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, 2017, Intel Corporation.
26 * lustre/lfsck/lfsck_namespace.c
28 * Author: Fan, Yong <fan.yong@intel.com>
31 #define DEBUG_SUBSYSTEM S_LFSCK
33 #include <lu_object.h>
34 #include <dt_object.h>
35 #include <md_object.h>
36 #include <lustre_fid.h>
37 #include <lustre_lib.h>
38 #include <lustre_net.h>
40 #include "lfsck_internal.h"
42 #define LFSCK_NAMESPACE_MAGIC_V1 0xA0629D03
43 #define LFSCK_NAMESPACE_MAGIC_V2 0xA0621A0B
44 #define LFSCK_NAMESPACE_MAGIC_V3 0xA06249FF
46 /* For Lustre-2.x (x <= 6), the namespace LFSCK used LFSCK_NAMESPACE_MAGIC_V1
47 * as the trace file magic. When downgrade to such old release, the old LFSCK
48 * will not recognize the new LFSCK_NAMESPACE_MAGIC_V2 in the new trace file,
49 * then it will reset the whole LFSCK, and will not cause start failure. The
50 * similar case will happen when upgrade from such old release. */
51 #define LFSCK_NAMESPACE_MAGIC LFSCK_NAMESPACE_MAGIC_V3
53 enum lfsck_nameentry_check {
54 LFSCK_NAMEENTRY_DEAD = 1, /* The object has been unlinked. */
55 LFSCK_NAMEENTRY_REMOVED = 2, /* The entry has been removed. */
56 LFSCK_NAMEENTRY_RECREATED = 3, /* The entry has been recreated. */
59 static struct lfsck_namespace_req *
60 lfsck_namespace_assistant_req_init(struct lfsck_instance *lfsck,
61 struct lfsck_assistant_object *lso,
62 struct lu_dirent *ent, __u16 type)
64 struct lfsck_namespace_req *lnr;
67 size = sizeof(*lnr) + (ent->lde_namelen & ~3) + 4;
70 return ERR_PTR(-ENOMEM);
72 INIT_LIST_HEAD(&lnr->lnr_lar.lar_list);
73 lnr->lnr_lar.lar_parent = lfsck_assistant_object_get(lso);
74 lnr->lnr_lmv = lfsck_lmv_get(lfsck->li_lmv);
75 lnr->lnr_fid = ent->lde_fid;
76 lnr->lnr_dir_cookie = ent->lde_hash;
77 lnr->lnr_attr = ent->lde_attrs;
80 lnr->lnr_namelen = ent->lde_namelen;
81 memcpy(lnr->lnr_name, ent->lde_name, ent->lde_namelen);
86 static void lfsck_namespace_assistant_req_fini(const struct lu_env *env,
87 struct lfsck_assistant_req *lar)
89 struct lfsck_namespace_req *lnr =
90 container_of0(lar, struct lfsck_namespace_req, lnr_lar);
92 if (lnr->lnr_lmv != NULL)
93 lfsck_lmv_put(env, lnr->lnr_lmv);
95 lfsck_assistant_object_put(env, lar->lar_parent);
96 OBD_FREE(lnr, lnr->lnr_size);
99 static void lfsck_namespace_le_to_cpu(struct lfsck_namespace *dst,
100 struct lfsck_namespace *src)
102 dst->ln_magic = le32_to_cpu(src->ln_magic);
103 dst->ln_status = le32_to_cpu(src->ln_status);
104 dst->ln_flags = le32_to_cpu(src->ln_flags);
105 dst->ln_success_count = le32_to_cpu(src->ln_success_count);
106 dst->ln_run_time_phase1 = le64_to_cpu(src->ln_run_time_phase1);
107 dst->ln_run_time_phase2 = le64_to_cpu(src->ln_run_time_phase2);
108 dst->ln_time_last_complete = le64_to_cpu(src->ln_time_last_complete);
109 dst->ln_time_latest_start = le64_to_cpu(src->ln_time_latest_start);
110 dst->ln_time_last_checkpoint =
111 le64_to_cpu(src->ln_time_last_checkpoint);
112 lfsck_position_le_to_cpu(&dst->ln_pos_latest_start,
113 &src->ln_pos_latest_start);
114 lfsck_position_le_to_cpu(&dst->ln_pos_last_checkpoint,
115 &src->ln_pos_last_checkpoint);
116 lfsck_position_le_to_cpu(&dst->ln_pos_first_inconsistent,
117 &src->ln_pos_first_inconsistent);
118 dst->ln_items_checked = le64_to_cpu(src->ln_items_checked);
119 dst->ln_items_repaired = le64_to_cpu(src->ln_items_repaired);
120 dst->ln_items_failed = le64_to_cpu(src->ln_items_failed);
121 dst->ln_dirs_checked = le64_to_cpu(src->ln_dirs_checked);
122 dst->ln_objs_checked_phase2 = le64_to_cpu(src->ln_objs_checked_phase2);
123 dst->ln_objs_repaired_phase2 =
124 le64_to_cpu(src->ln_objs_repaired_phase2);
125 dst->ln_objs_failed_phase2 = le64_to_cpu(src->ln_objs_failed_phase2);
126 dst->ln_objs_nlink_repaired = le64_to_cpu(src->ln_objs_nlink_repaired);
127 fid_le_to_cpu(&dst->ln_fid_latest_scanned_phase2,
128 &src->ln_fid_latest_scanned_phase2);
129 dst->ln_dirent_repaired = le64_to_cpu(src->ln_dirent_repaired);
130 dst->ln_linkea_repaired = le64_to_cpu(src->ln_linkea_repaired);
131 dst->ln_mul_linked_checked = le64_to_cpu(src->ln_mul_linked_checked);
132 dst->ln_mul_linked_repaired = le64_to_cpu(src->ln_mul_linked_repaired);
133 dst->ln_unknown_inconsistency =
134 le64_to_cpu(src->ln_unknown_inconsistency);
135 dst->ln_unmatched_pairs_repaired =
136 le64_to_cpu(src->ln_unmatched_pairs_repaired);
137 dst->ln_dangling_repaired = le64_to_cpu(src->ln_dangling_repaired);
138 dst->ln_mul_ref_repaired = le64_to_cpu(src->ln_mul_ref_repaired);
139 dst->ln_bad_type_repaired = le64_to_cpu(src->ln_bad_type_repaired);
140 dst->ln_lost_dirent_repaired =
141 le64_to_cpu(src->ln_lost_dirent_repaired);
142 dst->ln_striped_dirs_scanned =
143 le64_to_cpu(src->ln_striped_dirs_scanned);
144 dst->ln_striped_dirs_repaired =
145 le64_to_cpu(src->ln_striped_dirs_repaired);
146 dst->ln_striped_dirs_failed =
147 le64_to_cpu(src->ln_striped_dirs_failed);
148 dst->ln_striped_dirs_disabled =
149 le64_to_cpu(src->ln_striped_dirs_disabled);
150 dst->ln_striped_dirs_skipped =
151 le64_to_cpu(src->ln_striped_dirs_skipped);
152 dst->ln_striped_shards_scanned =
153 le64_to_cpu(src->ln_striped_shards_scanned);
154 dst->ln_striped_shards_repaired =
155 le64_to_cpu(src->ln_striped_shards_repaired);
156 dst->ln_striped_shards_failed =
157 le64_to_cpu(src->ln_striped_shards_failed);
158 dst->ln_striped_shards_skipped =
159 le64_to_cpu(src->ln_striped_shards_skipped);
160 dst->ln_name_hash_repaired = le64_to_cpu(src->ln_name_hash_repaired);
161 dst->ln_local_lpf_scanned = le64_to_cpu(src->ln_local_lpf_scanned);
162 dst->ln_local_lpf_moved = le64_to_cpu(src->ln_local_lpf_moved);
163 dst->ln_local_lpf_skipped = le64_to_cpu(src->ln_local_lpf_skipped);
164 dst->ln_local_lpf_failed = le64_to_cpu(src->ln_local_lpf_failed);
165 dst->ln_bitmap_size = le32_to_cpu(src->ln_bitmap_size);
166 dst->ln_time_latest_reset = le64_to_cpu(src->ln_time_latest_reset);
167 dst->ln_linkea_overflow_cleared =
168 le64_to_cpu(src->ln_linkea_overflow_cleared);
169 dst->ln_agent_entries_repaired =
170 le64_to_cpu(src->ln_agent_entries_repaired);
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);
243 dst->ln_agent_entries_repaired =
244 cpu_to_le64(src->ln_agent_entries_repaired);
247 static void lfsck_namespace_record_failure(const struct lu_env *env,
248 struct lfsck_instance *lfsck,
249 struct lfsck_namespace *ns)
251 struct lfsck_position pos;
253 ns->ln_items_failed++;
254 lfsck_pos_fill(env, lfsck, &pos, false);
255 if (lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent) ||
256 lfsck_pos_is_eq(&pos, &ns->ln_pos_first_inconsistent) < 0) {
257 ns->ln_pos_first_inconsistent = pos;
259 CDEBUG(D_LFSCK, "%s: namespace LFSCK hit first non-repaired "
260 "inconsistency at the pos [%llu, "DFID", %#llx]\n",
261 lfsck_lfsck2name(lfsck),
262 ns->ln_pos_first_inconsistent.lp_oit_cookie,
263 PFID(&ns->ln_pos_first_inconsistent.lp_dir_parent),
264 ns->ln_pos_first_inconsistent.lp_dir_cookie);
269 * Load the MDT bitmap from the lfsck_namespace trace file.
271 * \param[in] env pointer to the thread context
272 * \param[in] com pointer to the lfsck component
274 * \retval 0 for success
275 * \retval negative error number on failure or data corruption
277 static int lfsck_namespace_load_bitmap(const struct lu_env *env,
278 struct lfsck_component *com)
280 struct dt_object *obj = com->lc_obj;
281 struct lfsck_assistant_data *lad = com->lc_data;
282 struct lfsck_namespace *ns = com->lc_file_ram;
283 struct cfs_bitmap *bitmap = lad->lad_bitmap;
289 if (com->lc_lfsck->li_mdt_descs.ltd_tgts_bitmap->size >
291 nbits = com->lc_lfsck->li_mdt_descs.ltd_tgts_bitmap->size;
293 nbits = ns->ln_bitmap_size;
295 if (unlikely(nbits < BITS_PER_LONG))
296 nbits = BITS_PER_LONG;
298 if (nbits > bitmap->size) {
299 __u32 new_bits = bitmap->size;
300 struct cfs_bitmap *new_bitmap;
302 while (new_bits < nbits)
305 new_bitmap = CFS_ALLOCATE_BITMAP(new_bits);
306 if (new_bitmap == NULL)
309 lad->lad_bitmap = new_bitmap;
310 CFS_FREE_BITMAP(bitmap);
314 if (ns->ln_bitmap_size == 0) {
315 clear_bit(LAD_INCOMPLETE, &lad->lad_flags);
316 CFS_RESET_BITMAP(bitmap);
321 size = (ns->ln_bitmap_size + 7) >> 3;
322 rc = dt_xattr_get(env, obj,
323 lfsck_buf_get(env, bitmap->data, size),
324 XATTR_NAME_LFSCK_BITMAP);
326 RETURN(rc >= 0 ? -EINVAL : rc);
328 if (cfs_bitmap_check_empty(bitmap))
329 clear_bit(LAD_INCOMPLETE, &lad->lad_flags);
331 set_bit(LAD_INCOMPLETE, &lad->lad_flags);
337 * Load namespace LFSCK statistics information from the trace file.
