\n"
3764 "dangling_%s: %llu\n"
3765 "multiple_referenced_%s: %llu\n"
3766 "bad_file_type_%s: %llu\n"
3767 "lost_dirent_%s: %llu\n"
3768 "local_lost_found_scanned: %llu\n"
3769 "local_lost_found_moved: %llu\n"
3770 "local_lost_found_skipped: %llu\n"
3771 "local_lost_found_failed: %llu\n"
3772 "striped_dirs_scanned: %llu\n"
3773 "striped_dirs_%s: %llu\n"
3774 "striped_dirs_failed: %llu\n"
3775 "striped_dirs_disabled: %llu\n"
3776 "striped_dirs_skipped: %llu\n"
3777 "striped_shards_scanned: %llu\n"
3778 "striped_shards_%s: %llu\n"
3779 "striped_shards_failed: %llu\n"
3780 "striped_shards_skipped: %llu\n"
3781 "name_hash_%s: %llu\n"
3782 "linkea_overflow_%s: %llu\n"
3783 "success_count: %u\n"
3784 "run_time_phase1: %lld seconds\n"
3785 "run_time_phase2: %lld seconds\n",
3788 dryrun ? "inconsistent" : "updated",
3789 ns->ln_items_repaired,
3790 dryrun ? "inconsistent" : "updated",
3791 ns->ln_objs_repaired_phase2,
3792 ns->ln_items_failed,
3793 ns->ln_objs_failed_phase2,
3794 ns->ln_dirs_checked,
3795 postfix, ns->ln_dirent_repaired,
3796 postfix, ns->ln_linkea_repaired,
3797 postfix, ns->ln_objs_nlink_repaired,
3798 ns->ln_mul_linked_checked,
3799 postfix, ns->ln_mul_linked_repaired,
3800 ns->ln_unknown_inconsistency,
3801 postfix, ns->ln_unmatched_pairs_repaired,
3802 postfix, ns->ln_dangling_repaired,
3803 postfix, ns->ln_mul_ref_repaired,
3804 postfix, ns->ln_bad_type_repaired,
3805 postfix, ns->ln_lost_dirent_repaired,
3806 ns->ln_local_lpf_scanned,
3807 ns->ln_local_lpf_moved,
3808 ns->ln_local_lpf_skipped,
3809 ns->ln_local_lpf_failed,
3810 ns->ln_striped_dirs_scanned,
3811 postfix, ns->ln_striped_dirs_repaired,
3812 ns->ln_striped_dirs_failed,
3813 ns->ln_striped_dirs_disabled,
3814 ns->ln_striped_dirs_skipped,
3815 ns->ln_striped_shards_scanned,
3816 postfix, ns->ln_striped_shards_repaired,
3817 ns->ln_striped_shards_failed,
3818 ns->ln_striped_shards_skipped,
3819 postfix, ns->ln_name_hash_repaired,
3820 dryrun ? "inconsistent" : "cleared",
3821 ns->ln_linkea_overflow_cleared,
3822 ns->ln_success_count,
3827 static void lfsck_namespace_release_lmv(const struct lu_env *env,
3828 struct lfsck_component *com)
3830 struct lfsck_instance *lfsck = com->lc_lfsck;
3831 struct lfsck_namespace *ns = com->lc_file_ram;
3833 while (!list_empty(&lfsck->li_list_lmv)) {
3834 struct lfsck_lmv_unit *llu;
3835 struct lfsck_lmv *llmv;
3837 llu = list_entry(lfsck->li_list_lmv.next,
3838 struct lfsck_lmv_unit, llu_link);
3839 llmv = &llu->llu_lmv;
3841 LASSERTF(atomic_read(&llmv->ll_ref) == 1,
3842 "still in using: %u\n",
3843 atomic_read(&llmv->ll_ref));
3845 ns->ln_striped_dirs_skipped++;
3846 lfsck_lmv_put(env, llmv);
3850 static int lfsck_namespace_check_for_double_scan(const struct lu_env *env,
3851 struct lfsck_component *com,
3852 struct dt_object *obj)
3854 struct lu_attr *la = &lfsck_env_info(env)->lti_la;
3857 rc = dt_attr_get(env, obj, la);
3861 /* zero-linkEA object may be orphan, but it also maybe because
3862 * of upgrading. Currently, we cannot record it for double scan.
3863 * Because it may cause the LFSCK trace file to be too large. */
3865 /* "la_ctime" == 1 means that it has ever been removed from
3866 * backend /lost+found directory but not been added back to
3867 * the normal namespace yet. */
3869 if ((S_ISREG(lfsck_object_type(obj)) && la->la_nlink > 1) ||
3870 unlikely(la->la_ctime == 1))
3871 rc = lfsck_namespace_trace_update(env, com, lfsck_dto2fid(obj),
3872 LNTF_CHECK_LINKEA, true);
3877 /* namespace APIs */
3879 static int lfsck_namespace_reset(const struct lu_env *env,
3880 struct lfsck_component *com, bool init)
3882 struct lfsck_instance *lfsck = com->lc_lfsck;
3883 struct lfsck_namespace *ns = com->lc_file_ram;
3884 struct lfsck_assistant_data *lad = com->lc_data;
3885 struct dt_object *root;
3889 root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
3891 GOTO(log, rc = PTR_ERR(root));
3893 if (unlikely(!dt_try_as_dir(env, root)))
3894 GOTO(put, rc = -ENOTDIR);
3896 down_write(&com->lc_sem);
3898 memset(ns, 0, sizeof(*ns));
3900 __u32 count = ns->ln_success_count;
3901 time64_t last_time = ns->ln_time_last_complete;
3903 memset(ns, 0, sizeof(*ns));
3904 ns->ln_success_count = count;
3905 ns->ln_time_last_complete = last_time;
3907 ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
3908 ns->ln_status = LS_INIT;
3909 ns->ln_time_latest_reset = ktime_get_real_seconds();
3911 rc = lfsck_load_one_trace_file(env, com, root, &com->lc_obj,
3912 &dt_lfsck_namespace_features,
3913 LFSCK_NAMESPACE, true);
3917 rc = lfsck_load_sub_trace_files(env, com, &dt_lfsck_namespace_features,
3918 LFSCK_NAMESPACE, true);
3922 lad->lad_incomplete = 0;
3923 CFS_RESET_BITMAP(lad->lad_bitmap);
3925 rc = lfsck_namespace_store(env, com);
3930 up_write(&com->lc_sem);
3933 lfsck_object_put(env, root);
3935 CDEBUG(D_LFSCK, "%s: namespace LFSCK reset: rc = %d\n",
3936 lfsck_lfsck2name(lfsck), rc);
3941 lfsck_namespace_fail(const struct lu_env *env, struct lfsck_component *com,
3944 struct lfsck_namespace *ns = com->lc_file_ram;
3946 down_write(&com->lc_sem);
3948 com->lc_new_checked++;
3949 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
3950 up_write(&com->lc_sem);
3953 static void lfsck_namespace_close_dir(const struct lu_env *env,
3954 struct lfsck_component *com)
3956 struct lfsck_namespace *ns = com->lc_file_ram;
3957 struct lfsck_assistant_data *lad = com->lc_data;
3958 struct lfsck_assistant_object *lso = NULL;
3959 struct lfsck_instance *lfsck = com->lc_lfsck;
3960 struct lfsck_lmv *llmv = lfsck->li_lmv;
3961 struct lfsck_namespace_req *lnr;
3963 sizeof(*lnr) + LFSCK_TMPBUF_LEN;
3964 bool wakeup = false;
3970 OBD_ALLOC(lnr, size);
3972 ns->ln_striped_dirs_skipped++;
3977 lso = lfsck_assistant_object_init(env, lfsck_dto2fid(lfsck->li_obj_dir),
3978 NULL, lfsck->li_pos_current.lp_oit_cookie, true);
3980 OBD_FREE(lnr, size);
3981 ns->ln_striped_dirs_skipped++;
3986 /* Generate a dummy request to indicate that all shards' name entry
3987 * in this striped directory has been scanned for the first time. */
3988 INIT_LIST_HEAD(&lnr->lnr_lar.lar_list);
3989 lnr->lnr_lar.lar_parent = lso;
3990 lnr->lnr_lmv = lfsck_lmv_get(llmv);
3991 lnr->lnr_fid = *lfsck_dto2fid(lfsck->li_obj_dir);
3992 lnr->lnr_dir_cookie = MDS_DIR_END_OFF;
3993 lnr->lnr_size = size;
3995 spin_lock(&lad->lad_lock);
3996 if (lad->lad_assistant_status < 0) {
3997 spin_unlock(&lad->lad_lock);
3998 lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar);
3999 ns->ln_striped_dirs_skipped++;
4004 list_add_tail(&lnr->lnr_lar.lar_list, &lad->lad_req_list);
4005 if (lad->lad_prefetched == 0)
4008 lad->lad_prefetched++;
4009 spin_unlock(&lad->lad_lock);
4011 wake_up_all(&lad->lad_thread.t_ctl_waitq);
4016 static int lfsck_namespace_open_dir(const struct lu_env *env,
4017 struct lfsck_component *com)
4019 struct lfsck_instance *lfsck = com->lc_lfsck;
4020 struct lfsck_namespace *ns = com->lc_file_ram;
4021 struct lfsck_lmv *llmv = lfsck->li_lmv;
4028 if (llmv->ll_lmv_master) {
4029 struct lmv_mds_md_v1 *lmv = &llmv->ll_lmv;
4031 if (lmv->lmv_master_mdt_index != lfsck_dev_idx(lfsck)) {
4032 lmv->lmv_master_mdt_index =
4033 lfsck_dev_idx(lfsck);
4034 ns->ln_flags |= LF_INCONSISTENT;
4035 llmv->ll_lmv_updated = 1;
4038 rc = lfsck_namespace_verify_stripe_slave(env, com,
4039 lfsck->li_obj_dir, llmv);
4042 RETURN(rc > 0 ? 0 : rc);
4045 static int lfsck_namespace_checkpoint(const struct lu_env *env,
4046 struct lfsck_component *com, bool init)
4048 struct lfsck_instance *lfsck = com->lc_lfsck;
4049 struct lfsck_namespace *ns = com->lc_file_ram;
4053 rc = lfsck_checkpoint_generic(env, com);
4058 down_write(&com->lc_sem);
4060 ns->ln_pos_latest_start = lfsck->li_pos_checkpoint;
4062 ns->ln_pos_last_checkpoint = lfsck->li_pos_checkpoint;
4063 ns->ln_run_time_phase1 += ktime_get_seconds() -
4064 lfsck->li_time_last_checkpoint;
4065 ns->ln_time_last_checkpoint = ktime_get_real_seconds();
4066 ns->ln_items_checked += com->lc_new_checked;
4067 com->lc_new_checked = 0;
4070 rc = lfsck_namespace_store(env, com);
4071 up_write(&com->lc_sem);
4074 CDEBUG(D_LFSCK, "%s: namespace LFSCK checkpoint at the pos [%llu"
4075 ", "DFID", %#llx], status = %d: rc = %d\n",
4076 lfsck_lfsck2name(lfsck), lfsck->li_pos_current.lp_oit_cookie,
4077 PFID(&lfsck->li_pos_current.lp_dir_parent),
4078 lfsck->li_pos_current.lp_dir_cookie, ns->ln_status, rc);
4080 return rc > 0 ? 0 : rc;
4083 static int lfsck_namespace_prep(const struct lu_env *env,
4084 struct lfsck_component *com,
4085 struct lfsck_start_param *lsp)
4087 struct lfsck_instance *lfsck = com->lc_lfsck;
4088 struct lfsck_namespace *ns = com->lc_file_ram;
4089 struct lfsck_position *pos = &com->lc_pos_start;
4092 rc = lfsck_namespace_load_bitmap(env, com);
4093 if (rc != 0 || ns->ln_status == LS_COMPLETED) {
4094 rc = lfsck_namespace_reset(env, com, false);
4096 rc = lfsck_set_param(env, lfsck, lsp->lsp_start, true);
4099 CDEBUG(D_LFSCK, "%s: namespace LFSCK prep failed: "
4100 "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
4106 down_write(&com->lc_sem);
4107 ns->ln_time_latest_start = ktime_get_real_seconds();
4108 spin_lock(&lfsck->li_lock);
4110 if (ns->ln_flags & LF_SCANNED_ONCE) {
4111 if (!lfsck->li_drop_dryrun ||
4112 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent)) {
4113 ns->ln_status = LS_SCANNING_PHASE2;
4114 list_move_tail(&com->lc_link,
4115 &lfsck->li_list_double_scan);
4116 if (!list_empty(&com->lc_link_dir))
4117 list_del_init(&com->lc_link_dir);
4118 lfsck_pos_set_zero(pos);
4120 ns->ln_status = LS_SCANNING_PHASE1;
4121 ns->ln_run_time_phase1 = 0;
4122 ns->ln_run_time_phase2 = 0;
4123 ns->ln_items_checked = 0;
4124 ns->ln_items_repaired = 0;
4125 ns->ln_items_failed = 0;
4126 ns->ln_dirs_checked = 0;
4127 ns->ln_objs_checked_phase2 = 0;
4128 ns->ln_objs_repaired_phase2 = 0;
4129 ns->ln_objs_failed_phase2 = 0;