339 * \param[in] env pointer to the thread context
340 * \param[in] com pointer to the lfsck component
342 * \retval 0 for success
343 * \retval negative error number on failure
345 static int lfsck_namespace_load(const struct lu_env *env,
346 struct lfsck_component *com)
348 int len = com->lc_file_size;
351 rc = dt_xattr_get(env, com->lc_obj,
352 lfsck_buf_get(env, com->lc_file_disk, len),
353 XATTR_NAME_LFSCK_NAMESPACE);
355 struct lfsck_namespace *ns = com->lc_file_ram;
357 lfsck_namespace_le_to_cpu(ns,
358 (struct lfsck_namespace *)com->lc_file_disk);
359 if (ns->ln_magic != LFSCK_NAMESPACE_MAGIC) {
360 CDEBUG(D_LFSCK, "%s: invalid lfsck_namespace magic "
361 "%#x != %#x\n", lfsck_lfsck2name(com->lc_lfsck),
362 ns->ln_magic, LFSCK_NAMESPACE_MAGIC);
367 } else if (rc != -ENODATA) {
368 CDEBUG(D_LFSCK, "%s: fail to load lfsck_namespace, "
369 "expected = %d: rc = %d\n",
370 lfsck_lfsck2name(com->lc_lfsck), len, rc);
378 static int lfsck_namespace_store(const struct lu_env *env,
379 struct lfsck_component *com)
381 struct dt_object *obj = com->lc_obj;
382 struct lfsck_instance *lfsck = com->lc_lfsck;
383 struct lfsck_namespace *ns = com->lc_file_ram;
384 struct lfsck_assistant_data *lad = com->lc_data;
385 struct dt_device *dev = lfsck_obj2dev(obj);
386 struct cfs_bitmap *bitmap = NULL;
387 struct thandle *handle;
389 int len = com->lc_file_size;
394 bitmap = lad->lad_bitmap;
395 nbits = bitmap->size;
398 LASSERTF((nbits & 7) == 0, "Invalid nbits %u\n", nbits);
401 ns->ln_bitmap_size = nbits;
402 lfsck_namespace_cpu_to_le((struct lfsck_namespace *)com->lc_file_disk,
404 handle = dt_trans_create(env, dev);
406 GOTO(log, rc = PTR_ERR(handle));
408 rc = dt_declare_xattr_set(env, obj,
409 lfsck_buf_get(env, com->lc_file_disk, len),
410 XATTR_NAME_LFSCK_NAMESPACE, 0, handle);
414 if (bitmap != NULL) {
415 rc = dt_declare_xattr_set(env, obj,
416 lfsck_buf_get(env, bitmap->data, nbits >> 3),
417 XATTR_NAME_LFSCK_BITMAP, 0, handle);
422 rc = dt_trans_start_local(env, dev, handle);
426 rc = dt_xattr_set(env, obj,
427 lfsck_buf_get(env, com->lc_file_disk, len),
428 XATTR_NAME_LFSCK_NAMESPACE, 0, handle);
429 if (rc == 0 && bitmap != NULL)
430 rc = dt_xattr_set(env, obj,
431 lfsck_buf_get(env, bitmap->data, nbits >> 3),
432 XATTR_NAME_LFSCK_BITMAP, 0, handle);
437 dt_trans_stop(env, dev, handle);
441 CDEBUG(D_LFSCK, "%s: fail to store lfsck_namespace: rc = %d\n",
442 lfsck_lfsck2name(lfsck), rc);
446 static int lfsck_namespace_init(const struct lu_env *env,
447 struct lfsck_component *com)
449 struct lfsck_namespace *ns = com->lc_file_ram;
452 memset(ns, 0, sizeof(*ns));
453 ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
454 ns->ln_status = LS_INIT;
455 ns->ln_time_latest_reset = ktime_get_real_seconds();
456 down_write(&com->lc_sem);
457 rc = lfsck_namespace_store(env, com);
459 rc = lfsck_load_sub_trace_files(env, com,
460 &dt_lfsck_namespace_features, LFSCK_NAMESPACE, true);
461 up_write(&com->lc_sem);
467 * Update the namespace LFSCK trace file for the given @fid
469 * \param[in] env pointer to the thread context
470 * \param[in] com pointer to the lfsck component
471 * \param[in] fid the fid which flags to be updated in the lfsck
473 * \param[in] add true if add new flags, otherwise remove flags
475 * \retval 0 for success or nothing to be done
476 * \retval negative error number on failure
478 int lfsck_namespace_trace_update(const struct lu_env *env,
479 struct lfsck_component *com,
480 const struct lu_fid *fid,
481 const __u8 flags, bool add)
483 struct lfsck_instance *lfsck = com->lc_lfsck;
484 struct dt_object *obj;
485 struct lu_fid *key = &lfsck_env_info(env)->lti_fid3;
486 struct dt_device *dev;
487 struct thandle *th = NULL;
496 if (unlikely(!fid_is_sane(fid)))
499 idx = lfsck_sub_trace_file_fid2idx(fid);
500 mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
501 obj = com->lc_sub_trace_objs[idx].lsto_obj;
502 if (unlikely(obj == NULL)) {
503 mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
507 lfsck_object_get(obj);
508 dev = lfsck_obj2dev(obj);
509 fid_cpu_to_be(key, fid);
510 rc = dt_lookup(env, obj, (struct dt_rec *)&old,
511 (const struct dt_key *)key);
514 GOTO(unlock, rc = 0);
518 } else if (rc == 0) {
520 if ((old & flags) == flags)
521 GOTO(unlock, rc = 0);
525 if ((old & flags) == 0)
526 GOTO(unlock, rc = 0);
534 th = dt_trans_create(env, dev);
536 GOTO(log, rc = PTR_ERR(th));
539 rc = dt_declare_delete(env, obj,
540 (const struct dt_key *)key, th);
546 rc = dt_declare_insert(env, obj,
547 (const struct dt_rec *)&new,
548 (const struct dt_key *)key, th);
553 rc = dt_trans_start_local(env, dev, th);
558 rc = dt_delete(env, obj, (const struct dt_key *)key, th);
564 rc = dt_insert(env, obj, (const struct dt_rec *)&new,
565 (const struct dt_key *)key, th);
573 if (th != NULL && !IS_ERR(th))
574 dt_trans_stop(env, dev, th);
576 CDEBUG(D_LFSCK, "%s: namespace LFSCK %s flags for "DFID" in the "
577 "trace file, flags %x, old %x, new %x: rc = %d\n",
578 lfsck_lfsck2name(lfsck), add ? "add" : "del", PFID(fid),
579 (__u32)flags, (__u32)old, (__u32)new, rc);
582 mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
583 lfsck_object_put(env, obj);
588 int lfsck_namespace_check_exist(const struct lu_env *env,
589 struct dt_object *dir,
590 struct dt_object *obj, const char *name)
592 struct lu_fid *fid = &lfsck_env_info(env)->lti_fid;
596 if (unlikely(lfsck_is_dead_obj(obj)))
597 RETURN(LFSCK_NAMEENTRY_DEAD);
599 rc = dt_lookup(env, dir, (struct dt_rec *)fid,
600 (const struct dt_key *)name);
602 RETURN(LFSCK_NAMEENTRY_REMOVED);
607 if (!lu_fid_eq(fid, lfsck_dto2fid(obj)))
608 RETURN(LFSCK_NAMEENTRY_RECREATED);
613 static int lfsck_declare_namespace_exec_dir(const struct lu_env *env,
614 struct dt_object *obj,
615 struct thandle *handle)
619 /* For remote updating LINKEA, there may be further LFSCK action
620 * on remote MDT after the updating, so update the LINKEA ASAP. */
621 if (dt_object_remote(obj))
624 /* For destroying all invalid linkEA entries. */
625 rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, handle);
627 /* For insert new linkEA entry. */
628 rc = dt_declare_xattr_set(env, obj,
629 lfsck_buf_get_const(env, NULL, MAX_LINKEA_SIZE),
630 XATTR_NAME_LINK, 0, handle);
634 int __lfsck_links_read(const struct lu_env *env, struct dt_object *obj,
635 struct linkea_data *ldata, bool with_rec)
639 if (ldata->ld_buf->lb_buf == NULL)
642 if (!dt_object_exists(obj))
645 rc = dt_xattr_get(env, obj, ldata->ld_buf, XATTR_NAME_LINK);
647 /* Buf was too small, figure out what we need. */
648 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LINK);
649 if (unlikely(rc == 0))
655 lu_buf_realloc(ldata->ld_buf, rc);
656 if (ldata->ld_buf->lb_buf == NULL)
659 rc = dt_xattr_get(env, obj, ldata->ld_buf, XATTR_NAME_LINK);
662 if (unlikely(rc == 0))
667 rc = linkea_init_with_rec(ldata);
669 rc = linkea_init(ldata);
676 * Remove linkEA for the given object.
678 * The caller should take the ldlm lock before the calling.
680 * \param[in] env pointer to the thread context
681 * \param[in] com pointer to the lfsck component
682 * \param[in] obj pointer to the dt_object to be handled
684 * \retval 0 for repaired cases
685 * \retval negative error number on failure
687 static int lfsck_namespace_links_remove(const struct lu_env *env,
688 struct lfsck_component *com,
689 struct dt_object *obj)
691 struct lfsck_instance *lfsck = com->lc_lfsck;
692 struct dt_device *dev = lfsck_obj2dev(obj);
693 struct thandle *th = NULL;
697 LASSERT(dt_object_remote(obj) == 0);
699 th = dt_trans_create(env, dev);
701 GOTO(log, rc = PTR_ERR(th));
703 rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, th);
707 rc = dt_trans_start_local(env, dev, th);
711 dt_write_lock(env, obj, 0);
712 if (unlikely(lfsck_is_dead_obj(obj)))
713 GOTO(unlock, rc = -ENOENT);
715 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
716 GOTO(unlock, rc = 0);
718 rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, th);
723 dt_write_unlock(env, obj);
726 dt_trans_stop(env, dev, th);
729 CDEBUG(D_LFSCK, "%s: namespace LFSCK remove invalid linkEA "
730 "for the object "DFID": rc = %d\n",
731 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
734 struct lfsck_namespace *ns = com->lc_file_ram;
736 ns->ln_flags |= LF_INCONSISTENT;
742 static int lfsck_links_write(const struct lu_env *env, struct dt_object *obj,
743 struct linkea_data *ldata, struct thandle *handle)
748 lfsck_buf_init(&buf, ldata->ld_buf->lb_buf, ldata->ld_leh->leh_len);
751 rc = dt_xattr_set(env, obj, &buf, XATTR_NAME_LINK, 0, handle);
752 if (unlikely(rc == -ENOSPC)) {
753 rc = linkea_overflow_shrink(ldata);
754 if (likely(rc > 0)) {
763 static inline bool linkea_reclen_is_valid(const struct linkea_data *ldata)
765 if (ldata->ld_reclen <= 0)
768 if ((char *)ldata->ld_lee + ldata->ld_reclen >
769 (char *)ldata->ld_leh + ldata->ld_leh->leh_len)
775 static inline bool linkea_entry_is_valid(const struct linkea_data *ldata,
776 const struct lu_name *cname,
777 const struct lu_fid *pfid)
779 if (!linkea_reclen_is_valid(ldata))
782 if (cname->ln_namelen <= 0 || cname->ln_namelen > NAME_MAX)
785 if (!fid_is_sane(pfid))
791 static int lfsck_namespace_unpack_linkea_entry(struct linkea_data *ldata,
792 struct lu_name *cname,
794 char *buf, const int buflen)
796 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, cname, pfid);
797 if (unlikely(!linkea_entry_is_valid(ldata, cname, pfid)))
800 /* To guarantee the 'name' is terminated with '0'. */
801 memcpy(buf, cname->ln_name, cname->ln_namelen);
802 buf[cname->ln_namelen] = 0;
803 cname->ln_name = buf;
808 static void lfsck_linkea_del_buf(struct linkea_data *ldata,
809 const struct lu_name *lname)
811 LASSERT(ldata->ld_leh != NULL && ldata->ld_lee != NULL);
813 /* If current record is corrupted, all the subsequent
814 * records will be dropped. */
815 if (unlikely(!linkea_reclen_is_valid(ldata))) {
816 void *ptr = ldata->ld_lee;
818 ldata->ld_leh->leh_len = sizeof(struct link_ea_header);
819 ldata->ld_leh->leh_reccount = 0;
820 linkea_first_entry(ldata);
821 while (ldata->ld_lee != NULL &&
822 (char *)ldata->ld_lee < (char *)ptr) {
823 int reclen = (ldata->ld_lee->lee_reclen[0] << 8) |
824 ldata->ld_lee->lee_reclen[1];
826 ldata->ld_leh->leh_len += reclen;
827 ldata->ld_leh->leh_reccount++;
828 ldata->ld_lee = (struct link_ea_entry *)
829 ((char *)ldata->ld_lee + reclen);
832 ldata->ld_lee = NULL;
834 linkea_del_buf(ldata, lname);
838 static int lfsck_namespace_filter_linkea_entry(struct linkea_data *ldata,
839 struct lu_name *cname,
843 struct link_ea_entry *oldlee;
847 oldlee = ldata->ld_lee;
848 oldlen = ldata->ld_reclen;
849 linkea_next_entry(ldata);
850 while (ldata->ld_lee != NULL) {
851 ldata->ld_reclen = (ldata->ld_lee->lee_reclen[0] << 8) |
852 ldata->ld_lee->lee_reclen[1];
853 if (unlikely(!linkea_reclen_is_valid(ldata))) {
854 lfsck_linkea_del_buf(ldata, NULL);
855 LASSERT(ldata->ld_lee == NULL);
856 } else if (unlikely(ldata->ld_reclen == oldlen &&
857 memcmp(ldata->ld_lee, oldlee, oldlen) == 0)) {
862 lfsck_linkea_del_buf(ldata, cname);
864 linkea_next_entry(ldata);
867 ldata->ld_lee = oldlee;
868 ldata->ld_reclen = oldlen;
874 * Insert orphan into .lustre/lost+found/MDTxxxx/ locally.
876 * Add the specified orphan MDT-object to the .lustre/lost+found/MDTxxxx/
877 * with the given type to generate the name, the detailed rules for name
878 * have been described as following.
880 * The function also generates the linkEA corresponding to the name entry
881 * under the .lustre/lost+found/MDTxxxx/ for the orphan MDT-object.
883 * \param[in] env pointer to the thread context
884 * \param[in] com pointer to the lfsck component
885 * \param[in] orphan pointer to the orphan MDT-object
886 * \param[in] infix additional information for the orphan name, such as
887 * the FID for original
888 * \param[in] type the type for describing why the orphan MDT-object is
889 * created. The rules are as following:
891 * type "D": The MDT-object is a directory, it may knows its parent
892 * but because there is no valid linkEA, the LFSCK cannot
893 * know where to put it back to the namespace.
894 * type "O": The MDT-object has no linkEA, and there is no name
895 * entry that references the MDT-object.
897 * type "S": The orphan MDT-object is a shard of a striped directory
899 * \see lfsck_layout_recreate_parent() for more types.
901 * The orphan name will be like:
902 * ${FID}-${infix}-${type}-${conflict_version}
904 * \param[out] count if some others inserted some linkEA entries by race,
905 * then return the linkEA entries count.
907 * \retval positive number for repaired cases
908 * \retval 0 if needs to repair nothing
909 * \retval negative error number on failure
911 static int lfsck_namespace_insert_orphan(const struct lu_env *env,
912 struct lfsck_component *com,
913 struct dt_object *orphan,
914 const char *infix, const char *type,
917 struct lfsck_thread_info *info = lfsck_env_info(env);
918 struct lu_name *cname = &info->lti_name;
919 struct dt_insert_rec *rec = &info->lti_dt_rec;
920 struct lu_attr *la = &info->lti_la2;
921 const struct lu_fid *cfid = lfsck_dto2fid(orphan);
922 const struct lu_fid *pfid;
924 struct lfsck_instance *lfsck = com->lc_lfsck;
925 struct dt_device *dev = lfsck_obj2dev(orphan);
926 struct dt_object *parent;
927 struct thandle *th = NULL;
928 struct lfsck_lock_handle *pllh = &info->lti_llh;
929 struct lustre_handle clh = { 0 };
930 struct linkea_data ldata2 = { NULL };
931 struct lu_buf linkea_buf;
938 cname->ln_name = NULL;
939 if (unlikely(lfsck->li_lpf_obj == NULL))
940 GOTO(log, rc = -ENXIO);
942 parent = lfsck->li_lpf_obj;
943 pfid = lfsck_dto2fid(parent);
947 namelen = snprintf(info->lti_key, NAME_MAX, DFID"%s-%s-%d",
948 PFID(cfid), infix, type, idx++);
949 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
950 (const struct dt_key *)info->lti_key);
951 if (rc != 0 && rc != -ENOENT)
954 if (unlikely(rc == 0 && lu_fid_eq(cfid, &tfid)))
956 } while (rc == 0 && !exist);
958 rc = lfsck_lock(env, lfsck, parent, info->lti_key, pllh,
959 MDS_INODELOCK_UPDATE, LCK_PW);
963 /* Re-check whether the name conflict with othrs after taken
965 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
966 (const struct dt_key *)info->lti_key);
968 if (!lu_fid_eq(cfid, &tfid)) {
975 } else if (rc != -ENOENT) {
981 cname->ln_name = info->lti_key;
982 cname->ln_namelen = namelen;
983 rc = linkea_links_new(&ldata2, &info->lti_linkea_buf2,
988 rc = lfsck_ibits_lock(env, lfsck, orphan, &clh,
989 MDS_INODELOCK_UPDATE | MDS_INODELOCK_LOOKUP |
990 MDS_INODELOCK_XATTR, LCK_EX);
994 lfsck_buf_init(&linkea_buf, ldata2.ld_buf->lb_buf,
995 ldata2.ld_leh->leh_len);
996 th = dt_trans_create(env, dev);
998 GOTO(log, rc = PTR_ERR(th));
1000 if (S_ISDIR(lfsck_object_type(orphan))) {
1001 rc = dt_declare_delete(env, orphan,
1002 (const struct dt_key *)dotdot, th);
1006 rec->rec_type = S_IFDIR;
1007 rec->rec_fid = pfid;
1008 rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
1009 (const struct dt_key *)dotdot, th);
1014 rc = dt_declare_xattr_set(env, orphan, &linkea_buf,
1015 XATTR_NAME_LINK, 0, th);
1020 rec->rec_type = lfsck_object_type(orphan) & S_IFMT;
1021 rec->rec_fid = cfid;
1022 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1023 (const struct dt_key *)cname->ln_name,
1028 if (S_ISDIR(rec->rec_type)) {
1029 rc = dt_declare_ref_add(env, parent, th);
1035 memset(la, 0, sizeof(*la));
1036 la->la_ctime = ktime_get_real_seconds();
1037 la->la_valid = LA_CTIME;
1038 rc = dt_declare_attr_set(env, orphan, la, th);
1042 rc = dt_trans_start_local(env, dev, th);
1046 dt_write_lock(env, orphan, 0);
1047 rc = lfsck_links_read2_with_rec(env, orphan, &ldata2);
1048 if (likely(rc == -ENODATA || rc == -EINVAL)) {
1049 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1050 GOTO(unlock, rc = 1);
1052 if (S_ISDIR(lfsck_object_type(orphan))) {
1053 rc = dt_delete(env, orphan,
1054 (const struct dt_key *)dotdot, th);
1058 rec->rec_type = S_IFDIR;
1059 rec->rec_fid = pfid;
1060 rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1061 (const struct dt_key *)dotdot, th);
1066 rc = dt_xattr_set(env, orphan, &linkea_buf, XATTR_NAME_LINK, 0,
1069 if (rc == 0 && count != NULL)
1070 *count = ldata2.ld_leh->leh_reccount;
1074 dt_write_unlock(env, orphan);
1076 if (rc == 0 && !exist) {
1077 rec->rec_type = lfsck_object_type(orphan) & S_IFMT;
1078 rec->rec_fid = cfid;
1079 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1080 (const struct dt_key *)cname->ln_name, th);
1081 if (rc == 0 && S_ISDIR(rec->rec_type)) {
1082 dt_write_lock(env, parent, 0);
1083 rc = dt_ref_add(env, parent, th);
1084 dt_write_unlock(env, parent);
1089 rc = dt_attr_set(env, orphan, la, th);
1091 GOTO(stop, rc = (rc == 0 ? 1 : rc));
1094 dt_write_unlock(env, orphan);
1097 dt_trans_stop(env, dev, th);
1100 lfsck_ibits_unlock(&clh, LCK_EX);
1102 CDEBUG(D_LFSCK, "%s: namespace LFSCK insert orphan for the "
1103 "object "DFID", name = %s: rc = %d\n",
1104 lfsck_lfsck2name(lfsck), PFID(cfid),
1105 cname->ln_name != NULL ? cname->ln_name : "<NULL>", rc);
1108 struct lfsck_namespace *ns = com->lc_file_ram;
1110 ns->ln_flags |= LF_INCONSISTENT;
1117 * Add the specified name entry back to namespace.
1119 * If there is a linkEA entry that back references a name entry under
1120 * some parent directory, but such parent directory does not have the
1121 * claimed name entry. On the other hand, the linkEA entries count is
1122 * not larger than the MDT-object's hard link count. Under such case,
1123 * it is quite possible that the name entry is lost. Then the LFSCK
1124 * should add the name entry back to the namespace.
1126 * \param[in] env pointer to the thread context
1127 * \param[in] com pointer to the lfsck component
1128 * \param[in] parent pointer to the directory under which the name entry
1129 * will be inserted into
1130 * \param[in] child pointer to the object referenced by the name entry
1131 * that to be inserted into the parent
1132 * \param[in] name the name for the child in the parent directory
1134 * \retval positive number for repaired cases
1135 * \retval 0 if nothing to be repaired
1136 * \retval negative error number on failure
1138 static int lfsck_namespace_insert_normal(const struct lu_env *env,
1139 struct lfsck_component *com,
1140 struct dt_object *parent,
1141 struct dt_object *child,
1144 struct lfsck_thread_info *info = lfsck_env_info(env);
1145 struct lu_attr *la = &info->lti_la;
1146 struct dt_insert_rec *rec = &info->lti_dt_rec;
1147 struct lfsck_instance *lfsck = com->lc_lfsck;
1148 /* The child and its name may be on different MDTs. */
1149 const struct lu_fid *pfid = lfsck_dto2fid(parent);
1150 const struct lu_fid *cfid = lfsck_dto2fid(child);
1151 struct dt_device *dev = lfsck->li_next;
1152 struct thandle *th = NULL;
1153 struct lfsck_lock_handle *llh = &info->lti_llh;
1157 /* @parent/@child may be based on lfsck->li_bottom,
1158 * but here we need the object based on the lfsck->li_next. */
1160 parent = lfsck_object_locate(dev, parent);
1162 GOTO(log, rc = PTR_ERR(parent));
1164 if (unlikely(!dt_try_as_dir(env, parent)))
1165 GOTO(log, rc = -ENOTDIR);
1167 child = lfsck_object_locate(dev, child);
1169 GOTO(log, rc = PTR_ERR(child));
1171 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1174 rc = lfsck_lock(env, lfsck, parent, name, llh,
1175 MDS_INODELOCK_UPDATE, LCK_PW);
1179 th = dt_trans_create(env, dev);
1181 GOTO(unlock, rc = PTR_ERR(th));
1183 rec->rec_type = lfsck_object_type(child) & S_IFMT;
1184 rec->rec_fid = cfid;
1185 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1186 (const struct dt_key *)name, th);
1190 if (S_ISDIR(rec->rec_type)) {
1191 rc = dt_declare_ref_add(env, parent, th);
1196 memset(la, 0, sizeof(*la));
1197 la->la_ctime = ktime_get_real_seconds();
1198 la->la_valid = LA_CTIME;
1199 rc = dt_declare_attr_set(env, parent, la, th);
1203 rc = dt_declare_attr_set(env, child, la, th);
1207 rc = dt_trans_start_local(env, dev, th);
1211 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1212 (const struct dt_key *)name, th);
1216 if (S_ISDIR(rec->rec_type)) {
1217 dt_write_lock(env, parent, 0);
1218 rc = dt_ref_add(env, parent, th);
1219 dt_write_unlock(env, parent);
1224 la->la_ctime = ktime_get_real_seconds();
1225 rc = dt_attr_set(env, parent, la, th);
1229 rc = dt_attr_set(env, child, la, th);
1231 GOTO(stop, rc = (rc == 0 ? 1 : rc));
1234 dt_trans_stop(env, dev, th);
1240 CDEBUG(D_LFSCK, "%s: namespace LFSCK insert object "DFID" with "
1241 "the name %s and type %o to the parent "DFID": rc = %d\n",
1242 lfsck_lfsck2name(lfsck), PFID(cfid), name,
1243 lfsck_object_type(child) & S_IFMT, PFID(pfid), rc);
1246 struct lfsck_namespace *ns = com->lc_file_ram;
1248 ns->ln_flags |= LF_INCONSISTENT;
1250 ns->ln_lost_dirent_repaired++;
1257 * Create the specified orphan directory.
1259 * For the case that the parent MDT-object stored in some MDT-object's
1260 * linkEA entry is lost, the LFSCK will re-create the parent object as
1261 * an orphan and insert it into .lustre/lost+found/MDTxxxx/ directory
1262 * with the name ${FID}-P-${conflict_version}.
1264 * \param[in] env pointer to the thread context
1265 * \param[in] com pointer to the lfsck component
1266 * \param[in] orphan pointer to the orphan MDT-object to be created
1267 * \param[in] lmv pointer to master LMV EA that will be set to the orphan
1269 * \retval positive number for repaired cases
1270 * \retval negative error number on failure
1272 static int lfsck_namespace_create_orphan_dir(const struct lu_env *env,
1273 struct lfsck_component *com,
1274 struct dt_object *orphan,
1275 struct lmv_mds_md_v1 *lmv)
1277 struct lfsck_thread_info *info = lfsck_env_info(env);
1278 struct lu_attr *la = &info->lti_la;
1279 struct dt_allocation_hint *hint = &info->lti_hint;
1280 struct dt_object_format *dof = &info->lti_dof;
1281 struct lu_name *cname = &info->lti_name2;
1282 struct dt_insert_rec *rec = &info->lti_dt_rec;
1283 struct lmv_mds_md_v1 *lmv2 = &info->lti_lmv2;
1284 const struct lu_fid *cfid = lfsck_dto2fid(orphan);
1286 struct lfsck_instance *lfsck = com->lc_lfsck;
1287 struct lfsck_namespace *ns = com->lc_file_ram;
1288 struct dt_device *dev = lfsck_obj2dev(orphan);
1289 struct dt_object *parent = NULL;
1290 struct thandle *th = NULL;
1291 struct lfsck_lock_handle *llh = &info->lti_llh;
1292 struct linkea_data ldata = { NULL };
1293 struct lu_buf linkea_buf;
1294 struct lu_buf lmv_buf;
1302 LASSERT(!dt_object_exists(orphan));
1304 cname->ln_name = NULL;
1305 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1308 if (dt_object_remote(orphan)) {
1309 LASSERT(lfsck->li_lpf_root_obj != NULL);
1311 idx = lfsck_find_mdt_idx_by_fid(env, lfsck, cfid);
1313 GOTO(log, rc = idx);
1315 snprintf(name, 8, "MDT%04x", idx);
1316 rc = dt_lookup(env, lfsck->li_lpf_root_obj,
1317 (struct dt_rec *)&tfid,
1318 (const struct dt_key *)name);
1320 GOTO(log, rc = (rc == -ENOENT ? -ENXIO : rc));
1322 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
1324 GOTO(log, rc = PTR_ERR(parent));
1326 if (unlikely(!dt_try_as_dir(env, parent)))
1327 GOTO(log, rc = -ENOTDIR);
1329 if (unlikely(lfsck->li_lpf_obj == NULL))
1330 GOTO(log, rc = -ENXIO);
1332 parent = lfsck->li_lpf_obj;
1335 dev = lfsck_find_dev_by_fid(env, lfsck, cfid);
1337 GOTO(log, rc = PTR_ERR(dev));
1343 namelen = snprintf(name, 31, DFID"-P-%d",
1345 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
1346 (const struct dt_key *)name);
1347 if (rc != 0 && rc != -ENOENT)
1351 rc = lfsck_lock(env, lfsck, parent, name, llh,
1352 MDS_INODELOCK_UPDATE, LCK_PW);
1356 /* Re-check whether the name conflict with othrs after taken
1358 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
1359 (const struct dt_key *)name);
1360 if (unlikely(rc == 0)) {
1368 cname->ln_name = name;
1369 cname->ln_namelen = namelen;
1371 memset(la, 0, sizeof(*la));
1372 la->la_mode = S_IFDIR | 0700;
1373 la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
1374 LA_ATIME | LA_MTIME | LA_CTIME;
1376 orphan->do_ops->do_ah_init(env, hint, parent, orphan,
1377 la->la_mode & S_IFMT);
1379 memset(dof, 0, sizeof(*dof));
1380 dof->dof_type = dt_mode_to_dft(S_IFDIR);
1382 rc = linkea_links_new(&ldata, &info->lti_linkea_buf2,
1383 cname, lfsck_dto2fid(parent));
1387 th = dt_trans_create(env, dev);
1389 GOTO(unlock1, rc = PTR_ERR(th));
1391 /* Sync the remote transaction to guarantee that the subsequent
1392 * lock against the @orphan can find the @orphan in time. */
1393 if (dt_object_remote(orphan))
1396 rc = dt_declare_create(env, orphan, la, hint, dof, th);
1400 if (unlikely(!dt_try_as_dir(env, orphan)))
1401 GOTO(stop, rc = -ENOTDIR);
1403 rc = dt_declare_ref_add(env, orphan, th);
1407 rec->rec_type = S_IFDIR;
1408 rec->rec_fid = cfid;
1409 rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
1410 (const struct dt_key *)dot, th);
1414 rec->rec_fid = lfsck_dto2fid(parent);
1415 rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
1416 (const struct dt_key *)dotdot, th);
1421 lmv->lmv_magic = LMV_MAGIC;
1422 lmv->lmv_master_mdt_index = lfsck_dev_idx(lfsck);
1423 lfsck_lmv_header_cpu_to_le(lmv2, lmv);
1424 lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
1425 rc = dt_declare_xattr_set(env, orphan, &lmv_buf,
1426 XATTR_NAME_LMV, 0, th);
1431 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
1432 ldata.ld_leh->leh_len);
1433 rc = dt_declare_xattr_set(env, orphan, &linkea_buf,
1434 XATTR_NAME_LINK, 0, th);
1438 rec->rec_fid = cfid;
1439 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1440 (const struct dt_key *)name, th);
1442 rc = dt_declare_ref_add(env, parent, th);
1447 rc = dt_trans_start_local(env, dev, th);
1451 dt_write_lock(env, orphan, 0);
1452 rc = dt_create(env, orphan, la, hint, dof, th);
1456 rc = dt_ref_add(env, orphan, th);
1460 rec->rec_fid = cfid;
1461 rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1462 (const struct dt_key *)dot, th);
1466 rec->rec_fid = lfsck_dto2fid(parent);
1467 rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1468 (const struct dt_key *)dotdot, th);
1473 rc = dt_xattr_set(env, orphan, &lmv_buf, XATTR_NAME_LMV, 0, th);
1478 rc = dt_xattr_set(env, orphan, &linkea_buf,
1479 XATTR_NAME_LINK, 0, th);
1480 dt_write_unlock(env, orphan);
1484 rec->rec_fid = cfid;
1485 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1486 (const struct dt_key *)name, th);
1488 dt_write_lock(env, parent, 0);
1489 rc = dt_ref_add(env, parent, th);
1490 dt_write_unlock(env, parent);
1493 GOTO(stop, rc = (rc == 0 ? 1 : rc));
1496 dt_write_unlock(env, orphan);
1499 rc1 = dt_trans_stop(env, dev, th);
1500 if (rc1 != 0 && rc > 0)
1507 CDEBUG(D_LFSCK, "%s: namespace LFSCK create orphan dir for "
1508 "the object "DFID", name = %s: rc = %d\n",
1509 lfsck_lfsck2name(lfsck), PFID(cfid),
1510 cname->ln_name != NULL ? cname->ln_name : "<NULL>", rc);
1512 if (parent != NULL && !IS_ERR(parent) && parent != lfsck->li_lpf_obj)
1513 lfsck_object_put(env, parent);
1516 ns->ln_flags |= LF_INCONSISTENT;
1522 * Remove the specified entry from the linkEA.