4130 ns->ln_objs_nlink_repaired = 0;
4131 ns->ln_dirent_repaired = 0;
4132 ns->ln_linkea_repaired = 0;
4133 ns->ln_mul_linked_checked = 0;
4134 ns->ln_mul_linked_repaired = 0;
4135 ns->ln_unknown_inconsistency = 0;
4136 ns->ln_unmatched_pairs_repaired = 0;
4137 ns->ln_dangling_repaired = 0;
4138 ns->ln_mul_ref_repaired = 0;
4139 ns->ln_bad_type_repaired = 0;
4140 ns->ln_lost_dirent_repaired = 0;
4141 ns->ln_striped_dirs_scanned = 0;
4142 ns->ln_striped_dirs_repaired = 0;
4143 ns->ln_striped_dirs_failed = 0;
4144 ns->ln_striped_dirs_disabled = 0;
4145 ns->ln_striped_dirs_skipped = 0;
4146 ns->ln_striped_shards_scanned = 0;
4147 ns->ln_striped_shards_repaired = 0;
4148 ns->ln_striped_shards_failed = 0;
4149 ns->ln_striped_shards_skipped = 0;
4150 ns->ln_name_hash_repaired = 0;
4151 fid_zero(&ns->ln_fid_latest_scanned_phase2);
4152 if (list_empty(&com->lc_link_dir))
4153 list_add_tail(&com->lc_link_dir,
4154 &lfsck->li_list_dir);
4155 *pos = ns->ln_pos_first_inconsistent;
4158 ns->ln_status = LS_SCANNING_PHASE1;
4159 if (list_empty(&com->lc_link_dir))
4160 list_add_tail(&com->lc_link_dir,
4161 &lfsck->li_list_dir);
4162 if (!lfsck->li_drop_dryrun ||
4163 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent)) {
4164 *pos = ns->ln_pos_last_checkpoint;
4165 pos->lp_oit_cookie++;
4167 *pos = ns->ln_pos_first_inconsistent;
4171 spin_unlock(&lfsck->li_lock);
4172 up_write(&com->lc_sem);
4174 rc = lfsck_start_assistant(env, com, lsp);
4176 CDEBUG(D_LFSCK, "%s: namespace LFSCK prep done, start pos [%llu, "
4177 DFID", %#llx]: rc = %d\n",
4178 lfsck_lfsck2name(lfsck), pos->lp_oit_cookie,
4179 PFID(&pos->lp_dir_parent), pos->lp_dir_cookie, rc);
4184 static int lfsck_namespace_exec_oit(const struct lu_env *env,
4185 struct lfsck_component *com,
4186 struct dt_object *obj)
4188 struct lfsck_thread_info *info = lfsck_env_info(env);
4189 struct lfsck_namespace *ns = com->lc_file_ram;
4190 struct lfsck_instance *lfsck = com->lc_lfsck;
4191 const struct lu_fid *fid = lfsck_dto2fid(obj);
4192 struct lu_fid *pfid = &info->lti_fid2;
4193 struct lu_name *cname = &info->lti_name;
4194 struct lu_seq_range *range = &info->lti_range;
4195 struct seq_server_site *ss = lfsck_dev_site(lfsck);
4196 struct linkea_data ldata = { NULL };
4197 __u32 idx = lfsck_dev_idx(lfsck);
4201 rc = lfsck_links_read(env, obj, &ldata);
4205 /* -EINVAL means crashed linkEA, should be verified. */
4206 if (rc == -EINVAL) {
4207 rc = lfsck_namespace_trace_update(env, com, fid,
4208 LNTF_CHECK_LINKEA, true);
4210 struct lustre_handle lh = { 0 };
4212 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
4213 MDS_INODELOCK_UPDATE |
4214 MDS_INODELOCK_XATTR, LCK_EX);
4216 rc = lfsck_namespace_links_remove(env, com,
4218 lfsck_ibits_unlock(&lh, LCK_EX);
4222 GOTO(out, rc = (rc == -ENOENT ? 0 : rc));
4225 if (rc == -ENODATA || unlikely(!ldata.ld_leh->leh_reccount)) {
4226 rc = lfsck_namespace_check_for_double_scan(env, com, obj);
4234 /* Record multiple-linked object. */
4235 if (ldata.ld_leh->leh_reccount > 1) {
4236 rc = lfsck_namespace_trace_update(env, com, fid,
4237 LNTF_CHECK_LINKEA, true);
4242 linkea_first_entry(&ldata);
4243 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, cname, pfid);
4244 if (!fid_is_sane(pfid)) {
4245 rc = lfsck_namespace_trace_update(env, com, fid,
4246 LNTF_CHECK_PARENT, true);
4248 fld_range_set_mdt(range);
4249 rc = fld_local_lookup(env, ss->ss_server_fld,
4250 fid_seq(pfid), range);
4251 if ((rc == -ENOENT) ||
4252 (rc == 0 && range->lsr_index != idx))
4253 rc = lfsck_namespace_trace_update(env, com, fid,
4254 LNTF_CHECK_LINKEA, true);
4256 rc = lfsck_namespace_check_for_double_scan(env, com,
4263 down_write(&com->lc_sem);
4264 if (S_ISDIR(lfsck_object_type(obj)))
4265 ns->ln_dirs_checked++;
4267 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
4268 up_write(&com->lc_sem);
4273 static int lfsck_namespace_exec_dir(const struct lu_env *env,
4274 struct lfsck_component *com,
4275 struct lfsck_assistant_object *lso,
4276 struct lu_dirent *ent, __u16 type)
4278 struct lfsck_assistant_data *lad = com->lc_data;
4279 struct lfsck_instance *lfsck = com->lc_lfsck;
4280 struct lfsck_namespace_req *lnr;
4281 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4282 struct ptlrpc_thread *mthread = &lfsck->li_thread;
4283 struct ptlrpc_thread *athread = &lad->lad_thread;
4284 struct l_wait_info lwi = { 0 };
4285 bool wakeup = false;
4287 l_wait_event(mthread->t_ctl_waitq,
4288 lad->lad_prefetched < bk->lb_async_windows ||
4289 !thread_is_running(mthread) ||
4290 thread_is_stopped(athread),
4293 if (unlikely(!thread_is_running(mthread)) ||
4294 thread_is_stopped(athread))
4297 if (unlikely(lfsck_is_dead_obj(lfsck->li_obj_dir)))
4300 lnr = lfsck_namespace_assistant_req_init(com->lc_lfsck, lso, ent, type);
4302 struct lfsck_namespace *ns = com->lc_file_ram;
4304 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
4305 return PTR_ERR(lnr);
4308 spin_lock(&lad->lad_lock);
4309 if (lad->lad_assistant_status < 0) {
4310 spin_unlock(&lad->lad_lock);
4311 lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar);
4312 return lad->lad_assistant_status;
4315 list_add_tail(&lnr->lnr_lar.lar_list, &lad->lad_req_list);
4316 if (lad->lad_prefetched == 0)
4319 lad->lad_prefetched++;
4320 spin_unlock(&lad->lad_lock);
4322 wake_up_all(&lad->lad_thread.t_ctl_waitq);
4324 down_write(&com->lc_sem);
4325 com->lc_new_checked++;
4326 up_write(&com->lc_sem);
4331 static int lfsck_namespace_post(const struct lu_env *env,
4332 struct lfsck_component *com,
4333 int result, bool init)
4335 struct lfsck_instance *lfsck = com->lc_lfsck;
4336 struct lfsck_namespace *ns = com->lc_file_ram;
4340 lfsck_post_generic(env, com, &result);
4342 down_write(&com->lc_sem);
4343 lfsck_namespace_release_lmv(env, com);
4345 spin_lock(&lfsck->li_lock);
4347 ns->ln_pos_last_checkpoint = lfsck->li_pos_checkpoint;
4349 ns->ln_status = LS_SCANNING_PHASE2;
4350 ns->ln_flags |= LF_SCANNED_ONCE;
4351 ns->ln_flags &= ~LF_UPGRADE;
4352 list_del_init(&com->lc_link_dir);
4353 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
4354 } else if (result == 0) {
4355 if (lfsck->li_status != 0)
4356 ns->ln_status = lfsck->li_status;
4358 ns->ln_status = LS_STOPPED;
4359 if (ns->ln_status != LS_PAUSED) {
4360 list_del_init(&com->lc_link_dir);
4361 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4364 ns->ln_status = LS_FAILED;
4365 list_del_init(&com->lc_link_dir);
4366 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4368 spin_unlock(&lfsck->li_lock);
4371 ns->ln_run_time_phase1 += ktime_get_seconds() -
4372 lfsck->li_time_last_checkpoint;
4373 ns->ln_time_last_checkpoint = ktime_get_real_seconds();
4374 ns->ln_items_checked += com->lc_new_checked;
4375 com->lc_new_checked = 0;
4378 rc = lfsck_namespace_store(env, com);
4379 up_write(&com->lc_sem);
4381 CDEBUG(D_LFSCK, "%s: namespace LFSCK post done: rc = %d\n",
4382 lfsck_lfsck2name(lfsck), rc);
4388 lfsck_namespace_dump(const struct lu_env *env, struct lfsck_component *com,
4391 struct lfsck_instance *lfsck = com->lc_lfsck;
4392 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4393 struct lfsck_namespace *ns = com->lc_file_ram;
4395 down_read(&com->lc_sem);
4396 seq_printf(m, "name: lfsck_namespace\n"
4402 lfsck_status2name(ns->ln_status));
4404 lfsck_bits_dump(m, ns->ln_flags, lfsck_flags_names, "flags");
4406 lfsck_bits_dump(m, bk->lb_param, lfsck_param_names, "param");
4408 lfsck_time_dump(m, ns->ln_time_last_complete, "last_completed");
4410 lfsck_time_dump(m, ns->ln_time_latest_start, "latest_start");
4412 lfsck_time_dump(m, ns->ln_time_last_checkpoint, "last_checkpoint");
4414 lfsck_pos_dump(m, &ns->ln_pos_latest_start, "latest_start_position");
4416 lfsck_pos_dump(m, &ns->ln_pos_last_checkpoint,
4417 "last_checkpoint_position");
4419 lfsck_pos_dump(m, &ns->ln_pos_first_inconsistent,
4420 "first_failure_position");
4422 if (ns->ln_status == LS_SCANNING_PHASE1) {
4423 struct lfsck_position pos;
4424 time64_t duration = ktime_get_seconds() -
4425 lfsck->li_time_last_checkpoint;
4426 u64 checked = ns->ln_items_checked + com->lc_new_checked;
4427 u64 speed = checked;
4428 u64 new_checked = com->lc_new_checked;
4429 time64_t rtime = ns->ln_run_time_phase1 + duration;
4432 div_u64(new_checked, duration);
4435 div_u64(speed, rtime);
4437 lfsck_namespace_dump_statistics(m, ns, checked, 0, rtime, 0,
4438 bk->lb_param & LPF_DRYRUN);
4439 seq_printf(m, "average_speed_phase1: %llu items/sec\n"
4440 "average_speed_phase2: N/A\n"
4441 "average_speed_total: %llu items/sec\n"
4442 "real_time_speed_phase1: %llu items/sec\n"
4443 "real_time_speed_phase2: N/A\n",
4448 if (likely(lfsck->li_di_oit)) {
4449 const struct dt_it_ops *iops =
4450 &lfsck->li_obj_oit->do_index_ops->dio_it;
4452 /* The low layer otable-based iteration position may NOT
4453 * exactly match the namespace-based directory traversal
4454 * cookie. Generally, it is not a serious issue. But the
4455 * caller should NOT make assumption on that. */
4456 pos.lp_oit_cookie = iops->store(env, lfsck->li_di_oit);
4457 if (!lfsck->li_current_oit_processed)
4458 pos.lp_oit_cookie--;
4460 spin_lock(&lfsck->li_lock);
4461 if (lfsck->li_di_dir) {
4462 pos.lp_dir_cookie = lfsck->li_cookie_dir;
4463 if (pos.lp_dir_cookie >= MDS_DIR_END_OFF) {
4464 fid_zero(&pos.lp_dir_parent);
4465 pos.lp_dir_cookie = 0;
4468 *lfsck_dto2fid(lfsck->li_obj_dir);
4471 fid_zero(&pos.lp_dir_parent);
4472 pos.lp_dir_cookie = 0;
4474 spin_unlock(&lfsck->li_lock);
4476 pos = ns->ln_pos_last_checkpoint;
4479 lfsck_pos_dump(m, &pos, "current_position");
4480 } else if (ns->ln_status == LS_SCANNING_PHASE2) {
4481 time64_t duration = ktime_get_seconds() -
4482 com->lc_time_last_checkpoint;
4483 __u64 checked = ns->ln_objs_checked_phase2 +
4484 com->lc_new_checked;
4485 __u64 speed1 = ns->ln_items_checked;
4486 __u64 speed2 = checked;
4487 __u64 speed0 = speed1 + speed2;
4488 __u64 new_checked = com->lc_new_checked;
4489 time64_t rtime = ns->ln_run_time_phase2 + duration;
4490 time64_t time0 = ns->ln_run_time_phase1 + rtime;
4493 div_u64(new_checked, duration);
4495 if (ns->ln_run_time_phase1 != 0)
4496 div_u64(speed1, ns->ln_run_time_phase1);
4497 else if (ns->ln_items_checked != 0)
4501 div_u64(speed2, rtime);
4502 else if (checked != 0)
4506 div_u64(speed0, time0);
4508 lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked,
4510 ns->ln_run_time_phase1, rtime,
4511 bk->lb_param & LPF_DRYRUN);
4512 seq_printf(m, "average_speed_phase1: %llu items/sec\n"
4513 "average_speed_phase2: %llu objs/sec\n"
4514 "average_speed_total: %llu items/sec\n"
4515 "real_time_speed_phase1: N/A\n"
4516 "real_time_speed_phase2: %llu objs/sec\n"
4517 "current_position: "DFID"\n",
4522 PFID(&ns->ln_fid_latest_scanned_phase2));
4524 __u64 speed1 = ns->ln_items_checked;
4525 __u64 speed2 = ns->ln_objs_checked_phase2;
4526 __u64 speed0 = speed1 + speed2;
4527 time64_t time0 = ns->ln_run_time_phase1 + ns->ln_run_time_phase2;
4529 if (ns->ln_run_time_phase1 != 0)
4530 div_u64(speed1, ns->ln_run_time_phase1);
4531 else if (ns->ln_items_checked != 0)
4534 if (ns->ln_run_time_phase2 != 0)
4535 div_u64(speed2, ns->ln_run_time_phase2);
4536 else if (ns->ln_objs_checked_phase2 != 0)
4540 div_u64(speed0, time0);
4542 lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked,
4543 ns->ln_objs_checked_phase2,
4544 ns->ln_run_time_phase1,
4545 ns->ln_run_time_phase2,
4546 bk->lb_param & LPF_DRYRUN);
4547 seq_printf(m, "average_speed_phase1: %llu items/sec\n"
4548 "average_speed_phase2: %llu objs/sec\n"
4549 "average_speed_total: %llu items/sec\n"
4550 "real_time_speed_phase1: N/A\n"
4551 "real_time_speed_phase2: N/A\n"
4552 "current_position: N/A\n",
4558 up_read(&com->lc_sem);
4561 static int lfsck_namespace_double_scan(const struct lu_env *env,
4562 struct lfsck_component *com)
4564 struct lfsck_namespace *ns = com->lc_file_ram;
4565 struct lfsck_assistant_data *lad = com->lc_data;
4566 struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
4567 struct lfsck_tgt_desc *ltd;
4568 struct lfsck_tgt_desc *next;
4571 rc = lfsck_double_scan_generic(env, com, ns->ln_status);
4572 if (thread_is_stopped(&lad->lad_thread)) {
4573 LASSERT(list_empty(&lad->lad_req_list));
4574 LASSERT(list_empty(&lad->lad_mdt_phase1_list));
4576 spin_lock(<ds->ltd_lock);
4577 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4578 ltd_namespace_phase_list) {
4579 list_del_init(<d->ltd_namespace_phase_list);
4581 spin_unlock(<ds->ltd_lock);
4587 static void lfsck_namespace_data_release(const struct lu_env *env,
4588 struct lfsck_component *com)
4590 struct lfsck_assistant_data *lad = com->lc_data;
4591 struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
4592 struct lfsck_tgt_desc *ltd;
4593 struct lfsck_tgt_desc *next;
4595 LASSERT(lad != NULL);
4596 LASSERT(thread_is_init(&lad->lad_thread) ||
4597 thread_is_stopped(&lad->lad_thread));
4598 LASSERT(list_empty(&lad->lad_req_list));
4600 com->lc_data = NULL;
4601 lfsck_namespace_release_lmv(env, com);
4603 spin_lock(<ds->ltd_lock);
4604 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
4605 ltd_namespace_phase_list) {
4606 list_del_init(<d->ltd_namespace_phase_list);
4608 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4609 ltd_namespace_phase_list) {
4610 list_del_init(<d->ltd_namespace_phase_list);
4612 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_list,
4613 ltd_namespace_list) {
4614 list_del_init(<d->ltd_namespace_list);
4616 spin_unlock(<ds->ltd_lock);
4618 if (likely(lad->lad_bitmap != NULL))
4619 CFS_FREE_BITMAP(lad->lad_bitmap);
4624 static void lfsck_namespace_quit(const struct lu_env *env,
4625 struct lfsck_component *com)
4627 struct lfsck_assistant_data *lad = com->lc_data;
4628 struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
4629 struct lfsck_tgt_desc *ltd;
4630 struct lfsck_tgt_desc *next;
4632 LASSERT(lad != NULL);
4634 lfsck_quit_generic(env, com);
4636 LASSERT(thread_is_init(&lad->lad_thread) ||
4637 thread_is_stopped(&lad->lad_thread));
4638 LASSERT(list_empty(&lad->lad_req_list));
4640 lfsck_namespace_release_lmv(env, com);
4642 spin_lock(<ds->ltd_lock);
4643 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
4644 ltd_namespace_phase_list) {
4645 list_del_init(<d->ltd_namespace_phase_list);
4647 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4648 ltd_namespace_phase_list) {
4649 list_del_init(<d->ltd_namespace_phase_list);
4651 spin_unlock(<ds->ltd_lock);
4654 static int lfsck_namespace_in_notify(const struct lu_env *env,
4655 struct lfsck_component *com,
4656 struct lfsck_request *lr)
4658 struct lfsck_instance *lfsck = com->lc_lfsck;
4659 struct lfsck_namespace *ns = com->lc_file_ram;
4660 struct lfsck_assistant_data *lad = com->lc_data;
4661 struct lfsck_tgt_descs *ltds = &lfsck->li_mdt_descs;
4662 struct lfsck_tgt_desc *ltd;
4667 switch (lr->lr_event) {
4668 case LE_SET_LMV_MASTER: {
4669 struct dt_object *obj;
4671 obj = lfsck_object_find_bottom(env, lfsck, &lr->lr_fid);
4673 RETURN(PTR_ERR(obj));
4675 if (likely(dt_object_exists(obj)))
4676 rc = lfsck_namespace_notify_lmv_master_local(env, com,
4679 lfsck_object_put(env, obj);
4681 RETURN(rc > 0 ? 0 : rc);
4683 case LE_SET_LMV_SLAVE: {
4684 if (!(lr->lr_flags & LEF_RECHECK_NAME_HASH))
4685 ns->ln_striped_shards_repaired++;
4687 rc = lfsck_namespace_trace_update(env, com, &lr->lr_fid,
4688 LNTF_RECHECK_NAME_HASH, true);
4690 RETURN(rc > 0 ? 0 : rc);
4692 case LE_PHASE1_DONE:
4693 case LE_PHASE2_DONE:
4700 CDEBUG(D_LFSCK, "%s: namespace LFSCK handles notify %u from MDT %x, "
4701 "status %d, flags %x\n", lfsck_lfsck2name(lfsck), lr->lr_event,
4702 lr->lr_index, lr->lr_status, lr->lr_flags2);
4704 spin_lock(<ds->ltd_lock);
4705 ltd = lfsck_ltd2tgt(ltds, lr->lr_index);
4707 spin_unlock(<ds->ltd_lock);
4712 list_del_init(<d->ltd_namespace_phase_list);
4713 switch (lr->lr_event) {
4714 case LE_PHASE1_DONE:
4715 if (lr->lr_status <= 0) {
4716 ltd->ltd_namespace_done = 1;
4717 list_del_init(<d->ltd_namespace_list);
4718 CDEBUG(D_LFSCK, "%s: MDT %x failed/stopped at "
4719 "phase1 for namespace LFSCK: rc = %d.\n",
4720 lfsck_lfsck2name(lfsck),
4721 ltd->ltd_index, lr->lr_status);
4722 ns->ln_flags |= LF_INCOMPLETE;
4727 if (lr->lr_flags2 & LF_INCOMPLETE)
4728 ns->ln_flags |= LF_INCOMPLETE;
4730 if (list_empty(<d->ltd_namespace_list))
4731 list_add_tail(<d->ltd_namespace_list,
4732 &lad->lad_mdt_list);
4733 list_add_tail(<d->ltd_namespace_phase_list,
4734 &lad->lad_mdt_phase2_list);
4736 case LE_PHASE2_DONE:
4737 ltd->ltd_namespace_done = 1;
4738 list_del_init(<d->ltd_namespace_list);
4742 ltd->ltd_namespace_done = 1;
4743 list_del_init(<d->ltd_namespace_list);
4744 if (!(lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT)) {
4746 "%s: the peer MDT %x exit namespace LFSCK\n",
4747 lfsck_lfsck2name(lfsck), ltd->ltd_index);
4748 ns->ln_flags |= LF_INCOMPLETE;
4754 spin_unlock(<ds->ltd_lock);
4756 if (fail && lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) {
4757 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
4759 memset(stop, 0, sizeof(*stop));
4760 stop->ls_status = lr->lr_status;
4761 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
4762 lfsck_stop(env, lfsck->li_bottom, stop);
4763 } else if (lfsck_phase2_next_ready(lad)) {
4764 wake_up_all(&lad->lad_thread.t_ctl_waitq);
4770 static void lfsck_namespace_repaired(struct lfsck_namespace *ns, __u64 *count)
4772 *count += ns->ln_objs_nlink_repaired;
4773 *count += ns->ln_dirent_repaired;
4774 *count += ns->ln_linkea_repaired;
4775 *count += ns->ln_mul_linked_repaired;
4776 *count += ns->ln_unmatched_pairs_repaired;
4777 *count += ns->ln_dangling_repaired;
4778 *count += ns->ln_mul_ref_repaired;
4779 *count += ns->ln_bad_type_repaired;
4780 *count += ns->ln_lost_dirent_repaired;
4781 *count += ns->ln_striped_dirs_disabled;
4782 *count += ns->ln_striped_dirs_repaired;
4783 *count += ns->ln_striped_shards_repaired;
4784 *count += ns->ln_name_hash_repaired;
4785 *count += ns->ln_local_lpf_moved;
4788 static int lfsck_namespace_query_all(const struct lu_env *env,
4789 struct lfsck_component *com,
4790 __u32 *mdts_count, __u64 *repaired)
4792 struct lfsck_namespace *ns = com->lc_file_ram;
4793 struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
4794 struct lfsck_tgt_desc *ltd;
4799 rc = lfsck_query_all(env, com);
4803 down_read(<ds->ltd_rw_sem);
4804 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
4805 ltd = lfsck_ltd2tgt(ltds, idx);
4806 LASSERT(ltd != NULL);
4808 mdts_count[ltd->ltd_namespace_status]++;
4809 *repaired += ltd->ltd_namespace_repaired;
4811 up_read(<ds->ltd_rw_sem);
4813 down_read(&com->lc_sem);
4814 mdts_count[ns->ln_status]++;
4815 lfsck_namespace_repaired(ns, repaired);
4816 up_read(&com->lc_sem);
4821 static int lfsck_namespace_query(const struct lu_env *env,
4822 struct lfsck_component *com,
4823 struct lfsck_request *req,
4824 struct lfsck_reply *rep,
4825 struct lfsck_query *que, int idx)
4827 struct lfsck_namespace *ns = com->lc_file_ram;
4831 LASSERT(com->lc_lfsck->li_master);
4833 rc = lfsck_namespace_query_all(env, com,
4834 que->lu_mdts_count[idx],
4835 &que->lu_repaired[idx]);
4837 down_read(&com->lc_sem);
4838 rep->lr_status = ns->ln_status;
4839 if (req->lr_flags & LEF_QUERY_ALL)
4840 lfsck_namespace_repaired(ns, &rep->lr_repaired);
4841 up_read(&com->lc_sem);
4847 static struct lfsck_operations lfsck_namespace_ops = {
4848 .lfsck_reset = lfsck_namespace_reset,
4849 .lfsck_fail = lfsck_namespace_fail,
4850 .lfsck_close_dir = lfsck_namespace_close_dir,
4851 .lfsck_open_dir = lfsck_namespace_open_dir,
4852 .lfsck_checkpoint = lfsck_namespace_checkpoint,
4853 .lfsck_prep = lfsck_namespace_prep,