1524 * Locate the linkEA entry with the given @cname and @pfid, then
1525 * remove this entry or the other entries those are repeated with
1528 * \param[in] env pointer to the thread context
1529 * \param[in] com pointer to the lfsck component
1530 * \param[in] obj pointer to the dt_object to be handled
1531 * \param[in,out]ldata pointer to the buffer that holds the linkEA
1532 * \param[in] cname the name for the child in the parent directory
1533 * \param[in] pfid the parent directory's FID for the linkEA
1534 * \param[in] next if true, then remove the first found linkEA
1535 * entry, and move the ldata->ld_lee to next entry
1537 * \retval positive number for repaired cases
1538 * \retval 0 if nothing to be repaired
1539 * \retval negative error number on failure
1541 static int lfsck_namespace_shrink_linkea(const struct lu_env *env,
1542 struct lfsck_component *com,
1543 struct dt_object *obj,
1544 struct linkea_data *ldata,
1545 struct lu_name *cname,
1546 struct lu_fid *pfid,
1549 struct lfsck_instance *lfsck = com->lc_lfsck;
1550 struct dt_device *dev = lfsck_obj2dev(obj);
1551 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
1552 struct thandle *th = NULL;
1553 struct lustre_handle lh = { 0 };
1554 struct linkea_data ldata_new = { NULL };
1555 struct lu_buf linkea_buf;
1560 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
1561 MDS_INODELOCK_UPDATE | MDS_INODELOCK_XATTR,
1567 lfsck_linkea_del_buf(ldata, cname);
1569 lfsck_namespace_filter_linkea_entry(ldata, cname, pfid,
1571 if (ldata->ld_leh->leh_reccount > 0 ||
1572 unlikely(ldata->ld_leh->leh_overflow_time)) {
1573 lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
1574 ldata->ld_leh->leh_len);
1575 buflen = linkea_buf.lb_len;
1579 th = dt_trans_create(env, dev);
1581 GOTO(unlock1, rc = PTR_ERR(th));
1584 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
1585 XATTR_NAME_LINK, 0, th);
1587 rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, th);
1591 rc = dt_trans_start_local(env, dev, th);
1595 dt_write_lock(env, obj, 0);
1596 if (unlikely(lfsck_is_dead_obj(obj)))
1597 GOTO(unlock2, rc = -ENOENT);
1599 rc = lfsck_links_read2_with_rec(env, obj, &ldata_new);
1601 GOTO(unlock2, rc = (rc == -ENODATA ? 0 : rc));
1603 /* The specified linkEA entry has been removed by race. */
1604 rc = linkea_links_find(&ldata_new, cname, pfid);
1606 GOTO(unlock2, rc = 0);
1608 if (bk->lb_param & LPF_DRYRUN)
1609 GOTO(unlock2, rc = 1);
1612 lfsck_linkea_del_buf(&ldata_new, cname);
1614 lfsck_namespace_filter_linkea_entry(&ldata_new, cname, pfid,
1618 * linkea may change because it doesn't take lock in the first read, if
1619 * it becomes larger, restart from beginning.
1621 if ((ldata_new.ld_leh->leh_reccount > 0 ||
1622 unlikely(ldata_new.ld_leh->leh_overflow_time)) &&
1623 buflen < ldata_new.ld_leh->leh_len) {
1624 dt_write_unlock(env, obj);
1625 dt_trans_stop(env, dev, th);
1626 lfsck_buf_init(&linkea_buf, ldata_new.ld_buf->lb_buf,
1627 ldata_new.ld_leh->leh_len);
1628 buflen = linkea_buf.lb_len;
1633 rc = lfsck_links_write(env, obj, &ldata_new, th);
1635 rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, th);
1637 GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
1640 dt_write_unlock(env, obj);
1643 dt_trans_stop(env, dev, th);
1646 lfsck_ibits_unlock(&lh, LCK_EX);
1649 CDEBUG(D_LFSCK, "%s: namespace LFSCK remove %s linkEA entry "
1650 "for the object: "DFID", parent "DFID", name %.*s\n",
1651 lfsck_lfsck2name(lfsck), next ? "invalid" : "redundant",
1652 PFID(lfsck_dto2fid(obj)), PFID(pfid), cname->ln_namelen,
1656 struct lfsck_namespace *ns = com->lc_file_ram;
1658 ns->ln_flags |= LF_INCONSISTENT;
1665 * Conditionally remove the specified entry from the linkEA.
1667 * Take the parent lock firstly, then check whether the specified
1668 * name entry exists or not: if yes, do nothing; otherwise, call
1669 * lfsck_namespace_shrink_linkea() to remove the linkea entry.
1671 * \param[in] env pointer to the thread context
1672 * \param[in] com pointer to the lfsck component
1673 * \param[in] parent pointer to the parent directory
1674 * \param[in] child pointer to the child object that holds the linkEA
1675 * \param[in,out]ldata pointer to the buffer that holds the linkEA
1676 * \param[in] cname the name for the child in the parent directory
1677 * \param[in] pfid the parent directory's FID for the linkEA
1679 * \retval positive number for repaired cases
1680 * \retval 0 if nothing to be repaired
1681 * \retval negative error number on failure
1683 static int lfsck_namespace_shrink_linkea_cond(const struct lu_env *env,
1684 struct lfsck_component *com,
1685 struct dt_object *parent,
1686 struct dt_object *child,
1687 struct linkea_data *ldata,
1688 struct lu_name *cname,
1689 struct lu_fid *pfid)
1691 struct lfsck_thread_info *info = lfsck_env_info(env);
1692 struct lu_fid *cfid = &info->lti_fid3;
1693 struct lfsck_lock_handle *llh = &info->lti_llh;
1697 rc = lfsck_lock(env, com->lc_lfsck, parent, cname->ln_name, llh,
1698 MDS_INODELOCK_UPDATE, LCK_PR);
1702 dt_read_lock(env, parent, 0);
1703 if (unlikely(lfsck_is_dead_obj(parent))) {
1704 dt_read_unlock(env, parent);
1706 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
1712 rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
1713 (const struct dt_key *)cname->ln_name);
1714 dt_read_unlock(env, parent);
1716 /* It is safe to release the ldlm lock, because when the logic come
1717 * here, we have got all the needed information above whether the
1718 * linkEA entry is valid or not. It is not important that others
1719 * may add new linkEA entry after the ldlm lock released. If other
1720 * has removed the specified linkEA entry by race, then it is OK,
1721 * because the subsequent lfsck_namespace_shrink_linkea() can handle
1724 if (rc == -ENOENT) {
1725 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
1734 /* The LFSCK just found some internal status of cross-MDTs
1735 * create operation. That is normal. */
1736 if (lu_fid_eq(cfid, lfsck_dto2fid(child))) {
1737 linkea_next_entry(ldata);
1742 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata, cname,
1749 * Conditionally replace name entry in the parent.
1751 * As required, the LFSCK may re-create the lost MDT-object for dangling
1752 * name entry, but such repairing may be wrong because of bad FID in the
1753 * name entry. As the LFSCK processing, the real MDT-object may be found,
1754 * then the LFSCK should check whether the former re-created MDT-object
1755 * has been modified or not, if not, then destroy it and update the name
1756 * entry in the parent to reference the real MDT-object.
1758 * \param[in] env pointer to the thread context
1759 * \param[in] com pointer to the lfsck component
1760 * \param[in] parent pointer to the parent directory
1761 * \param[in] child pointer to the MDT-object that may be the real
1762 * MDT-object corresponding to the name entry in parent
1763 * \param[in] cfid the current FID in the name entry
1764 * \param[in] cname contains the name of the child in the parent directory
1766 * \retval positive number for repaired cases
1767 * \retval 0 if nothing to be repaired
1768 * \retval negative error number on failure
1770 static int lfsck_namespace_replace_cond(const struct lu_env *env,
1771 struct lfsck_component *com,
1772 struct dt_object *parent,
1773 struct dt_object *child,
1774 const struct lu_fid *cfid,
1775 const struct lu_name *cname)
1777 struct lfsck_thread_info *info = lfsck_env_info(env);
1778 struct lu_attr *la = &info->lti_la;
1779 struct dt_insert_rec *rec = &info->lti_dt_rec;
1781 struct lfsck_instance *lfsck = com->lc_lfsck;
1782 /* The child and its name may be on different MDTs. */
1783 struct dt_device *dev = lfsck->li_next;
1784 const char *name = cname->ln_name;
1785 const struct lu_fid *pfid = lfsck_dto2fid(parent);
1786 struct dt_object *cobj = NULL;
1787 struct lfsck_lock_handle *pllh = &info->lti_llh;
1788 struct lustre_handle clh = { 0 };
1789 struct linkea_data ldata = { NULL };
1790 struct thandle *th = NULL;
1795 /* @parent/@child may be based on lfsck->li_bottom,
1796 * but here we need the object based on the lfsck->li_next. */
1798 parent = lfsck_object_locate(dev, parent);
1800 GOTO(log, rc = PTR_ERR(parent));
1802 if (unlikely(!dt_try_as_dir(env, parent)))
1803 GOTO(log, rc = -ENOTDIR);
1805 rc = lfsck_lock(env, lfsck, parent, name, pllh,
1806 MDS_INODELOCK_UPDATE, LCK_PW);
1810 if (!fid_is_sane(cfid)) {
1815 cobj = lfsck_object_find_by_dev(env, dev, cfid);
1818 if (rc == -ENOENT) {
1826 if (!dt_object_exists(cobj)) {
1831 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
1832 (const struct dt_key *)name);
1833 if (rc == -ENOENT) {
1841 /* Someone changed the name entry, cannot replace it. */
1842 if (!lu_fid_eq(cfid, &tfid))
1845 /* lock the object to be destroyed. */
1846 rc = lfsck_ibits_lock(env, lfsck, cobj, &clh,
1847 MDS_INODELOCK_UPDATE |
1848 MDS_INODELOCK_UPDATE | MDS_INODELOCK_XATTR,
1853 if (unlikely(lfsck_is_dead_obj(cobj))) {
1858 rc = dt_attr_get(env, cobj, la);
1862 /* The object has been modified by other(s), or it is not created by
1863 * LFSCK, the two cases are indistinguishable. So cannot replace it. */
1864 if (la->la_ctime != 0)
1867 if (S_ISREG(la->la_mode)) {
1868 rc = dt_xattr_get(env, cobj, &LU_BUF_NULL, XATTR_NAME_LOV);
1869 /* If someone has created related OST-object(s),
1871 if ((rc > 0) || (rc < 0 && rc != -ENODATA))
1872 GOTO(log, rc = (rc > 0 ? 0 : rc));
1876 dt_read_lock(env, child, 0);
1877 rc = lfsck_links_read2_with_rec(env, child, &ldata);
1878 dt_read_unlock(env, child);
1880 /* Someone changed the child, no need to replace. */
1887 rc = linkea_links_find(&ldata, cname, pfid);
1888 /* Someone moved the child, no need to replace. */
1892 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1895 th = dt_trans_create(env, dev);
1897 GOTO(log, rc = PTR_ERR(th));
1900 rc = dt_declare_destroy(env, cobj, th);
1905 rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
1909 rec->rec_type = S_IFDIR;
1910 rec->rec_fid = lfsck_dto2fid(child);
1911 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1912 (const struct dt_key *)name, th);
1916 rc = dt_trans_start_local(env, dev, th);
1921 rc = dt_destroy(env, cobj, th);
1926 /* The old name entry maybe not exist. */
1927 rc = dt_delete(env, parent, (const struct dt_key *)name, th);
1928 if (rc != 0 && rc != -ENOENT)
1931 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1932 (const struct dt_key *)name, th);
1934 GOTO(stop, rc = (rc == 0 ? 1 : rc));
1937 dt_trans_stop(env, dev, th);
1940 lfsck_ibits_unlock(&clh, LCK_EX);
1943 if (cobj != NULL && !IS_ERR(cobj))
1944 lfsck_object_put(env, cobj);
1946 CDEBUG(D_LFSCK, "%s: namespace LFSCK conditionally destroy the "
1947 "object "DFID" because of conflict with the object "DFID
1948 " under the parent "DFID" with name %s: rc = %d\n",
1949 lfsck_lfsck2name(lfsck), PFID(cfid),
1950 PFID(lfsck_dto2fid(child)), PFID(pfid), name, rc);
1956 * Overwrite the linkEA for the object with the given ldata.
1958 * The caller should take the ldlm lock before the calling.
1960 * \param[in] env pointer to the thread context
1961 * \param[in] com pointer to the lfsck component
1962 * \param[in] obj pointer to the dt_object to be handled
1963 * \param[in] ldata pointer to the new linkEA data
1965 * \retval positive number for repaired cases
1966 * \retval 0 if nothing to be repaired
1967 * \retval negative error number on failure
1969 int lfsck_namespace_rebuild_linkea(const struct lu_env *env,
1970 struct lfsck_component *com,
1971 struct dt_object *obj,
1972 struct linkea_data *ldata)
1974 struct lfsck_instance *lfsck = com->lc_lfsck;
1975 struct dt_device *dev = lfsck_obj2dev(obj);
1976 struct thandle *th = NULL;
1977 struct lu_buf linkea_buf;
1981 th = dt_trans_create(env, dev);
1983 GOTO(log, rc = PTR_ERR(th));
1985 lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
1986 ldata->ld_leh->leh_len);
1987 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
1988 XATTR_NAME_LINK, 0, th);
1992 rc = dt_trans_start_local(env, dev, th);
1996 dt_write_lock(env, obj, 0);
1997 if (unlikely(lfsck_is_dead_obj(obj)))
1998 GOTO(unlock, rc = 0);
2000 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2001 GOTO(unlock, rc = 1);
2003 rc = dt_xattr_set(env, obj, &linkea_buf,
2004 XATTR_NAME_LINK, 0, th);
2006 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2009 dt_write_unlock(env, obj);
2012 dt_trans_stop(env, dev, th);
2015 CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild linkEA for the "
2016 "object "DFID": rc = %d\n",
2017 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
2020 struct lfsck_namespace *ns = com->lc_file_ram;
2022 ns->ln_flags |= LF_INCONSISTENT;
2029 * Repair invalid name entry.
2031 * If the name entry contains invalid information, such as bad file type
2032 * or (and) corrupted object FID, then either remove the name entry or
2033 * udpate the name entry with the given (right) information.
2035 * \param[in] env pointer to the thread context
2036 * \param[in] com pointer to the lfsck component
2037 * \param[in] parent pointer to the parent directory
2038 * \param[in] child pointer to the object referenced by the name entry
2039 * \param[in] name the old name of the child under the parent directory
2040 * \param[in] name2 the new name of the child under the parent directory
2041 * \param[in] type the type claimed by the name entry
2042 * \param[in] update update the name entry if true; otherwise, remove it
2043 * \param[in] dec decrease the parent nlink count if true
2045 * \retval positive number for repaired successfully
2046 * \retval 0 if nothing to be repaired
2047 * \retval negative error number on failure
2049 int lfsck_namespace_repair_dirent(const struct lu_env *env,
2050 struct lfsck_component *com,
2051 struct dt_object *parent,
2052 struct dt_object *child,
2053 const char *name, const char *name2,
2054 __u16 type, bool update, bool dec)
2056 struct lfsck_thread_info *info = lfsck_env_info(env);
2057 struct dt_insert_rec *rec = &info->lti_dt_rec;
2058 const struct lu_fid *pfid = lfsck_dto2fid(parent);
2059 struct lu_fid cfid = {0};
2061 struct lfsck_instance *lfsck = com->lc_lfsck;
2062 struct dt_device *dev = lfsck->li_next;
2063 struct thandle *th = NULL;
2064 struct lfsck_lock_handle *llh = &info->lti_llh;
2065 struct lustre_handle lh = { 0 };
2070 cfid = *lfsck_dto2fid(child);
2071 parent = lfsck_object_locate(dev, parent);
2073 GOTO(log, rc = PTR_ERR(parent));
2075 if (unlikely(!dt_try_as_dir(env, parent)))
2076 GOTO(log, rc = -ENOTDIR);
2078 if (!update || strcmp(name, name2) == 0)
2079 rc = lfsck_lock(env, lfsck, parent, name, llh,
2080 MDS_INODELOCK_UPDATE, LCK_PW);
2082 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
2083 MDS_INODELOCK_UPDATE, LCK_PW);
2087 th = dt_trans_create(env, dev);
2089 GOTO(unlock1, rc = PTR_ERR(th));
2091 rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
2096 rec->rec_type = lfsck_object_type(child) & S_IFMT;
2097 LASSERT(!fid_is_zero(&cfid));
2098 rec->rec_fid = &cfid;
2099 rc = dt_declare_insert(env, parent,
2100 (const struct dt_rec *)rec,
2101 (const struct dt_key *)name2, th);
2106 if (dec && S_ISDIR(type)) {
2107 rc = dt_declare_ref_del(env, parent, th);
2112 rc = dt_trans_start_local(env, dev, th);
2117 dt_write_lock(env, parent, 0);
2118 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
2119 (const struct dt_key *)name);
2120 /* Someone has removed the bad name entry by race. */
2122 GOTO(unlock2, rc = 0);
2127 /* Someone has removed the bad name entry and reused it for other
2128 * object by race. */
2129 if (!lu_fid_eq(&tfid, &cfid))
2130 GOTO(unlock2, rc = 0);
2132 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2133 GOTO(unlock2, rc = 1);
2135 rc = dt_delete(env, parent, (const struct dt_key *)name, th);
2140 rc = dt_insert(env, parent,
2141 (const struct dt_rec *)rec,
2142 (const struct dt_key *)name2, th);
2147 if (dec && S_ISDIR(type)) {
2148 rc = dt_ref_del(env, parent, th);
2153 GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
2156 dt_write_unlock(env, parent);
2159 dt_trans_stop(env, dev, th);
2161 /* We are not sure whether the child will become orphan or not.
2162 * Record it in the LFSCK trace file for further checking in
2163 * the second-stage scanning. */
2164 if (!update && !dec && child && rc == 0)
2165 lfsck_namespace_trace_update(env, com, &cfid,
2166 LNTF_CHECK_LINKEA, true);
2169 /* It is harmless even if unlock the unused lock_handle */
2170 lfsck_ibits_unlock(&lh, LCK_PW);
2174 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found bad name "
2175 "entry for: parent "DFID", child "DFID", name %s, type "
2176 "in name entry %o, type claimed by child %o. repair it "
2177 "by %s with new name2 %s: rc = %d\n",
2178 lfsck_lfsck2name(lfsck), PFID(pfid), PFID(&cfid),
2179 name, type, update ? lfsck_object_type(child) : 0,
2180 update ? "updating" : "removing", name2, rc);
2183 struct lfsck_namespace *ns = com->lc_file_ram;
2185 ns->ln_flags |= LF_INCONSISTENT;
2192 * Update the ".." name entry for the given object.
2194 * The object's ".." is corrupted, this function will update the ".." name
2195 * entry with the given pfid, and the linkEA with the given ldata.
2197 * The caller should take the ldlm lock before the calling.
2199 * \param[in] env pointer to the thread context
2200 * \param[in] com pointer to the lfsck component
2201 * \param[in] obj pointer to the dt_object to be handled
2202 * \param[in] pfid the new fid for the object's ".." name entry
2203 * \param[in] cname the name for the @obj in the parent directory
2205 * \retval positive number for repaired cases
2206 * \retval 0 if nothing to be repaired
2207 * \retval negative error number on failure
2209 static int lfsck_namespace_repair_unmatched_pairs(const struct lu_env *env,
2210 struct lfsck_component *com,
2211 struct dt_object *obj,
2212 const struct lu_fid *pfid,
2213 struct lu_name *cname)
2215 struct lfsck_thread_info *info = lfsck_env_info(env);
2216 struct dt_insert_rec *rec = &info->lti_dt_rec;
2217 struct lfsck_instance *lfsck = com->lc_lfsck;
2218 struct dt_device *dev = lfsck_obj2dev(obj);
2219 struct thandle *th = NULL;
2220 struct linkea_data ldata = { NULL };
2221 struct lu_buf linkea_buf;
2225 LASSERT(!dt_object_remote(obj));
2226 LASSERT(S_ISDIR(lfsck_object_type(obj)));
2228 rc = linkea_links_new(&ldata, &info->lti_big_buf, cname, pfid);
2232 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
2233 ldata.ld_leh->leh_len);
2235 th = dt_trans_create(env, dev);
2237 GOTO(log, rc = PTR_ERR(th));
2239 rc = dt_declare_delete(env, obj, (const struct dt_key *)dotdot, th);
2243 rec->rec_type = S_IFDIR;
2244 rec->rec_fid = pfid;
2245 rc = dt_declare_insert(env, obj, (const struct dt_rec *)rec,
2246 (const struct dt_key *)dotdot, th);
2250 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
2251 XATTR_NAME_LINK, 0, th);
2255 rc = dt_trans_start_local(env, dev, th);
2259 dt_write_lock(env, obj, 0);
2260 if (unlikely(lfsck_is_dead_obj(obj)))
2261 GOTO(unlock, rc = 0);
2263 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2264 GOTO(unlock, rc = 1);
2266 /* The old ".." name entry maybe not exist. */
2267 dt_delete(env, obj, (const struct dt_key *)dotdot, th);
2269 rc = dt_insert(env, obj, (const struct dt_rec *)rec,
2270 (const struct dt_key *)dotdot, th);
2274 rc = lfsck_links_write(env, obj, &ldata, th);
2276 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2279 dt_write_unlock(env, obj);
2282 dt_trans_stop(env, dev, th);
2285 CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild dotdot name entry for "
2286 "the object "DFID", new parent "DFID": rc = %d\n",
2287 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)),
2291 struct lfsck_namespace *ns = com->lc_file_ram;
2293 ns->ln_flags |= LF_INCONSISTENT;
2300 * Handle orphan @obj during Double Scan Directory.
2302 * Remove the @obj's current (invalid) linkEA entries, and insert
2303 * it in the directory .lustre/lost+found/MDTxxxx/ with the name:
2304 * ${FID}-${PFID}-D-${conflict_version}
2306 * The caller should take the ldlm lock before the calling.
2308 * \param[in] env pointer to the thread context
2309 * \param[in] com pointer to the lfsck component
2310 * \param[in] obj pointer to the orphan object to be handled
2311 * \param[in] pfid the new fid for the object's ".." name entry
2312 * \param[in,out] lh ldlm lock handler for the given @obj
2313 * \param[out] type to tell the caller what the inconsistency is
2315 * \retval positive number for repaired cases
2316 * \retval 0 if nothing to be repaired
2317 * \retval negative error number on failure
2320 lfsck_namespace_dsd_orphan(const struct lu_env *env,
2321 struct lfsck_component *com,
2322 struct dt_object *obj,
2323 const struct lu_fid *pfid,
2324 struct lustre_handle *lh,
2325 enum lfsck_namespace_inconsistency_type *type)
2327 struct lfsck_thread_info *info = lfsck_env_info(env);
2328 struct lfsck_namespace *ns = com->lc_file_ram;
2332 /* Remove the unrecognized linkEA. */
2333 rc = lfsck_namespace_links_remove(env, com, obj);
2334 lfsck_ibits_unlock(lh, LCK_EX);
2335 if (rc < 0 && rc != -ENODATA)
2338 *type = LNIT_MUL_REF;
2340 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
2341 * ever tried to verify some remote MDT-object that resides on this
2342 * MDT, but this MDT failed to respond such request. So means there
2343 * may be some remote name entry on other MDT that references this
2344 * object with another name, so we cannot know whether this linkEA
2345 * is valid or not. So keep it there and maybe resolved when next
2347 if (ns->ln_flags & LF_INCOMPLETE)
2350 /* The unique linkEA is invalid, even if the ".." name entry may be
2351 * valid, we still cannot know via which name entry this directory
2352 * will be referenced. Then handle it as pure orphan. */
2353 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2354 "-"DFID, PFID(pfid));
2355 rc = lfsck_namespace_insert_orphan(env, com, obj,
2356 info->lti_tmpbuf, "D", NULL);
2362 * Double Scan Directory object for single linkEA entry case.