4854 .lfsck_exec_oit = lfsck_namespace_exec_oit,
4855 .lfsck_exec_dir = lfsck_namespace_exec_dir,
4856 .lfsck_post = lfsck_namespace_post,
4857 .lfsck_dump = lfsck_namespace_dump,
4858 .lfsck_double_scan = lfsck_namespace_double_scan,
4859 .lfsck_data_release = lfsck_namespace_data_release,
4860 .lfsck_quit = lfsck_namespace_quit,
4861 .lfsck_in_notify = lfsck_namespace_in_notify,
4862 .lfsck_query = lfsck_namespace_query,
4866 * Repair dangling name entry.
4868 * For the name entry with dangling reference, we need to repare the
4869 * inconsistency according to the LFSCK sponsor's requirement:
4871 * 1) Keep the inconsistency there and report the inconsistency case,
4872 * then give the chance to the application to find related issues,
4873 * and the users can make the decision about how to handle it with
4874 * more human knownledge. (by default)
4876 * 2) Re-create the missing MDT-object with the FID information.
4878 * \param[in] env pointer to the thread context
4879 * \param[in] com pointer to the lfsck component
4880 * \param[in] parent pointer to the dir object that contains the dangling
4882 * \param[in] child pointer to the object corresponding to the dangling
4884 * \param[in] lnr pointer to the namespace request that contains the
4885 * name's name, parent object, parent's LMV, and ect.
4887 * \retval positive number if no need to repair
4888 * \retval zero for repaired successfully
4889 * \retval negative error number on failure
4891 int lfsck_namespace_repair_dangling(const struct lu_env *env,
4892 struct lfsck_component *com,
4893 struct dt_object *parent,
4894 struct dt_object *child,
4895 struct lfsck_namespace_req *lnr)
4897 struct lfsck_thread_info *info = lfsck_env_info(env);
4898 struct lu_attr *la = &info->lti_la;
4899 struct dt_allocation_hint *hint = &info->lti_hint;
4900 struct dt_object_format *dof = &info->lti_dof;
4901 struct dt_insert_rec *rec = &info->lti_dt_rec;
4902 struct lmv_mds_md_v1 *lmv2 = &info->lti_lmv2;
4903 const struct lu_name *cname;
4904 const struct lu_fid *pfid = lfsck_dto2fid(parent);
4905 const struct lu_fid *cfid = lfsck_dto2fid(child);
4906 struct linkea_data ldata = { NULL };
4907 struct lfsck_lock_handle *llh = &info->lti_llh;
4908 struct lu_buf linkea_buf;
4909 struct lu_buf lmv_buf;
4910 struct lfsck_instance *lfsck = com->lc_lfsck;
4911 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4912 struct dt_device *dev = lfsck->li_next;
4913 struct thandle *th = NULL;
4915 __u16 type = lnr->lnr_type;
4919 cname = lfsck_name_get_const(env, lnr->lnr_name, lnr->lnr_namelen);
4920 if (bk->lb_param & LPF_CREATE_MDTOBJ)
4925 if (!create || bk->lb_param & LPF_DRYRUN)
4928 /* We may need to create the sub-objects of the @child via LOD,
4929 * so make the modification based on lfsck->li_next. */
4931 parent = lfsck_object_locate(dev, parent);
4933 GOTO(log, rc = PTR_ERR(parent));
4935 if (unlikely(!dt_try_as_dir(env, parent)))
4936 GOTO(log, rc = -ENOTDIR);
4938 child = lfsck_object_locate(dev, child);
4940 GOTO(log, rc = PTR_ERR(child));
4942 rc = linkea_links_new(&ldata, &info->lti_linkea_buf2,
4947 rc = lfsck_lock(env, lfsck, parent, lnr->lnr_name, llh,
4948 MDS_INODELOCK_UPDATE, LCK_PR);
4952 rc = lfsck_namespace_check_exist(env, parent, child, lnr->lnr_name);
4956 /* Set the ctime as zero, then others can know it is created for
4957 * repairing dangling name entry by LFSCK. And if the LFSCK made
4958 * wrong decision and the real MDT-object has been found later,
4959 * then the LFSCK has chance to fix the incosistency properly. */
4960 memset(la, 0, sizeof(*la));
4961 la->la_mode = (type & S_IFMT) | 0600;
4962 la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
4963 LA_ATIME | LA_MTIME | LA_CTIME;
4965 child->do_ops->do_ah_init(env, hint, parent, child,
4966 la->la_mode & S_IFMT);
4968 memset(dof, 0, sizeof(*dof));
4969 dof->dof_type = dt_mode_to_dft(type);
4970 /* If the target is a regular file, then the LFSCK will only create
4971 * the MDT-object without stripes (dof->dof_reg.striped = 0). related
4972 * OST-objects will be created when write open. */
4974 th = dt_trans_create(env, dev);
4976 GOTO(log, rc = PTR_ERR(th));
4978 /* 1a. create child. */
4979 rc = dt_declare_create(env, child, la, hint, dof, th);
4983 if (S_ISDIR(type)) {
4984 if (unlikely(!dt_try_as_dir(env, child)))
4985 GOTO(stop, rc = -ENOTDIR);
4987 /* 2a. increase child nlink */
4988 rc = dt_declare_ref_add(env, child, th);
4992 /* 3a. insert dot into child dir */
4993 rec->rec_type = S_IFDIR;
4994 rec->rec_fid = cfid;
4995 rc = dt_declare_insert(env, child,
4996 (const struct dt_rec *)rec,
4997 (const struct dt_key *)dot, th);
5001 /* 4a. insert dotdot into child dir */
5002 rec->rec_fid = pfid;
5003 rc = dt_declare_insert(env, child,
5004 (const struct dt_rec *)rec,
5005 (const struct dt_key *)dotdot, th);
5009 /* 5a. generate slave LMV EA. */
5010 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
5013 idx = lfsck_shard_name_to_index(env,
5014 lnr->lnr_name, lnr->lnr_namelen,
5016 if (unlikely(idx < 0))
5017 GOTO(stop, rc = idx);
5019 *lmv2 = lnr->lnr_lmv->ll_lmv;
5020 lmv2->lmv_magic = LMV_MAGIC_STRIPE;
5021 lmv2->lmv_master_mdt_index = idx;
5023 lfsck_lmv_header_cpu_to_le(lmv2, lmv2);
5024 lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
5025 rc = dt_declare_xattr_set(env, child, &lmv_buf,
5026 XATTR_NAME_LMV, 0, th);
5032 /* 6a. insert linkEA for child */
5033 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
5034 ldata.ld_leh->leh_len);
5035 rc = dt_declare_xattr_set(env, child, &linkea_buf,
5036 XATTR_NAME_LINK, 0, th);
5040 rc = dt_trans_start_local(env, dev, th);
5042 GOTO(stop, rc = (rc == -EEXIST ? 1 : rc));
5044 dt_write_lock(env, child, 0);
5045 /* 1b. create child */
5046 rc = dt_create(env, child, la, hint, dof, th);
5048 GOTO(unlock, rc = (rc == -EEXIST ? 1 : rc));
5050 if (S_ISDIR(type)) {
5051 /* 2b. increase child nlink */
5052 rc = dt_ref_add(env, child, th);
5056 /* 3b. insert dot into child dir */
5057 rec->rec_type = S_IFDIR;
5058 rec->rec_fid = cfid;
5059 rc = dt_insert(env, child, (const struct dt_rec *)rec,
5060 (const struct dt_key *)dot, th, 1);
5064 /* 4b. insert dotdot into child dir */
5065 rec->rec_fid = pfid;
5066 rc = dt_insert(env, child, (const struct dt_rec *)rec,
5067 (const struct dt_key *)dotdot, th, 1);
5071 /* 5b. generate slave LMV EA. */
5072 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
5073 rc = dt_xattr_set(env, child, &lmv_buf, XATTR_NAME_LMV,
5080 /* 6b. insert linkEA for child. */
5081 rc = dt_xattr_set(env, child, &linkea_buf,
5082 XATTR_NAME_LINK, 0, th);
5087 dt_write_unlock(env, child);
5090 dt_trans_stop(env, dev, th);
5094 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found dangling "
5095 "reference for: parent "DFID", child "DFID", type %u, "
5096 "name %s. %s: rc = %d\n", lfsck_lfsck2name(lfsck),
5097 PFID(pfid), PFID(cfid), type, cname->ln_name,
5098 create ? "Create the lost MDT-object as required" :
5099 "Keep the MDT-object there by default", rc);
5102 struct lfsck_namespace *ns = com->lc_file_ram;
5104 ns->ln_flags |= LF_INCONSISTENT;
5110 static int lfsck_namespace_assistant_handler_p1(const struct lu_env *env,
5111 struct lfsck_component *com,
5112 struct lfsck_assistant_req *lar)
5114 struct lfsck_thread_info *info = lfsck_env_info(env);
5115 struct lu_attr *la = &info->lti_la;
5116 struct lfsck_instance *lfsck = com->lc_lfsck;
5117 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5118 struct lfsck_namespace *ns = com->lc_file_ram;
5119 struct lfsck_assistant_data *lad = com->lc_data;
5120 struct linkea_data ldata = { NULL };
5121 const struct lu_name *cname;
5122 struct thandle *handle = NULL;
5123 struct lfsck_namespace_req *lnr =
5124 container_of0(lar, struct lfsck_namespace_req, lnr_lar);
5125 struct dt_object *dir = NULL;
5126 struct dt_object *obj = NULL;
5127 struct lfsck_assistant_object *lso = lar->lar_parent;
5128 const struct lu_fid *pfid = &lso->lso_fid;
5129 struct dt_device *dev = NULL;
5130 struct lustre_handle lh = { 0 };
5131 bool repaired = false;
5132 bool dtlocked = false;
5136 bool bad_hash = false;
5137 bool bad_linkea = false;
5141 enum lfsck_namespace_inconsistency_type type = LNIT_NONE;
5148 if (lnr->lnr_attr & LUDA_UPGRADE) {
5149 ns->ln_flags |= LF_UPGRADE;
5150 ns->ln_dirent_repaired++;
5152 } else if (lnr->lnr_attr & LUDA_REPAIR) {
5153 ns->ln_flags |= LF_INCONSISTENT;
5154 ns->ln_dirent_repaired++;
5158 if (unlikely(fid_is_zero(&lnr->lnr_fid))) {
5159 if (strcmp(lnr->lnr_name, dotdot) != 0)
5162 rc = lfsck_namespace_trace_update(env, com, pfid,
5163 LNTF_CHECK_PARENT, true);
5168 if (unlikely(!fid_is_sane(&lnr->lnr_fid))) {
5169 CDEBUG(D_LFSCK, "%s: dir scan find invalid FID "DFID
5170 " for the name entry %.*s under "DFID"\n",
5171 lfsck_lfsck2name(lfsck), PFID(&lnr->lnr_fid),
5172 lnr->lnr_namelen, lnr->lnr_name, PFID(pfid));
5174 if (strcmp(lnr->lnr_name, dotdot) != 0)
5175 /* invalid FID means bad name entry, remove it. */
5176 type = LNIT_BAD_DIRENT;
5178 /* If the parent FID is invalid, we cannot remove
5179 * the ".." entry directly. */
5180 rc = lfsck_namespace_trace_update(env, com, pfid,
5181 LNTF_CHECK_PARENT, true);
5186 if (unlikely(lnr->lnr_dir_cookie == MDS_DIR_END_OFF)) {
5187 rc = lfsck_namespace_striped_dir_rescan(env, com, lnr);
5192 if (fid_seq_is_dot(fid_seq(&lnr->lnr_fid)))
5195 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
5196 rc = lfsck_namespace_handle_striped_master(env, com, lnr);
5201 idx = lfsck_find_mdt_idx_by_fid(env, lfsck, &lnr->lnr_fid);
5203 GOTO(out, rc = idx);
5205 if (idx == lfsck_dev_idx(lfsck)) {
5206 if (unlikely(strcmp(lnr->lnr_name, dotdot) == 0))
5209 dev = lfsck->li_bottom;
5211 struct lfsck_tgt_desc *ltd;
5213 /* Usually, some local filesystem consistency verification
5214 * tools can guarantee the local namespace tree consistenct.