2364 * The given @child has unique linkEA entry. If the linkEA entry is valid,
2365 * then check whether the name is in the namespace or not, if not, add the
2366 * missing name entry back to namespace. If the linkEA entry is invalid,
2367 * then remove it and insert the @child in the .lustre/lost+found/MDTxxxx/
2370 * \param[in] env pointer to the thread context
2371 * \param[in] com pointer to the lfsck component
2372 * \param[in] child pointer to the directory to be double scanned
2373 * \param[in] pfid the FID corresponding to the ".." entry
2374 * \param[in] ldata pointer to the linkEA data for the given @child
2375 * \param[in,out] lh ldlm lock handler for the given @child
2376 * \param[out] type to tell the caller what the inconsistency is
2377 * \param[in] retry if found inconsistency, but the caller does not hold
2378 * ldlm lock on the @child, then set @retry as true
2379 * \param[in] unknown set if does not know how to repair the inconsistency
2381 * \retval positive number for repaired cases
2382 * \retval 0 if nothing to be repaired
2383 * \retval negative error number on failure
2386 lfsck_namespace_dsd_single(const struct lu_env *env,
2387 struct lfsck_component *com,
2388 struct dt_object *child,
2389 const struct lu_fid *pfid,
2390 struct linkea_data *ldata,
2391 struct lustre_handle *lh,
2392 enum lfsck_namespace_inconsistency_type *type,
2393 bool *retry, bool *unknown)
2395 struct lfsck_thread_info *info = lfsck_env_info(env);
2396 struct lu_name *cname = &info->lti_name;
2397 const struct lu_fid *cfid = lfsck_dto2fid(child);
2399 struct lfsck_namespace *ns = com->lc_file_ram;
2400 struct lfsck_instance *lfsck = com->lc_lfsck;
2401 struct dt_object *parent = NULL;
2402 struct lmv_mds_md_v1 *lmv;
2406 rc = lfsck_namespace_unpack_linkea_entry(ldata, cname, &tfid,
2408 sizeof(info->lti_key));
2409 /* The unique linkEA entry with bad parent will be handled as orphan. */
2411 if (!lustre_handle_is_used(lh) && retry != NULL)
2414 rc = lfsck_namespace_dsd_orphan(env, com, child,
2420 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
2422 GOTO(out, rc = PTR_ERR(parent));
2424 /* We trust the unique linkEA entry in spite of whether it matches the
2425 * ".." name entry or not. Because even if the linkEA entry is wrong
2426 * and the ".." name entry is right, we still cannot know via which
2427 * name entry the child will be referenced, since all known entries
2428 * have been verified during the first-stage scanning. */
2429 if (!dt_object_exists(parent)) {
2430 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2431 * has ever tried to verify some remote MDT-object that resides
2432 * on this MDT, but this MDT failed to respond such request. So
2433 * means there may be some remote name entry on other MDT that
2434 * references this object with another name, so we cannot know
2435 * whether this linkEA is valid or not. So keep it there and
2436 * maybe resolved when next LFSCK run. */
2437 if (ns->ln_flags & LF_INCOMPLETE)
2440 if (!lustre_handle_is_used(lh) && retry != NULL) {
2446 lfsck_ibits_unlock(lh, LCK_EX);
2449 lmv = &info->lti_lmv;
2450 rc = lfsck_read_stripe_lmv(env, lfsck, child, lmv);
2451 if (rc != 0 && rc != -ENODATA)
2454 if (rc == -ENODATA || lmv->lmv_magic != LMV_MAGIC_STRIPE) {
2456 } else if (lfsck_shard_name_to_index(env,
2457 cname->ln_name, cname->ln_namelen,
2458 S_IFDIR, cfid) < 0) {
2459 /* It is an invalid name entry, we
2460 * cannot trust the parent also. */
2461 rc = lfsck_namespace_shrink_linkea(env, com, child,
2462 ldata, cname, &tfid, true);
2466 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2467 "-"DFID, PFID(pfid));
2468 rc = lfsck_namespace_insert_orphan(env, com, child,
2469 info->lti_tmpbuf, "S", NULL);
2474 /* Create the lost parent as an orphan. */
2475 rc = lfsck_namespace_create_orphan_dir(env, com, parent, lmv);
2477 /* Add the missing name entry to the parent. */
2478 rc = lfsck_namespace_insert_normal(env, com, parent,
2479 child, cname->ln_name);
2480 if (unlikely(rc == -EEXIST)) {
2481 /* Unfortunately, someone reused the name
2482 * under the parent by race. So we have
2483 * to remove the linkEA entry from
2484 * current child object. It means that the
2485 * LFSCK cannot recover the system
2486 * totally back to its original status,
2487 * but it is necessary to make the
2488 * current system to be consistent. */
2489 rc = lfsck_namespace_shrink_linkea(env,
2491 cname, &tfid, true);
2493 snprintf(info->lti_tmpbuf,
2494 sizeof(info->lti_tmpbuf),
2495 "-"DFID, PFID(pfid));
2496 rc = lfsck_namespace_insert_orphan(env,
2497 com, child, info->lti_tmpbuf,
2504 } /* !dt_object_exists(parent) */
2506 /* The unique linkEA entry with bad parent will be handled as orphan. */
2507 if (unlikely(!dt_try_as_dir(env, parent))) {
2508 if (!lustre_handle_is_used(lh) && retry != NULL)
2511 rc = lfsck_namespace_dsd_orphan(env, com, child,
2517 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
2518 (const struct dt_key *)cname->ln_name);
2519 if (rc == -ENOENT) {
2520 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2521 * has ever tried to verify some remote MDT-object that resides
2522 * on this MDT, but this MDT failed to respond such request. So
2523 * means there may be some remote name entry on other MDT that
2524 * references this object with another name, so we cannot know
2525 * whether this linkEA is valid or not. So keep it there and
2526 * maybe resolved when next LFSCK run. */
2527 if (ns->ln_flags & LF_INCOMPLETE)
2530 if (!lustre_handle_is_used(lh) && retry != NULL) {
2536 lfsck_ibits_unlock(lh, LCK_EX);
2537 rc = lfsck_namespace_check_name(env, lfsck, parent, child,
2545 /* It is an invalid name entry, drop it. */
2546 if (unlikely(rc > 0)) {
2547 rc = lfsck_namespace_shrink_linkea(env, com, child,
2548 ldata, cname, &tfid, true);
2550 snprintf(info->lti_tmpbuf,
2551 sizeof(info->lti_tmpbuf),
2552 "-"DFID, PFID(pfid));
2553 rc = lfsck_namespace_insert_orphan(env, com,
2554 child, info->lti_tmpbuf, "D", NULL);
2560 /* Add the missing name entry back to the namespace. */
2561 rc = lfsck_namespace_insert_normal(env, com, parent, child,
2563 if (unlikely(rc == -ESTALE))
2564 /* It may happen when the remote object has been
2565 * removed, but the local MDT is not aware of that. */
2568 if (unlikely(rc == -EEXIST)) {
2569 /* Unfortunately, someone reused the name under the
2570 * parent by race. So we have to remove the linkEA
2571 * entry from current child object. It means that the
2572 * LFSCK cannot recover the system totally back to
2573 * its original status, but it is necessary to make
2574 * the current system to be consistent.
2576 * It also may be because of the LFSCK found some
2577 * internal status of create operation. Under such
2578 * case, nothing to be done. */
2579 rc = lfsck_namespace_shrink_linkea_cond(env, com,
2580 parent, child, ldata, cname, &tfid);
2582 snprintf(info->lti_tmpbuf,
2583 sizeof(info->lti_tmpbuf),
2584 "-"DFID, PFID(pfid));
2585 rc = lfsck_namespace_insert_orphan(env, com,
2586 child, info->lti_tmpbuf, "D", NULL);
2591 } /* rc == -ENOENT */
2596 if (!lu_fid_eq(&tfid, cfid)) {
2597 if (!lustre_handle_is_used(lh) && retry != NULL) {
2603 lfsck_ibits_unlock(lh, LCK_EX);
2604 /* The name entry references another MDT-object that
2605 * may be created by the LFSCK for repairing dangling
2606 * name entry. Try to replace it. */
2607 rc = lfsck_namespace_replace_cond(env, com, parent, child,
2610 rc = lfsck_namespace_dsd_orphan(env, com, child,
2616 /* Zero FID may because the remote directroy object has invalid linkEA,
2617 * or lost linkEA. Under such case, the LFSCK on this MDT does not know
2618 * how to repair the inconsistency, but the namespace LFSCK on the MDT
2619 * where its name entry resides may has more information (name, FID) to
2620 * repair such inconsistency. So here, keep the inconsistency to avoid
2621 * some imporper repairing. */
2622 if (fid_is_zero(pfid)) {
2629 /* The ".." name entry is wrong, update it. */
2630 if (!lu_fid_eq(pfid, lfsck_dto2fid(parent))) {
2631 if (!lustre_handle_is_used(lh) && retry != NULL) {
2637 *type = LNIT_UNMATCHED_PAIRS;
2638 rc = lfsck_namespace_repair_unmatched_pairs(env, com, child,
2639 lfsck_dto2fid(parent), cname);
2645 if (parent != NULL && !IS_ERR(parent))
2646 lfsck_object_put(env, parent);
2652 * Double Scan Directory object for multiple linkEA entries case.
2654 * The given @child has multiple linkEA entries. There is at most one linkEA
2655 * entry will be valid, all the others will be removed. Firstly, the function
2656 * will try to find out the linkEA entry for which the name entry exists under
2657 * the given parent (@pfid). If there is no linkEA entry that matches the given
2658 * ".." name entry, then tries to find out the first linkEA entry that both the
2659 * parent and the name entry exist to rebuild a new ".." name entry.
2661 * \param[in] env pointer to the thread context
2662 * \param[in] com pointer to the lfsck component
2663 * \param[in] child pointer to the directory to be double scanned
2664 * \param[in] pfid the FID corresponding to the ".." entry
2665 * \param[in] ldata pointer to the linkEA data for the given @child
2666 * \param[in,out] lh ldlm lock handler for the given @child
2667 * \param[out] type to tell the caller what the inconsistency is
2668 * \param[in] lpf true if the ".." entry is under lost+found/MDTxxxx/
2669 * \param[in] unknown set if does not know how to repair the inconsistency
2671 * \retval positive number for repaired cases
2672 * \retval 0 if nothing to be repaired
2673 * \retval negative error number on failure
2676 lfsck_namespace_dsd_multiple(const struct lu_env *env,
2677 struct lfsck_component *com,
2678 struct dt_object *child,
2679 const struct lu_fid *pfid,
2680 struct linkea_data *ldata,
2681 struct lustre_handle *lh,
2682 enum lfsck_namespace_inconsistency_type *type,
2683 bool lpf, bool *unknown)
2685 struct lfsck_thread_info *info = lfsck_env_info(env);
2686 struct lu_name *cname = &info->lti_name;
2687 const struct lu_fid *cfid = lfsck_dto2fid(child);
2688 struct lu_fid *pfid2 = &info->lti_fid3;
2690 struct lfsck_namespace *ns = com->lc_file_ram;
2691 struct lfsck_instance *lfsck = com->lc_lfsck;
2692 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
2693 struct dt_object *parent = NULL;
2694 struct linkea_data ldata_new = { NULL };
2695 int dirent_count = 0;
2701 while (ldata->ld_lee != NULL) {
2702 rc = lfsck_namespace_unpack_linkea_entry(ldata, cname, &tfid,
2704 sizeof(info->lti_key));
2705 /* Drop invalid linkEA entry. */
2707 lfsck_linkea_del_buf(ldata, cname);
2711 /* Drop repeated linkEA entries. */
2712 lfsck_namespace_filter_linkea_entry(ldata, cname, &tfid, true);
2714 /* If current dotdot is the .lustre/lost+found/MDTxxxx/,
2715 * then it is possible that: the directry object has ever
2716 * been lost, but its name entry was there. In the former
2717 * LFSCK run, during the first-stage scanning, the LFSCK
2718 * found the dangling name entry, but it did not recreate
2719 * the lost object, and when moved to the second-stage
2720 * scanning, some children objects of the lost directory
2721 * object were found, then the LFSCK recreated such lost
2722 * directory object as an orphan.