5215 * So the LFSCK will only verify the remote directory. */
5216 if (unlikely(strcmp(lnr->lnr_name, dotdot) == 0)) {
5217 rc = lfsck_namespace_trace_update(env, com, pfid,
5218 LNTF_CHECK_PARENT, true);
5223 ltd = lfsck_ltd2tgt(&lfsck->li_mdt_descs, idx);
5224 if (unlikely(ltd == NULL)) {
5225 CDEBUG(D_LFSCK, "%s: cannot talk with MDT %x which "
5226 "did not join the namespace LFSCK\n",
5227 lfsck_lfsck2name(lfsck), idx);
5228 lfsck_lad_set_bitmap(env, com, idx);
5230 GOTO(out, rc = -ENODEV);
5236 obj = lfsck_object_find_by_dev(env, dev, &lnr->lnr_fid);
5238 GOTO(out, rc = PTR_ERR(obj));
5240 cname = lfsck_name_get_const(env, lnr->lnr_name, lnr->lnr_namelen);
5241 if (dt_object_exists(obj) == 0) {
5245 dir = lfsck_assistant_object_load(env, lfsck, lso);
5249 GOTO(trace, rc == -ENOENT ? 0 : rc);
5253 rc = lfsck_namespace_check_exist(env, dir, obj, lnr->lnr_name);
5255 if (!lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5256 lnr->lnr_name, lnr->lnr_namelen)) {
5257 type = LNIT_BAD_DIRENT;
5262 type = LNIT_DANGLING;
5263 rc = lfsck_namespace_repair_dangling(env, com, dir,
5272 if (!(bk->lb_param & LPF_DRYRUN) && lad->lad_advance_lock) {
5275 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
5276 MDS_INODELOCK_UPDATE |
5277 MDS_INODELOCK_XATTR, LCK_EX);
5281 handle = dt_trans_create(env, dev);
5283 GOTO(out, rc = PTR_ERR(handle));
5285 rc = lfsck_declare_namespace_exec_dir(env, obj, handle);
5289 rc = dt_trans_start_local(env, dev, handle);
5293 dt_write_lock(env, obj, 0);
5297 rc = lfsck_links_read(env, obj, &ldata);
5298 if (unlikely(rc == -ENOENT)) {
5299 if (handle != NULL) {
5300 dt_write_unlock(env, obj);
5303 dt_trans_stop(env, dev, handle);
5306 lfsck_ibits_unlock(&lh, LCK_EX);
5309 /* It may happen when the remote object has been removed,
5310 * but the local MDT is not aware of that. */
5312 } else if (rc == 0) {
5313 count = ldata.ld_leh->leh_reccount;
5314 rc = linkea_links_find(&ldata, cname, pfid);
5316 (count == 1 || !S_ISDIR(lfsck_object_type(obj)))) {
5317 if ((lfsck_object_type(obj) & S_IFMT) !=
5319 ns->ln_flags |= LF_INCONSISTENT;
5320 type = LNIT_BAD_TYPE;
5326 /* If the name entry hash does not match the slave striped
5327 * directory, and the name entry does not match also, then
5328 * it is quite possible that name entry is corrupted. */
5329 if (!lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5330 lnr->lnr_name, lnr->lnr_namelen)) {
5331 ns->ln_flags |= LF_INCONSISTENT;
5332 type = LNIT_BAD_DIRENT;
5337 /* If the file type stored in the name entry does not match
5338 * the file type claimed by the object, and the object does
5339 * not recognize the name entry, then it is quite possible
5340 * that the name entry is corrupted. */
5341 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type) {
5342 ns->ln_flags |= LF_INCONSISTENT;
5343 type = LNIT_BAD_DIRENT;
5348 /* For sub-dir object, we cannot make sure whether the sub-dir
5349 * back references the parent via ".." name entry correctly or
5350 * not in the LFSCK first-stage scanning. It may be that the
5351 * (remote) sub-dir ".." name entry has no parent FID after
5352 * file-level backup/restore and its linkEA may be wrong.
5353 * So under such case, we should replace the linkEA according
5354 * to current name entry. But this needs to be done during the
5355 * LFSCK second-stage scanning. The LFSCK will record the name
5356 * entry for further possible using. */
5360 } else if (unlikely(rc == -EINVAL)) {
5361 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type)
5362 type = LNIT_BAD_TYPE;
5365 /* The magic crashed, we are not sure whether there are more
5366 * corrupt data in the linkea, so remove all linkea entries. */
5370 } else if (rc == -ENODATA) {
5371 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type)
5372 type = LNIT_BAD_TYPE;
5379 if (bk->lb_param & LPF_DRYRUN) {
5381 ns->ln_flags |= LF_UPGRADE;
5383 ns->ln_flags |= LF_INCONSISTENT;
5384 ns->ln_linkea_repaired++;
5390 if (!lustre_handle_is_used(&lh)) {
5398 LASSERT(handle != NULL);
5401 dir = lfsck_assistant_object_load(env, lfsck, lso);
5405 GOTO(stop, rc == -ENOENT ? 0 : rc);
5409 rc = lfsck_namespace_check_exist(env, dir, obj, lnr->lnr_name);
5414 if (!remove && newdata)
5415 ns->ln_flags |= LF_UPGRADE;
5416 else if (remove || !(ns->ln_flags & LF_UPGRADE))
5417 ns->ln_flags |= LF_INCONSISTENT;
5422 rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, handle);
5423 if (rc != 0 && rc != -ENOENT && rc != -ENODATA)
5428 rc = linkea_data_new(&ldata,
5429 &lfsck_env_info(env)->lti_linkea_buf);
5434 rc = linkea_add_buf(&ldata, cname, pfid);
5436 rc = lfsck_links_write(env, obj, &ldata, handle);
5440 count = ldata.ld_leh->leh_reccount;
5441 if (!S_ISDIR(lfsck_object_type(obj)) ||
5442 !dt_object_remote(obj)) {
5443 ns->ln_linkea_repaired++;
5453 dt_write_unlock(env, obj);
5455 if (handle != NULL && !IS_ERR(handle))
5456 dt_trans_stop(env, dev, handle);
5459 lfsck_ibits_unlock(&lh, LCK_EX);
5461 if (!name_is_dot_or_dotdot(lnr->lnr_name, lnr->lnr_namelen) &&
5462 !lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5463 lnr->lnr_name, lnr->lnr_namelen) &&
5464 type != LNIT_BAD_DIRENT) {
5465 ns->ln_flags |= LF_INCONSISTENT;
5469 dir = lfsck_assistant_object_load(env, lfsck, lso);
5473 GOTO(trace, rc == -ENOENT ? 0 : rc);
5477 rc = lfsck_namespace_repair_bad_name_hash(env, com, dir,
5478 lnr->lnr_lmv, lnr->lnr_name);
5484 if (type != LNIT_NONE && dir == NULL) {
5485 dir = lfsck_assistant_object_load(env, lfsck, lso);
5489 GOTO(trace, rc == -ENOENT ? 0 : rc);
5496 rc = lfsck_namespace_repair_dirent(env, com, dir,
5497 obj, lnr->lnr_name, lnr->lnr_name,
5498 lnr->lnr_type, true, false);
5502 case LNIT_BAD_DIRENT:
5504 /* XXX: This is a bad dirent, we do not know whether
5505 * the original name entry reference a regular
5506 * file or a directory, then keep the parent's
5507 * nlink count unchanged here. */
5508 rc = lfsck_namespace_repair_dirent(env, com, dir,
5509 obj, lnr->lnr_name, lnr->lnr_name,
5510 lnr->lnr_type, false, false);
5518 if (obj != NULL && count == 1 &&
5519 S_ISREG(lfsck_object_type(obj)))
5520 dt_attr_get(env, obj, la);
5524 down_write(&com->lc_sem);
5526 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail to handle "
5527 "the entry: "DFID", parent "DFID", name %.*s: rc = %d\n",
5528 lfsck_lfsck2name(lfsck), PFID(&lnr->lnr_fid), PFID(pfid),
5529 lnr->lnr_namelen, lnr->lnr_name, rc);
5531 lfsck_namespace_record_failure(env, lfsck, ns);
5532 if ((rc == -ENOTCONN || rc == -ESHUTDOWN || rc == -EREMCHG ||
5533 rc == -ETIMEDOUT || rc == -EHOSTDOWN ||
5534 rc == -EHOSTUNREACH || rc == -EINPROGRESS) &&
5535 dev != NULL && dev != lfsck->li_bottom)
5536 lfsck_lad_set_bitmap(env, com, idx);
5538 if (!(bk->lb_param & LPF_FAILOUT))
5542 ns->ln_items_repaired++;
5544 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant "
5545 "repaired the entry: "DFID", parent "DFID
5546 ", name %.*s, type %d\n",
5547 lfsck_lfsck2name(lfsck),
5548 PFID(&lnr->lnr_fid), PFID(pfid),
5549 lnr->lnr_namelen, lnr->lnr_name, type);
5553 ns->ln_dangling_repaired++;
5556 ns->ln_bad_type_repaired++;
5558 case LNIT_BAD_DIRENT:
5559 ns->ln_dirent_repaired++;
5565 if (bk->lb_param & LPF_DRYRUN &&
5566 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent))
5567 lfsck_pos_fill(env, lfsck,
5568 &ns->ln_pos_first_inconsistent,
5573 ns->ln_name_hash_repaired++;
5575 /* Not count repeatedly. */
5577 ns->ln_items_repaired++;
5579 CDEBUG(D_LFSCK, "%s: namespace LFSCK "
5580 "assistant repaired the entry: "
5583 lfsck_lfsck2name(lfsck),
5584 PFID(&lnr->lnr_fid), PFID(pfid),
5585 lnr->lnr_namelen, lnr->lnr_name);
5588 if (bk->lb_param & LPF_DRYRUN &&
5589 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent))
5590 lfsck_pos_fill(env, lfsck,
5591 &ns->ln_pos_first_inconsistent,
5598 if (count > 1 || la->la_nlink > 1)
5599 ns->ln_mul_linked_checked++;
5601 up_write(&com->lc_sem);
5603 if (obj != NULL && !IS_ERR(obj))
5604 lfsck_object_put(env, obj);
5606 if (dir != NULL && !IS_ERR(dir))
5607 lfsck_object_put(env, dir);
5609 lad->lad_advance_lock = bad_linkea;
5615 * Handle one orphan under the backend /lost+found directory
5617 * Insert the orphan FID into the namespace LFSCK trace file for further
5618 * processing (via the subsequent namespace LFSCK second-stage scanning).
5619 * At the same time, remove the orphan name entry from backend /lost+found
5620 * directory. There is an interval between the orphan name entry removed
5621 * from the backend /lost+found directory and the orphan FID in the LFSCK
5622 * trace file handled. In such interval, the LFSCK can be reset, then
5623 * all the FIDs recorded in the namespace LFSCK trace file will be dropped.
5624 * To guarantee that the orphans can be found when LFSCK run next time
5625 * without e2fsck again, when remove the orphan name entry, the LFSCK
5626 * will set the orphan's ctime attribute as 1. Since normal applications
5627 * cannot change the object's ctime attribute as 1. Then when LFSCK run
5628 * next time, it can record the object (that ctime is 1) in the namespace
5629 * LFSCK trace file during the first-stage scanning.
5631 * \param[in] env pointer to the thread context
5632 * \param[in] com pointer to the lfsck component
5633 * \param[in] parent pointer to the object for the backend /lost+found
5634 * \param[in] ent pointer to the name entry for the target under the
5635 * backend /lost+found
5637 * \retval positive for repaired
5638 * \retval 0 if needs to repair nothing
5639 * \retval negative error number on failure
5641 static int lfsck_namespace_scan_local_lpf_one(const struct lu_env *env,
5642 struct lfsck_component *com,
5643 struct dt_object *parent,
5644 struct lu_dirent *ent)
5646 struct lfsck_thread_info *info = lfsck_env_info(env);
5647 struct lu_fid *key = &info->lti_fid;
5648 struct lu_attr *la = &info->lti_la;
5649 struct lfsck_instance *lfsck = com->lc_lfsck;
5650 struct dt_object *obj;
5651 struct dt_device *dev = lfsck->li_bottom;
5652 struct dt_object *child = NULL;
5653 struct thandle *th = NULL;
5660 child = lfsck_object_find_by_dev(env, dev, &ent->lde_fid);
5662 RETURN(PTR_ERR(child));
5664 LASSERT(dt_object_exists(child));
5665 LASSERT(!dt_object_remote(child));
5667 idx = lfsck_sub_trace_file_fid2idx(&ent->lde_fid);
5668 obj = com->lc_sub_trace_objs[idx].lsto_obj;
5669 fid_cpu_to_be(key, &ent->lde_fid);
5670 rc = dt_lookup(env, obj, (struct dt_rec *)&flags,
5671 (const struct dt_key *)key);
5674 flags |= LNTF_CHECK_ORPHAN;
5675 } else if (rc == -ENOENT) {
5676 flags = LNTF_CHECK_ORPHAN;
5681 th = dt_trans_create(env, dev);
5683 GOTO(out, rc = PTR_ERR(th));
5685 /* a1. remove name entry from backend /lost+found */
5686 rc = dt_declare_delete(env, parent,
5687 (const struct dt_key *)ent->lde_name, th);
5691 if (S_ISDIR(lfsck_object_type(child))) {
5692 /* a2. decrease parent's nlink */
5693 rc = dt_declare_ref_del(env, parent, th);
5699 /* a3. remove child's FID from the LFSCK trace file. */
5700 rc = dt_declare_delete(env, obj,
5701 (const struct dt_key *)key, th);
5705 /* a4. set child's ctime as 1 */
5706 memset(la, 0, sizeof(*la));
5708 la->la_valid = LA_CTIME;
5709 rc = dt_declare_attr_set(env, child, la, th);
5714 /* a5. insert child's FID into the LFSCK trace file. */
5715 rc = dt_declare_insert(env, obj, (const struct dt_rec *)&flags,
5716 (const struct dt_key *)key, th);
5720 rc = dt_trans_start_local(env, dev, th);
5724 /* b1. remove name entry from backend /lost+found */
5725 rc = dt_delete(env, parent, (const struct dt_key *)ent->lde_name, th);
5729 if (S_ISDIR(lfsck_object_type(child))) {
5730 /* b2. decrease parent's nlink */
5731 dt_write_lock(env, parent, 0);
5732 rc = dt_ref_del(env, parent, th);
5733 dt_write_unlock(env, parent);
5739 /* a3. remove child's FID from the LFSCK trace file. */
5740 rc = dt_delete(env, obj, (const struct dt_key *)key, th);
5744 /* b4. set child's ctime as 1 */
5745 rc = dt_attr_set(env, child, la, th);
5750 /* b5. insert child's FID into the LFSCK trace file. */
5751 rc = dt_insert(env, obj, (const struct dt_rec *)&flags,
5752 (const struct dt_key *)key, th, 1);
5754 GOTO(stop, rc = (rc == 0 ? 1 : rc));
5757 dt_trans_stop(env, dev, th);
5760 lfsck_object_put(env, child);
5766 * Handle orphans under the backend /lost+found directory
5768 * Some backend checker, such as e2fsck for ldiskfs may find some orphans
5769 * and put them under the backend /lost+found directory that is invisible
5770 * to client. The LFSCK will scan such directory, for the original client
5771 * visible orphans, add their fids into the namespace LFSCK trace file,
5772 * then the subsenquent namespace LFSCK second-stage scanning can handle
5773 * them as other objects to be double scanned: either move back to normal
5774 * namespace, or to the global visible orphan directory:
5775 * /ROOT/.lustre/lost+found/MDTxxxx/
5777 * \param[in] env pointer to the thread context
5778 * \param[in] com pointer to the lfsck component
5780 static void lfsck_namespace_scan_local_lpf(const struct lu_env *env,
5781 struct lfsck_component *com)
5783 struct lfsck_thread_info *info = lfsck_env_info(env);
5784 struct lu_dirent *ent =
5785 (struct lu_dirent *)info->lti_key;
5786 struct lu_seq_range *range = &info->lti_range;
5787 struct lfsck_instance *lfsck = com->lc_lfsck;
5788 struct ptlrpc_thread *thread = &lfsck->li_thread;
5789 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5790 struct lfsck_namespace *ns = com->lc_file_ram;
5791 struct dt_object *parent;
5792 const struct dt_it_ops *iops;
5794 struct seq_server_site *ss = lfsck_dev_site(lfsck);
5796 __u32 idx = lfsck_dev_idx(lfsck);
5801 parent = lfsck_object_find_by_dev(env, lfsck->li_bottom,
5802 &LU_BACKEND_LPF_FID);
5803 if (IS_ERR(parent)) {
5804 CERROR("%s: fail to find backend /lost+found: rc = %ld\n",
5805 lfsck_lfsck2name(lfsck), PTR_ERR(parent));
5809 /* It is normal that the /lost+found does not exist for ZFS backend. */
5810 if (!dt_object_exists(parent))
5813 if (unlikely(!dt_try_as_dir(env, parent)))
5814 GOTO(out, rc = -ENOTDIR);
5816 CDEBUG(D_LFSCK, "%s: start to scan backend /lost+found\n",
5817 lfsck_lfsck2name(lfsck));
5819 com->lc_new_scanned = 0;
5820 iops = &parent->do_index_ops->dio_it;
5821 di = iops->init(env, parent, LUDA_64BITHASH | LUDA_TYPE);
5823 GOTO(out, rc = PTR_ERR(di));
5825 rc = iops->load(env, di, 0);
5827 rc = iops->next(env, di);
5832 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
5833 unlikely(!thread_is_running(thread)))
5836 rc = iops->rec(env, di, (struct dt_rec *)ent,
5837 LUDA_64BITHASH | LUDA_TYPE);
5839 rc = lfsck_unpack_ent(ent, &cookie, &type);
5841 if (unlikely(rc != 0)) {
5842 CDEBUG(D_LFSCK, "%s: fail to iterate backend "
5843 "/lost+found: rc = %d\n",
5844 lfsck_lfsck2name(lfsck), rc);
5849 /* skip dot and dotdot entries */
5850 if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen))
5853 if (!fid_seq_in_fldb(fid_seq(&ent->lde_fid)))
5856 if (fid_is_norm(&ent->lde_fid)) {
5857 fld_range_set_mdt(range);
5858 rc = fld_local_lookup(env, ss->ss_server_fld,
5859 fid_seq(&ent->lde_fid), range);
5862 } else if (idx != 0) {
5863 /* If the returned FID is IGIF, then there are three
5866 * 1) The object is upgraded from old Lustre-1.8 with
5867 * IGIF assigned to such object.
5868 * 2) The object is a backend local object and is
5869 * invisible to client.
5870 * 3) The object lost its LMV EA, and since there is
5871 * no FID-in-dirent for the orphan in the backend
5872 * /lost+found directory, then the low layer will
5873 * return IGIF for such object.
5875 * For MDTx (x != 0), it is either case 2) or case 3),
5876 * but from the LFSCK view, they are indistinguishable.
5877 * To be safe, the LFSCK will keep it there and report
5878 * some message, then the adminstrator can handle that
5881 * For MDT0, it is more possible the case 1). The LFSCK
5882 * will handle the orphan as an upgraded object. */
5883 CDEBUG(D_LFSCK, "%s: the orphan %.*s with IGIF "DFID
5884 "in the backend /lost+found on the MDT %04x, "
5885 "to be safe, skip it.\n",
5886 lfsck_lfsck2name(lfsck), ent->lde_namelen,
5887 ent->lde_name, PFID(&ent->lde_fid), idx);
5891 rc = lfsck_namespace_scan_local_lpf_one(env, com, parent, ent);
5894 down_write(&com->lc_sem);
5895 com->lc_new_scanned++;
5896 ns->ln_local_lpf_scanned++;
5898 ns->ln_local_lpf_moved++;
5900 ns->ln_local_lpf_skipped++;
5902 ns->ln_local_lpf_failed++;
5903 up_write(&com->lc_sem);
5905 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
5909 lfsck_control_speed_by_self(com);
5910 if (unlikely(!thread_is_running(thread))) {
5915 rc = iops->next(env, di);
5919 iops->fini(env, di);
5924 CDEBUG(D_LFSCK, "%s: stop to scan backend /lost+found: rc = %d\n",
5925 lfsck_lfsck2name(lfsck), rc);
5927 lfsck_object_put(env, parent);
5931 * Rescan the striped directory after the master LMV EA reset.