2724 * When the LFSCK runs again, if the dangling name is still
2725 * there, the LFSCK should move the orphan directory object
2726 * back to the normal namespace. */
2727 if (!lpf && !fid_is_zero(pfid) &&
2728 !lu_fid_eq(pfid, &tfid) && once) {
2729 linkea_next_entry(ldata);
2733 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
2735 RETURN(PTR_ERR(parent));
2737 if (!dt_object_exists(parent)) {
2738 lfsck_object_put(env, parent);
2739 if (ldata->ld_leh->leh_reccount > 1) {
2740 /* If it is NOT the last linkEA entry, then
2741 * there is still other chance to make the
2742 * child to be visible via other parent, then
2743 * remove this linkEA entry. */
2744 lfsck_linkea_del_buf(ldata, cname);
2751 /* The linkEA entry with bad parent will be removed. */
2752 if (unlikely(!dt_try_as_dir(env, parent))) {
2753 lfsck_object_put(env, parent);
2754 lfsck_linkea_del_buf(ldata, cname);
2758 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
2759 (const struct dt_key *)cname->ln_name);
2760 *pfid2 = *lfsck_dto2fid(parent);
2761 if (rc == -ENOENT) {
2762 lfsck_object_put(env, parent);
2763 linkea_next_entry(ldata);
2768 lfsck_object_put(env, parent);
2773 if (lu_fid_eq(&tfid, cfid)) {
2774 lfsck_object_put(env, parent);
2775 /* If the parent (that is declared via linkEA entry)
2776 * directory contains the specified child, but such
2777 * parent does not match the dotdot name entry, then
2778 * trust the linkEA. */
2779 if (!fid_is_zero(pfid) && !lu_fid_eq(pfid, pfid2)) {
2780 *type = LNIT_UNMATCHED_PAIRS;
2781 rc = lfsck_namespace_repair_unmatched_pairs(env,
2782 com, child, pfid2, cname);
2788 /* It is the most common case that we find the
2789 * name entry corresponding to the linkEA entry
2790 * that matches the ".." name entry. */
2791 rc = linkea_links_new(&ldata_new, &info->lti_big_buf,
2796 rc = lfsck_namespace_rebuild_linkea(env, com, child,
2801 lfsck_linkea_del_buf(ldata, cname);
2802 linkea_first_entry(ldata);
2803 /* There may be some invalid dangling name entries under
2804 * other parent directories, remove all of them. */
2805 while (ldata->ld_lee != NULL) {
2806 rc = lfsck_namespace_unpack_linkea_entry(ldata,
2807 cname, &tfid, info->lti_key,
2808 sizeof(info->lti_key));
2812 parent = lfsck_object_find_bottom(env, lfsck,
2814 if (IS_ERR(parent)) {
2815 rc = PTR_ERR(parent);
2816 if (rc != -ENOENT &&
2817 bk->lb_param & LPF_FAILOUT)
2823 if (!dt_object_exists(parent)) {
2824 lfsck_object_put(env, parent);
2828 rc = lfsck_namespace_repair_dirent(env, com,
2829 parent, child, cname->ln_name,
2830 cname->ln_name, S_IFDIR, false, true);
2831 lfsck_object_put(env, parent);
2833 if (bk->lb_param & LPF_FAILOUT)
2842 lfsck_linkea_del_buf(ldata, cname);
2845 ns->ln_dirent_repaired += dirent_count;
2848 } /* lu_fid_eq(&tfid, lfsck_dto2fid(child)) */
2850 lfsck_ibits_unlock(lh, LCK_EX);
2851 /* The name entry references another MDT-object that may be
2852 * created by the LFSCK for repairing dangling name entry.
2853 * Try to replace it. */
2854 rc = lfsck_namespace_replace_cond(env, com, parent, child,
2856 lfsck_object_put(env, parent);
2863 lfsck_linkea_del_buf(ldata, cname);
2864 } /* while (ldata->ld_lee != NULL) */
2866 /* If there is still linkEA overflow, return. */
2867 if (unlikely(ldata->ld_leh->leh_overflow_time))
2870 linkea_first_entry(ldata);
2871 if (ldata->ld_leh->leh_reccount == 1) {
2872 rc = lfsck_namespace_dsd_single(env, com, child, pfid, ldata,
2873 lh, type, NULL, unknown);
2878 /* All linkEA entries are invalid and removed, then handle the @child
2880 if (ldata->ld_leh->leh_reccount == 0) {
2881 rc = lfsck_namespace_dsd_orphan(env, com, child, pfid, lh,
2887 /* If the dangling name entry for the orphan directory object has
2888 * been remvoed, then just check whether the directory object is
2889 * still under the .lustre/lost+found/MDTxxxx/ or not. */
2895 /* There is no linkEA entry that matches the ".." name entry. Find
2896 * the first linkEA entry that both parent and name entry exist to
2897 * rebuild a new ".." name entry. */
2907 * Repair the object's nlink attribute.
2909 * If all the known name entries have been verified, then the object's hard
2910 * link attribute should match the object's linkEA entries count unless the
2911 * object's has too many hard link to be recorded in the linkEA. Such cases
2912 * should have been marked in the LFSCK trace file. Otherwise, trust the
2913 * linkEA to update the object's nlink attribute.
2915 * \param[in] env pointer to the thread context
2916 * \param[in] com pointer to the lfsck component
2917 * \param[in] obj pointer to the dt_object to be handled
2918 * \param[in,out] la pointer to buffer to object's attribute before
2919 * and after the repairing
2921 * \retval positive number for repaired cases
2922 * \retval 0 if nothing to be repaired
2923 * \retval negative error number on failure
2925 static int lfsck_namespace_repair_nlink(const struct lu_env *env,
2926 struct lfsck_component *com,
2927 struct dt_object *obj,
2930 struct lfsck_namespace *ns = com->lc_file_ram;
2931 struct lfsck_instance *lfsck = com->lc_lfsck;
2932 struct dt_device *dev = lfsck_obj2dev(obj);
2933 const struct lu_fid *cfid = lfsck_dto2fid(obj);
2934 struct thandle *th = NULL;
2935 struct linkea_data ldata = { NULL };
2936 struct lustre_handle lh = { 0 };
2937 __u32 old = la->la_nlink;
2941 LASSERT(!dt_object_remote(obj));
2943 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
2944 MDS_INODELOCK_UPDATE, LCK_PW);
2948 th = dt_trans_create(env, dev);
2950 GOTO(log, rc = PTR_ERR(th));
2952 la->la_valid = LA_NLINK;
2953 rc = dt_declare_attr_set(env, obj, la, th);
2957 rc = dt_trans_start_local(env, dev, th);
2961 dt_write_lock(env, obj, 0);
2962 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
2963 * ever tried to verify some remote MDT-object that resides on this
2964 * MDT, but this MDT failed to respond such request. So means there
2965 * may be some remote name entry on other MDT that references this
2966 * object with another name, so we cannot know whether this linkEA
2967 * is valid or not. So keep it there and maybe resolved when next
2969 if (ns->ln_flags & LF_INCOMPLETE)
2970 GOTO(unlock, rc = 0);
2972 rc = dt_attr_get(env, obj, la);
2974 GOTO(unlock, rc = (rc == -ENOENT ? 0 : rc));
2976 rc = lfsck_links_read2_with_rec(env, obj, &ldata);
2978 GOTO(unlock, rc = (rc == -ENODATA ? 0 : rc));
2980 /* XXX: Currently, we only update the nlink attribute if the known
2981 * linkEA entries is larger than the nlink attribute. That is
2983 if (la->la_nlink >= ldata.ld_leh->leh_reccount ||
2984 unlikely(la->la_nlink == 0 ||
2985 ldata.ld_leh->leh_overflow_time))
2986 GOTO(unlock, rc = 0);
2988 la->la_nlink = ldata.ld_leh->leh_reccount;
2989 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2990 GOTO(unlock, rc = 1);
2992 rc = dt_attr_set(env, obj, la, th);
2994 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2997 dt_write_unlock(env, obj);
3000 dt_trans_stop(env, dev, th);
3003 lfsck_ibits_unlock(&lh, LCK_PW);
3004 CDEBUG(D_LFSCK, "%s: namespace LFSCK repaired the object "DFID"'s "
3005 "nlink count from %u to %u: rc = %d\n",
3006 lfsck_lfsck2name(lfsck), PFID(cfid), old, la->la_nlink, rc);
3009 ns->ln_flags |= LF_INCONSISTENT;
3015 * Double scan the directory object for namespace LFSCK.
3017 * This function will verify the <parent, child> pairs in the namespace tree:
3018 * the parent references the child via some name entry that should be in the
3019 * child's linkEA entry, the child should back references the parent via its
3022 * The LFSCK will scan every linkEA entry in turn until find out the first
3023 * matched pairs. If found, then all other linkEA entries will be dropped.
3024 * If all the linkEA entries cannot match the ".." name entry, then there
3025 * are serveral possible cases:
3027 * 1) If there is only one linkEA entry, then trust it as long as the PFID
3028 * in the linkEA entry is valid.
3030 * 2) If there are multiple linkEA entries, then try to find the linkEA
3031 * that matches the ".." name entry. If found, then all other entries
3032 * are invalid; otherwise, it is quite possible that the ".." name entry
3033 * is corrupted. Under such case, the LFSCK will rebuild the ".." name
3034 * entry according to the first valid linkEA entry (both the parent and
3035 * the name entry should exist).
3037 * 3) If the directory object has no (valid) linkEA entry, then the
3038 * directory object will be handled as pure orphan and inserted
3039 * in the .lustre/lost+found/MDTxxxx/ with the name:
3040 * ${self_FID}-${PFID}-D-${conflict_version}
3042 * \param[in] env pointer to the thread context
3043 * \param[in] com pointer to the lfsck component
3044 * \param[in] child pointer to the directory object to be handled
3045 * \param[in] flags to indicate the specical checking on the @child
3047 * \retval positive number for repaired cases
3048 * \retval 0 if nothing to be repaired
3049 * \retval negative error number on failure
3051 static int lfsck_namespace_double_scan_dir(const struct lu_env *env,
3052 struct lfsck_component *com,
3053 struct dt_object *child, __u8 flags)
3055 struct lfsck_thread_info *info = lfsck_env_info(env);
3056 const struct lu_fid *cfid = lfsck_dto2fid(child);
3057 struct lu_fid *pfid = &info->lti_fid2;
3058 struct lfsck_namespace *ns = com->lc_file_ram;
3059 struct lfsck_instance *lfsck = com->lc_lfsck;
3060 struct lustre_handle lh = { 0 };
3061 struct linkea_data ldata = { NULL };
3062 bool unknown = false;
3065 enum lfsck_namespace_inconsistency_type type = LNIT_BAD_LINKEA;
3069 LASSERT(!dt_object_remote(child));
3071 if (flags & LNTF_UNCERTAIN_LMV) {
3072 if (flags & LNTF_RECHECK_NAME_HASH) {
3073 rc = lfsck_namespace_scan_shard(env, com, child);
3077 ns->ln_striped_shards_scanned++;
3079 ns->ln_striped_shards_skipped++;
3083 flags &= ~(LNTF_RECHECK_NAME_HASH | LNTF_UNCERTAIN_LMV);
3087 if (flags & (LNTF_CHECK_LINKEA | LNTF_CHECK_PARENT) &&
3088 !(lfsck->li_bookmark_ram.lb_param & LPF_ALL_TGT)) {
3089 CDEBUG(D_LFSCK, "%s: some MDT(s) maybe NOT take part in the"
3090 "the namespace LFSCK, then the LFSCK cannot guarantee"
3091 "all the name entries have been verified in first-stage"
3092 "scanning. So have to skip orphan related handling for"
3093 "the directory object "DFID" with remote name entry\n",
3094 lfsck_lfsck2name(lfsck), PFID(cfid));
3099 if (unlikely(!dt_try_as_dir(env, child)))
3100 GOTO(out, rc = -ENOTDIR);
3102 /* We only take ldlm lock on the @child when required. When the
3103 * logic comes here for the first time, it is always false. */
3107 rc = lfsck_ibits_lock(env, lfsck, child, &lh,
3108 MDS_INODELOCK_UPDATE |
3109 MDS_INODELOCK_XATTR, LCK_EX);
3114 dt_read_lock(env, child, 0);
3115 if (unlikely(lfsck_is_dead_obj(child))) {
3116 dt_read_unlock(env, child);
3121 rc = dt_lookup(env, child, (struct dt_rec *)pfid,
3122 (const struct dt_key *)dotdot);
3124 if (rc != -ENOENT && rc != -ENODATA && rc != -EINVAL) {
3125 dt_read_unlock(env, child);
3130 if (!lustre_handle_is_used(&lh)) {
3131 dt_read_unlock(env, child);
3136 } else if (lfsck->li_lpf_obj != NULL &&
3137 lu_fid_eq(pfid, lfsck_dto2fid(lfsck->li_lpf_obj))) {
3139 } else if (unlikely(!fid_is_sane(pfid))) {
3143 rc = lfsck_links_read(env, child, &ldata);
3144 dt_read_unlock(env, child);
3146 if (rc != -ENODATA && rc != -EINVAL)
3149 if (!lustre_handle_is_used(&lh))
3152 if (rc == -EINVAL && !fid_is_zero(pfid)) {
3153 /* Remove the corrupted linkEA. */
3154 rc = lfsck_namespace_links_remove(env, com, child);
3156 /* Here, because of the crashed linkEA, we
3157 * cannot know whether there is some parent
3158 * that references the child directory via
3159 * some name entry or not. So keep it there,
3160 * when the LFSCK run next time, if there is
3161 * some parent that references this object,
3162 * then the LFSCK can rebuild the linkEA;
3163 * otherwise, this object will be handled
3164 * as orphan as above. */
3167 /* 1. If we have neither ".." nor linkEA,
3168 * then it is an orphan.