5933 * Sometimes, the master LMV EA of the striped directory maybe lost, so when
5934 * the namespace LFSCK engine scan the striped directory for the first time,
5935 * it will be regarded as a normal directory. As the LFSCK processing, some
5936 * other LFSCK instance on other MDT will find the shard of this striped dir,
5937 * and find that the master MDT-object of the striped directory lost its LMV
5938 * EA, then such remote LFSCK instance will regenerate the master LMV EA and
5939 * notify the LFSCK instance on this MDT to rescan the striped directory.
5941 * \param[in] env pointer to the thread context
5942 * \param[in] com pointer to the lfsck component
5943 * \param[in] llu the lfsck_lmv_unit that contains the striped directory
5946 * \retval positive number for success
5947 * \retval 0 for LFSCK stopped/paused
5948 * \retval negative error number on failure
5950 static int lfsck_namespace_rescan_striped_dir(const struct lu_env *env,
5951 struct lfsck_component *com,
5952 struct lfsck_lmv_unit *llu)
5954 struct lfsck_thread_info *info = lfsck_env_info(env);
5955 struct lfsck_instance *lfsck = com->lc_lfsck;
5956 struct lfsck_assistant_data *lad = com->lc_data;
5957 struct dt_object *dir;
5958 const struct dt_it_ops *iops;
5960 struct lu_dirent *ent =
5961 (struct lu_dirent *)info->lti_key;
5962 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5963 struct ptlrpc_thread *thread = &lfsck->li_thread;
5964 struct lfsck_assistant_object *lso = NULL;
5965 struct lfsck_namespace_req *lnr;
5966 struct lfsck_assistant_req *lar;
5971 LASSERT(list_empty(&lad->lad_req_list));
5973 lfsck->li_lmv = &llu->llu_lmv;
5974 lfsck->li_obj_dir = lfsck_object_get(llu->llu_obj);
5975 rc = lfsck_open_dir(env, lfsck, 0);
5979 dir = lfsck->li_obj_dir;
5980 di = lfsck->li_di_dir;
5981 iops = &dir->do_index_ops->dio_it;
5983 rc = iops->rec(env, di, (struct dt_rec *)ent,
5984 lfsck->li_args_dir);
5986 rc = lfsck_unpack_ent(ent, &lfsck->li_cookie_dir,
5990 if (bk->lb_param & LPF_FAILOUT)
5996 if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen))
6000 lso = lfsck_assistant_object_init(env,
6001 lfsck_dto2fid(dir), NULL,
6002 lfsck->li_pos_current.lp_oit_cookie, true);
6004 if (bk->lb_param & LPF_FAILOUT)
6005 GOTO(out, rc = PTR_ERR(lso));
6012 lnr = lfsck_namespace_assistant_req_init(lfsck, lso, ent, type);
6014 if (bk->lb_param & LPF_FAILOUT)
6015 GOTO(out, rc = PTR_ERR(lnr));
6020 lar = &lnr->lnr_lar;
6021 rc = lfsck_namespace_assistant_handler_p1(env, com, lar);
6022 lfsck_namespace_assistant_req_fini(env, lar);
6023 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
6026 if (unlikely(!thread_is_running(thread)))
6030 rc = iops->next(env, di);
6034 if (lso != NULL && !IS_ERR(lso))
6035 lfsck_assistant_object_put(env, lso);
6037 lfsck_close_dir(env, lfsck, rc);
6041 /* The close_dir() may insert a dummy lnr in the lad->lad_req_list. */
6042 if (list_empty(&lad->lad_req_list))
6045 spin_lock(&lad->lad_lock);
6046 lar = list_entry(lad->lad_req_list.next, struct lfsck_assistant_req,
6048 list_del_init(&lar->lar_list);
6049 spin_unlock(&lad->lad_lock);
6051 rc = lfsck_namespace_assistant_handler_p1(env, com, lar);
6052 lfsck_namespace_assistant_req_fini(env, lar);
6054 RETURN(rc == 0 ? 1 : rc);
6058 lfsck_namespace_double_scan_one_trace_file(const struct lu_env *env,
6059 struct lfsck_component *com,
6060 struct dt_object *obj, bool first)
6062 struct lfsck_instance *lfsck = com->lc_lfsck;
6063 struct ptlrpc_thread *thread = &lfsck->li_thread;
6064 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
6065 struct lfsck_namespace *ns = com->lc_file_ram;
6066 const struct dt_it_ops *iops = &obj->do_index_ops->dio_it;
6067 struct dt_object *target;
6075 di = iops->init(env, obj, 0);
6077 RETURN(PTR_ERR(di));
6080 fid_cpu_to_be(&fid, &ns->ln_fid_latest_scanned_phase2);
6083 rc = iops->get(env, di, (const struct dt_key *)&fid);
6088 /* The start one either has been processed or does not exist,
6090 rc = iops->next(env, di);
6096 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
6097 unlikely(!thread_is_running(thread)))
6100 key = iops->key(env, di);
6109 fid_be_to_cpu(&fid, (const struct lu_fid *)key);
6110 if (!fid_is_sane(&fid)) {
6115 target = lfsck_object_find_bottom(env, lfsck, &fid);
6116 if (IS_ERR(target)) {
6117 rc = PTR_ERR(target);
6121 if (dt_object_exists(target)) {
6122 rc = iops->rec(env, di, (struct dt_rec *)&flags, 0);
6124 rc = lfsck_namespace_double_scan_one(env, com,
6131 lfsck_object_put(env, target);
6134 down_write(&com->lc_sem);
6135 com->lc_new_checked++;
6136 com->lc_new_scanned++;
6138 ns->ln_fid_latest_scanned_phase2 = fid;
6141 ns->ln_objs_repaired_phase2++;
6143 ns->ln_objs_failed_phase2++;
6144 up_write(&com->lc_sem);
6146 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
6149 if (unlikely(com->lc_time_next_checkpoint <=
6150 ktime_get_seconds()) &&
6151 com->lc_new_checked != 0) {
6152 down_write(&com->lc_sem);
6153 ns->ln_run_time_phase2 += ktime_get_seconds() -
6154 com->lc_time_last_checkpoint;
6155 ns->ln_time_last_checkpoint = ktime_get_real_seconds();
6156 ns->ln_objs_checked_phase2 += com->lc_new_checked;
6157 com->lc_new_checked = 0;
6158 lfsck_namespace_store(env, com);
6159 up_write(&com->lc_sem);
6161 com->lc_time_last_checkpoint = ktime_get_seconds();
6162 com->lc_time_next_checkpoint =
6163 com->lc_time_last_checkpoint +
6164 LFSCK_CHECKPOINT_INTERVAL;
6167 lfsck_control_speed_by_self(com);
6168 if (unlikely(!thread_is_running(thread)))
6171 rc = iops->next(env, di);
6180 iops->fini(env, di);
6185 static int lfsck_namespace_assistant_handler_p2(const struct lu_env *env,
6186 struct lfsck_component *com)
6188 struct lfsck_instance *lfsck = com->lc_lfsck;
6189 struct lfsck_namespace *ns = com->lc_file_ram;
6194 while (!list_empty(&lfsck->li_list_lmv)) {
6195 struct lfsck_lmv_unit *llu;
6197 spin_lock(&lfsck->li_lock);
6198 llu = list_entry(lfsck->li_list_lmv.next,
6199 struct lfsck_lmv_unit, llu_link);
6200 list_del_init(&llu->llu_link);
6201 spin_unlock(&lfsck->li_lock);
6203 rc = lfsck_namespace_rescan_striped_dir(env, com, llu);
6208 CDEBUG(D_LFSCK, "%s: namespace LFSCK phase2 scan start\n",
6209 lfsck_lfsck2name(lfsck));
6211 lfsck_namespace_scan_local_lpf(env, com);
6213 com->lc_new_checked = 0;
6214 com->lc_new_scanned = 0;
6215 com->lc_time_last_checkpoint = ktime_get_seconds();
6216 com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
6217 LFSCK_CHECKPOINT_INTERVAL;
6219 i = lfsck_sub_trace_file_fid2idx(&ns->ln_fid_latest_scanned_phase2);
6220 rc = lfsck_namespace_double_scan_one_trace_file(env, com,
6221 com->lc_sub_trace_objs[i].lsto_obj, true);
6222 while (rc > 0 && ++i < LFSCK_STF_COUNT)
6223 rc = lfsck_namespace_double_scan_one_trace_file(env, com,
6224 com->lc_sub_trace_objs[i].lsto_obj, false);
6226 CDEBUG(D_LFSCK, "%s: namespace LFSCK phase2 scan stop at the No. %d "
6227 "trace file: rc = %d\n", lfsck_lfsck2name(lfsck), i, rc);
6232 static void lfsck_namespace_assistant_fill_pos(const struct lu_env *env,
6233 struct lfsck_component *com,
6234 struct lfsck_position *pos)
6236 struct lfsck_assistant_data *lad = com->lc_data;
6237 struct lfsck_namespace_req *lnr;
6239 if (((struct lfsck_namespace *)(com->lc_file_ram))->ln_status !=
6243 if (list_empty(&lad->lad_req_list))
6246 lnr = list_entry(lad->lad_req_list.next,
6247 struct lfsck_namespace_req,
6249 pos->lp_oit_cookie = lnr->lnr_lar.lar_parent->lso_oit_cookie;
6250 pos->lp_dir_cookie = lnr->lnr_dir_cookie - 1;
6251 pos->lp_dir_parent = lnr->lnr_lar.lar_parent->lso_fid;
6254 static int lfsck_namespace_double_scan_result(const struct lu_env *env,
6255 struct lfsck_component *com,
6258 struct lfsck_instance *lfsck = com->lc_lfsck;
6259 struct lfsck_namespace *ns = com->lc_file_ram;
6261 down_write(&com->lc_sem);
6262 ns->ln_run_time_phase2 += ktime_get_seconds() -
6263 com->lc_time_last_checkpoint;
6264 ns->ln_time_last_checkpoint = ktime_get_real_seconds();
6265 ns->ln_objs_checked_phase2 += com->lc_new_checked;
6266 com->lc_new_checked = 0;
6269 if (ns->ln_flags & LF_INCOMPLETE)
6270 ns->ln_status = LS_PARTIAL;
6272 ns->ln_status = LS_COMPLETED;
6273 ns->ln_flags &= ~LF_SCANNED_ONCE;
6274 if (!(lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN))
6275 ns->ln_flags &= ~LF_INCONSISTENT;
6276 ns->ln_time_last_complete = ns->ln_time_last_checkpoint;
6277 ns->ln_success_count++;
6278 } else if (rc == 0) {
6279 if (lfsck->li_status != 0)
6280 ns->ln_status = lfsck->li_status;
6282 ns->ln_status = LS_STOPPED;
6284 ns->ln_status = LS_FAILED;
6287 rc = lfsck_namespace_store(env, com);
6288 up_write(&com->lc_sem);
6294 lfsck_namespace_assistant_sync_failures_interpret(const struct lu_env *env,
6295 struct ptlrpc_request *req,
6299 struct lfsck_async_interpret_args *laia = args;
6300 struct lfsck_tgt_desc *ltd = laia->laia_ltd;
6302 ltd->ltd_synced_failures = 1;
6309 * Notify remote LFSCK instances about former failures.
6311 * The local LFSCK instance has recorded which MDTs have ever failed to respond
6312 * some LFSCK verification requests (maybe because of network issues or the MDT
6313 * itself trouble). During the respond gap the MDT may missed some name entries
6314 * verification, then the MDT cannot know whether related MDT-objects have been
6315 * referenced by related name entries or not, then in the second-stage scanning,
6316 * these MDT-objects will be regarded as orphan, if the MDT-object contains bad
6317 * linkEA for back reference, then it will misguide the LFSCK to generate wrong
6318 * name entry for repairing the orphan.
6320 * To avoid above trouble, when layout LFSCK finishes the first-stage scanning,
6321 * it will scan the bitmap for the ever failed MDTs, and notify them that they
6322 * have ever missed some name entries verification and should skip the handling
6323 * for orphan MDT-objects.
6325 * \param[in] env pointer to the thread context
6326 * \param[in] com pointer to the lfsck component
6327 * \param[in] lr pointer to the lfsck request
6329 static void lfsck_namespace_assistant_sync_failures(const struct lu_env *env,
6330 struct lfsck_component *com,
6331 struct lfsck_request *lr)
6333 struct lfsck_async_interpret_args *laia =
6334 &lfsck_env_info(env)->lti_laia2;
6335 struct lfsck_assistant_data *lad = com->lc_data;
6336 struct lfsck_namespace *ns = com->lc_file_ram;
6337 struct lfsck_instance *lfsck = com->lc_lfsck;
6338 struct lfsck_tgt_descs *ltds = &lfsck->li_mdt_descs;
6339 struct lfsck_tgt_desc *ltd;
6340 struct ptlrpc_request_set *set;
6345 if (!lad->lad_incomplete)
6348 set = ptlrpc_prep_set();
6350 GOTO(out, rc = -ENOMEM);
6352 lr->lr_flags2 = ns->ln_flags | LF_INCOMPLETE;
6353 memset(laia, 0, sizeof(*laia));
6354 lad->lad_touch_gen++;
6356 down_read(<ds->ltd_rw_sem);
6357 cfs_foreach_bit(lad->lad_bitmap, idx) {
6358 ltd = lfsck_ltd2tgt(ltds, idx);
6362 laia->laia_ltd = ltd;
6363 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
6364 lfsck_namespace_assistant_sync_failures_interpret,
6365 laia, LFSCK_NOTIFY);
6367 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail "
6368 "to sync failure with MDT %x: rc = %d\n",
6369 lfsck_lfsck2name(lfsck), ltd->ltd_index, rc);
6371 up_read(<ds->ltd_rw_sem);
6373 rc = ptlrpc_set_wait(set);
6374 ptlrpc_set_destroy(set);
6380 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail "
6381 "to sync failure with MDTs, and related MDTs "
6382 "may handle orphan improperly: rc = %d\n",
6383 lfsck_lfsck2name(lfsck), rc);
6388 struct lfsck_assistant_operations lfsck_namespace_assistant_ops = {
6389 .la_handler_p1 = lfsck_namespace_assistant_handler_p1,
6390 .la_handler_p2 = lfsck_namespace_assistant_handler_p2,
6391 .la_fill_pos = lfsck_namespace_assistant_fill_pos,
6392 .la_double_scan_result = lfsck_namespace_double_scan_result,
6393 .la_req_fini = lfsck_namespace_assistant_req_fini,
6394 .la_sync_failures = lfsck_namespace_assistant_sync_failures,
6398 * Verify the specified linkEA entry for the given directory object.
6399 * If the object has no such linkEA entry or it has more other linkEA
6400 * entries, then re-generate the linkEA with the given information.
6402 * \param[in] env pointer to the thread context
6403 * \param[in] obj pointer to the dt_object to be handled
6404 * \param[in] cname the name for the child in the parent directory
6405 * \param[in] pfid the parent directory's FID for the linkEA
6407 * \retval 0 for success
6408 * \retval negative error number on failure
6410 int lfsck_verify_linkea(const struct lu_env *env, struct dt_object *obj,
6411 const struct lu_name *cname, const struct lu_fid *pfid)
6413 struct dt_device *dev = lfsck_obj2dev(obj);
6414 struct linkea_data ldata = { NULL };
6415 struct lu_buf linkea_buf;
6418 int fl = LU_XATTR_CREATE;
6422 LASSERT(S_ISDIR(lfsck_object_type(obj)));
6424 rc = lfsck_links_read_with_rec(env, obj, &ldata);
6425 if (rc == -ENODATA) {
6427 } else if (rc == 0) {
6428 fl = LU_XATTR_REPLACE;
6429 if (ldata.ld_leh->leh_reccount != 1) {
6432 rc = linkea_links_find(&ldata, cname, pfid);
6441 rc = linkea_links_new(&ldata, &lfsck_env_info(env)->lti_linkea_buf,
6446 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
6447 ldata.ld_leh->leh_len);
6448 th = dt_trans_create(env, dev);
6450 RETURN(PTR_ERR(th));
6452 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
6453 XATTR_NAME_LINK, fl, th);
6457 rc = dt_trans_start_local(env, dev, th);
6461 dt_write_lock(env, obj, 0);
6462 rc = dt_xattr_set(env, obj, &linkea_buf,
6463 XATTR_NAME_LINK, fl, th);
6464 dt_write_unlock(env, obj);
6469 dt_trans_stop(env, dev, th);
6474 * Get the name and parent directory's FID from the first linkEA entry.
6476 * \param[in] env pointer to the thread context
6477 * \param[in] obj pointer to the object which get linkEA from
6478 * \param[out] name pointer to the buffer to hold the name
6479 * in the first linkEA entry
6480 * \param[out] pfid pointer to the buffer to hold the parent
6481 * directory's FID in the first linkEA entry
6483 * \retval 0 for success
6484 * \retval negative error number on failure
6486 int lfsck_links_get_first(const struct lu_env *env, struct dt_object *obj,
6487 char *name, struct lu_fid *pfid)
6489 struct lu_name *cname = &lfsck_env_info(env)->lti_name;
6490 struct linkea_data ldata = { NULL };
6493 rc = lfsck_links_read_with_rec(env, obj, &ldata);
6497 linkea_first_entry(&ldata);
6498 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, cname, pfid);
6499 /* To guarantee the 'name' is terminated with '0'. */
6500 memcpy(name, cname->ln_name, cname->ln_namelen);
6501 name[cname->ln_namelen] = 0;
6507 * Update the object's name entry with the given FID.
6509 * \param[in] env pointer to the thread context
6510 * \param[in] lfsck pointer to the lfsck instance
6511 * \param[in] dir pointer to the directory that holds
6513 * \param[in] name the name for the entry to be updated
6514 * \param[in] fid the new FID for the name entry referenced
6515 * \param[in] type the type for the name entry to be updated
6517 * \retval 0 for success
6518 * \retval negative error number on failure
6520 int lfsck_update_name_entry(const struct lu_env *env,
6521 struct lfsck_instance *lfsck,
6522 struct dt_object *dir, const char *name,
6523 const struct lu_fid *fid, __u32 type)
6525 struct lfsck_thread_info *info = lfsck_env_info(env);
6526 struct dt_insert_rec *rec = &info->lti_dt_rec;
6527 struct lfsck_lock_handle *llh = &info->lti_llh;
6528 struct dt_device *dev = lfsck_obj2dev(dir);
6534 rc = lfsck_lock(env, lfsck, dir, name, llh,
6535 MDS_INODELOCK_UPDATE, LCK_PW);
6539 th = dt_trans_create(env, dev);
6541 GOTO(unlock, rc = PTR_ERR(th));
6543 rc = dt_declare_delete(env, dir, (const struct dt_key *)name, th);
6547 rec->rec_type = type;
6549 rc = dt_declare_insert(env, dir, (const struct dt_rec *)rec,
6550 (const struct dt_key *)name, th);
6554 rc = dt_declare_ref_add(env, dir, th);
6558 rc = dt_trans_start_local(env, dev, th);
6562 rc = dt_delete(env, dir, (const struct dt_key *)name, th);
6563 if (rc == -ENOENT) {
6571 rc = dt_insert(env, dir, (const struct dt_rec *)rec,
6572 (const struct dt_key *)name, th, 1);
6573 if (rc == 0 && S_ISDIR(type) && !exists) {
6574 dt_write_lock(env, dir, 0);
6575 rc = dt_ref_add(env, dir, th);
6576 dt_write_unlock(env, dir);
6582 dt_trans_stop(env, dev, th);
6586 CDEBUG(D_LFSCK, "%s: update name entry "DFID"/%s with the FID "DFID
6587 " and the type %o: rc = %d\n", lfsck_lfsck2name(lfsck),
6588 PFID(lfsck_dto2fid(dir)), name, PFID(fid), type, rc);
6593 int lfsck_namespace_setup(const struct lu_env *env,
6594 struct lfsck_instance *lfsck)
6596 struct lfsck_component *com;
6597 struct lfsck_namespace *ns;
6598 struct dt_object *root = NULL;
6599 struct dt_object *obj;
6604 LASSERT(lfsck->li_master);
6610 INIT_LIST_HEAD(&com->lc_link);
6611 INIT_LIST_HEAD(&com->lc_link_dir);
6612 init_rwsem(&com->lc_sem);
6613 atomic_set(&com->lc_ref, 1);
6614 com->lc_lfsck = lfsck;
6615 com->lc_type = LFSCK_TYPE_NAMESPACE;
6616 com->lc_ops = &lfsck_namespace_ops;
6617 com->lc_data = lfsck_assistant_data_init(
6618 &lfsck_namespace_assistant_ops,
6620 if (com->lc_data == NULL)
6621 GOTO(out, rc = -ENOMEM);
6623 com->lc_file_size = sizeof(struct lfsck_namespace);
6624 OBD_ALLOC(com->lc_file_ram, com->lc_file_size);
6625 if (com->lc_file_ram == NULL)
6626 GOTO(out, rc = -ENOMEM);
6628 OBD_ALLOC(com->lc_file_disk, com->lc_file_size);
6629 if (com->lc_file_disk == NULL)
6630 GOTO(out, rc = -ENOMEM);
6632 for (i = 0; i < LFSCK_STF_COUNT; i++)
6633 mutex_init(&com->lc_sub_trace_objs[i].lsto_mutex);
6635 root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
6637 GOTO(out, rc = PTR_ERR(root));
6639 if (unlikely(!dt_try_as_dir(env, root)))
6640 GOTO(out, rc = -ENOTDIR);
6642 obj = local_index_find_or_create(env, lfsck->li_los, root,
6644 S_IFREG | S_IRUGO | S_IWUSR,
6645 &dt_lfsck_namespace_features);
6647 GOTO(out, rc = PTR_ERR(obj));
6650 rc = lfsck_namespace_load(env, com);
6652 rc = lfsck_namespace_init(env, com);
6654 rc = lfsck_namespace_reset(env, com, true);
6656 rc = lfsck_load_sub_trace_files(env, com,
6657 &dt_lfsck_namespace_features, LFSCK_NAMESPACE, false);
6661 ns = com->lc_file_ram;
6662 switch (ns->ln_status) {
6667 spin_lock(&lfsck->li_lock);
6668 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
6669 spin_unlock(&lfsck->li_lock);
6672 CERROR("%s: unknown lfsck_namespace status %d\n",
6673 lfsck_lfsck2name(lfsck), ns->ln_status);
6675 case LS_SCANNING_PHASE1:
6676 case LS_SCANNING_PHASE2:
6677 /* No need to store the status to disk right now.
6678 * If the system crashed before the status stored,
6679 * it will be loaded back when next time. */
6680 ns->ln_status = LS_CRASHED;
6684 spin_lock(&lfsck->li_lock);
6685 list_add_tail(&com->lc_link, &lfsck->li_list_scan);
6686 list_add_tail(&com->lc_link_dir, &lfsck->li_list_dir);
6687 spin_unlock(&lfsck->li_lock);
6694 if (root != NULL && !IS_ERR(root))
6695 lfsck_object_put(env, root);
6697 lfsck_component_cleanup(env, com);
6698 CERROR("%s: fail to init namespace LFSCK component: rc = %d\n",
6699 lfsck_lfsck2name(lfsck), rc);