3170 * 2. If we only have the ".." name entry,
3171 * but no parent references this child
3172 * directory, then handle it as orphan. */
3173 lfsck_ibits_unlock(&lh, LCK_EX);
3174 type = LNIT_MUL_REF;
3176 /* If the LFSCK is marked as LF_INCOMPLETE,
3177 * then means some MDT has ever tried to
3178 * verify some remote MDT-object that resides
3179 * on this MDT, but this MDT failed to respond
3180 * such request. So means there may be some
3181 * remote name entry on other MDT that
3182 * references this object with another name,
3183 * so we cannot know whether this linkEA is
3184 * valid or not. So keep it there and maybe
3185 * resolved when next LFSCK run. */
3186 if (ns->ln_flags & LF_INCOMPLETE)
3189 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
3190 "-"DFID, PFID(pfid));
3191 rc = lfsck_namespace_insert_orphan(env, com, child,
3192 info->lti_tmpbuf, "D", NULL);
3198 linkea_first_entry(&ldata);
3199 /* This is the most common case: the object has unique linkEA entry. */
3200 if (ldata.ld_leh->leh_reccount == 1) {
3201 rc = lfsck_namespace_dsd_single(env, com, child, pfid, &ldata,
3202 &lh, &type, &retry, &unknown);
3204 LASSERT(!lustre_handle_is_used(&lh));
3213 if (!lustre_handle_is_used(&lh))
3216 if (unlikely(ldata.ld_leh->leh_reccount == 0)) {
3217 rc = lfsck_namespace_dsd_orphan(env, com, child, pfid, &lh,
3223 /* When we come here, the cases usually like that:
3224 * 1) The directory object has a corrupted linkEA entry. During the
3225 * first-stage scanning, the LFSCK cannot know such corruption,
3226 * then it appends the right linkEA entry according to the found
3227 * name entry after the bad one.
3229 * 2) The directory object has a right linkEA entry. During the
3230 * first-stage scanning, the LFSCK finds some bad name entry,
3231 * but the LFSCK cannot aware that at that time, then it adds
3232 * the bad linkEA entry for further processing. */
3233 rc = lfsck_namespace_dsd_multiple(env, com, child, pfid, &ldata,
3234 &lh, &type, lpf, &unknown);
3239 lfsck_ibits_unlock(&lh, LCK_EX);
3242 case LNIT_BAD_LINKEA:
3243 ns->ln_linkea_repaired++;
3245 case LNIT_UNMATCHED_PAIRS:
3246 ns->ln_unmatched_pairs_repaired++;
3249 ns->ln_mul_ref_repaired++;
3257 ns->ln_unknown_inconsistency++;
3263 lfsck_namespace_linkea_stale_overflow(struct linkea_data *ldata,
3264 struct lfsck_namespace *ns)
3266 /* Both the leh_overflow_time and ln_time_latest_reset are
3267 * local time based, so need NOT to care about clock drift
3268 * among the servers. */
3269 return ldata->ld_leh->leh_overflow_time &&
3270 ldata->ld_leh->leh_overflow_time < ns->ln_time_latest_reset;
3274 * Clear the object's linkEA overflow timestamp.
3276 * If the MDT-object has too many hard links as to the linkEA cannot hold
3277 * all of them, then overflow timestamp will be set in the linkEA header.
3278 * If some hard links are removed after that, then it is possible to hold
3279 * other missed linkEA entries. If the namespace LFSCK have added all the
3280 * related linkEA entries, then it will remove the overflow timestamp.
3282 * \param[in] env pointer to the thread context
3283 * \param[in] com pointer to the lfsck component
3284 * \param[in] ldata pointer to the linkEA data for the given @obj
3285 * \param[in] obj pointer to the dt_object to be handled
3287 * \retval positive number for repaired cases
3288 * \retval 0 if nothing to be repaired
3289 * \retval negative error number on failure
3291 static int lfsck_namespace_linkea_clear_overflow(const struct lu_env *env,
3292 struct lfsck_component *com,
3293 struct linkea_data *ldata,
3294 struct dt_object *obj)
3296 struct lfsck_namespace *ns = com->lc_file_ram;
3297 struct lfsck_instance *lfsck = com->lc_lfsck;
3298 struct dt_device *dev = lfsck_obj2dev(obj);
3299 struct thandle *th = NULL;
3300 struct lustre_handle lh = { 0 };
3301 struct lu_buf linkea_buf;
3305 LASSERT(!dt_object_remote(obj));
3307 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
3308 MDS_INODELOCK_UPDATE, LCK_PW);
3312 th = dt_trans_create(env, dev);
3314 GOTO(log, rc = PTR_ERR(th));
3316 rc = dt_declare_xattr_set(env, obj,
3317 lfsck_buf_get_const(env, NULL, MAX_LINKEA_SIZE),
3318 XATTR_NAME_LINK, 0, th);
3322 rc = dt_trans_start_local(env, dev, th);
3326 dt_write_lock(env, obj, 0);
3327 rc = lfsck_links_read(env, obj, ldata);
3331 if (unlikely(!lfsck_namespace_linkea_stale_overflow(ldata, ns)))
3332 GOTO(unlock, rc = 0);
3334 ldata->ld_leh->leh_overflow_time = 0;
3335 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
3336 GOTO(unlock, rc = 1);
3338 /* If all known entries are in the linkEA, then the 'leh_reccount'
3339 * should NOT be zero. */
3340 LASSERT(ldata->ld_leh->leh_reccount > 0);
3342 lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
3343 ldata->ld_leh->leh_len);
3344 rc = dt_xattr_set(env, obj, &linkea_buf, XATTR_NAME_LINK, 0, th);
3345 if (unlikely(rc == -ENOSPC))
3353 dt_write_unlock(env, obj);
3356 dt_trans_stop(env, dev, th);
3359 lfsck_ibits_unlock(&lh, LCK_PW);
3360 CDEBUG(D_LFSCK, "%s: clear linkea overflow timestamp for the object "
3362 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
3368 * Verify the object's agent entry.
3370 * If the object claims to have agent entry but the linkEA does not contain
3371 * remote parent, then remove the agent entry. Otherwise, if the object has
3372 * no agent entry but its linkEA contains remote parent, then will generate
3373 * agent entry for it.
3375 * \param[in] env pointer to the thread context
3376 * \param[in] com pointer to the lfsck component
3377 * \param[in] obj pointer to the dt_object to be handled
3379 * \retval positive number for repaired cases
3380 * \retval 0 if nothing to be repaired
3381 * \retval negative error number on failure
3383 static int lfsck_namespace_check_agent_entry(const struct lu_env *env,
3384 struct lfsck_component *com,
3385 struct dt_object *obj)
3387 struct linkea_data ldata = { NULL };
3388 struct lfsck_thread_info *info = lfsck_env_info(env);
3389 struct lfsck_namespace *ns = com->lc_file_ram;
3390 struct lfsck_instance *lfsck = com->lc_lfsck;
3391 struct lu_fid *pfid = &info->lti_fid2;
3392 struct lu_name *cname = &info->lti_name;
3393 struct lu_seq_range *range = &info->lti_range;
3394 struct seq_server_site *ss = lfsck_dev_site(lfsck);
3395 __u32 idx = lfsck_dev_idx(lfsck);
3397 bool remote = false;
3400 if (!(lfsck->li_bookmark_ram.lb_param & LPF_ALL_TGT))
3403 rc = lfsck_links_read_with_rec(env, obj, &ldata);
3404 if (rc == -ENOENT || rc == -ENODATA)
3407 if (rc && rc != -EINVAL)
3410 /* We check the agent entry again after verifying the linkEA
3411 * successfully. So invalid linkEA should be dryrun mode. */
3412 if (rc == -EINVAL || unlikely(!ldata.ld_leh->leh_reccount))
3415 linkea_first_entry(&ldata);
3416 while (ldata.ld_lee != NULL && !remote) {
3417 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
3419 if (!linkea_entry_is_valid(&ldata, cname, pfid))
3422 fld_range_set_mdt(range);
3423 rc = fld_server_lookup(env, ss->ss_server_fld,
3424 fid_seq(pfid), range);
3426 GOTO(out, rc = (rc == -ENOENT ? 0 : rc));
3428 if (range->lsr_index != idx)
3431 linkea_next_entry(&ldata);
3434 if ((lu_object_has_agent_entry(&obj->do_lu) && !remote) ||
3435 (!lu_object_has_agent_entry(&obj->do_lu) && remote)) {
3436 struct dt_device *dev = lfsck_obj2dev(obj);
3437 struct linkea_data ldata2 = { NULL };
3438 struct lustre_handle lh = { 0 };
3439 struct lu_buf linkea_buf;
3440 struct thandle *handle;
3442 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
3445 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
3446 MDS_INODELOCK_UPDATE |
3447 MDS_INODELOCK_XATTR, LCK_EX);
3451 handle = dt_trans_create(env, dev);
3453 GOTO(unlock, rc = PTR_ERR(handle));
3455 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
3456 ldata.ld_leh->leh_len);
3457 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
3458 XATTR_NAME_LINK, LU_XATTR_REPLACE, handle);
3462 rc = dt_trans_start_local(env, dev, handle);
3466 dt_write_lock(env, obj, 0);
3467 rc = lfsck_links_read2_with_rec(env, obj, &ldata2);
3469 if (rc == -ENOENT || rc == -ENODATA)
3474 /* If someone changed linkEA by race, then the agent
3475 * entry will be updated by lower layer automatically. */
3476 if (ldata.ld_leh->leh_len != ldata2.ld_leh->leh_len ||
3477 memcmp(ldata.ld_buf->lb_buf, ldata2.ld_buf->lb_buf,
3478 ldata.ld_leh->leh_len) != 0)
3479 GOTO(unlock2, rc = 0);
3481 rc = dt_xattr_set(env, obj, &linkea_buf, XATTR_NAME_LINK,
3482 LU_XATTR_REPLACE, handle);
3489 dt_write_unlock(env, obj);
3491 dt_trans_stop(env, dev, handle);
3493 lfsck_ibits_unlock(&lh, LCK_EX);
3500 ns->ln_agent_entries_repaired++;
3502 CDEBUG(D_LFSCK, "%s: repair agent entry for "DFID": rc = %d\n",
3503 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
3508 * Double scan the MDT-object for namespace LFSCK.
3510 * If the MDT-object contains invalid or repeated linkEA entries, then drop
3511 * those entries from the linkEA; if the linkEA becomes empty or the object
3512 * has no linkEA, then it is an orphan and will be added into the directory
3513 * .lustre/lost+found/MDTxxxx/; if the remote parent is lost, then recreate
3514 * the remote parent; if the name entry corresponding to some linkEA entry
3515 * is lost, then add the name entry back to the namespace.
3517 * \param[in] env pointer to the thread context
3518 * \param[in] com pointer to the lfsck component
3519 * \param[in] child pointer to the dt_object to be handled
3520 * \param[in] flags some hints to indicate how the @child should be handled
3522 * \retval positive number for repaired cases
3523 * \retval 0 if nothing to be repaired
3524 * \retval negative error number on failure
3526 static int lfsck_namespace_double_scan_one(const struct lu_env *env,
3527 struct lfsck_component *com,
3528 struct dt_object *child, __u8 flags)
3530 struct lfsck_thread_info *info = lfsck_env_info(env);
3531 struct lu_attr *la = &info->lti_la;
3532 struct lu_name *cname = &info->lti_name;
3533 struct lu_fid *pfid = &info->lti_fid;
3534 struct lu_fid *cfid = &info->lti_fid2;
3535 struct lfsck_instance *lfsck = com->lc_lfsck;
3536 struct lfsck_namespace *ns = com->lc_file_ram;
3537 struct dt_object *parent = NULL;
3538 struct linkea_data ldata = { NULL };
3539 bool repaired = false;
3544 dt_read_lock(env, child, 0);
3545 if (unlikely(lfsck_is_dead_obj(child))) {
3546 dt_read_unlock(env, child);
3551 if (S_ISDIR(lfsck_object_type(child))) {
3552 dt_read_unlock(env, child);
3553 rc = lfsck_namespace_double_scan_dir(env, com, child, flags);
3554 if (!rc && flags & LNTF_CHECK_AGENT_ENTRY)
3555 rc = lfsck_namespace_check_agent_entry(env, com, child);
3560 rc = lfsck_links_read(env, child, &ldata);
3561 dt_read_unlock(env, child);
3563 if (rc == -EINVAL) {
3564 struct lustre_handle lh = { 0 };
3566 rc = lfsck_ibits_lock(env, com->lc_lfsck, child, &lh,
3567 MDS_INODELOCK_UPDATE |
3568 MDS_INODELOCK_XATTR, LCK_EX);
3570 rc = lfsck_namespace_links_remove(env, com, child);
3571 lfsck_ibits_unlock(&lh, LCK_EX);
3580 if (!(ns->ln_flags & LF_INCOMPLETE) &&
3581 unlikely(lfsck_namespace_linkea_stale_overflow(&ldata, ns))) {
3582 rc = lfsck_namespace_linkea_clear_overflow(env, com, &ldata,
3588 ns->ln_linkea_overflow_cleared++;
3591 linkea_first_entry(&ldata);
3592 while (ldata.ld_lee != NULL) {
3593 rc = lfsck_namespace_unpack_linkea_entry(&ldata, cname, pfid,
3595 sizeof(info->lti_key));
3596 /* Invalid PFID in the linkEA entry. */
3598 rc = lfsck_namespace_shrink_linkea(env, com, child,
3599 &ldata, cname, pfid, true);
3609 rc = lfsck_namespace_filter_linkea_entry(&ldata, cname, pfid,
3611 /* Found repeated linkEA entries */
3613 rc = lfsck_namespace_shrink_linkea(env, com, child,
3614 &ldata, cname, pfid, false);
3626 parent = lfsck_object_find_bottom(env, lfsck, pfid);
3628 GOTO(out, rc = PTR_ERR(parent));
3630 if (!dt_object_exists(parent)) {
3633 if (ldata.ld_leh->leh_reccount > 1) {
3634 /* If it is NOT the last linkEA entry, then
3635 * there is still other chance to make the
3636 * child to be visible via other parent, then
3637 * remove this linkEA entry. */
3638 rc = lfsck_namespace_shrink_linkea(env, com,
3639 child, &ldata, cname, pfid, true);
3641 /* If the LFSCK is marked as LF_INCOMPLETE,
3642 * then means some MDT has ever tried to
3643 * verify some remote MDT-object that resides
3644 * on this MDT, but this MDT failed to respond
3645 * such request. So means there may be some
3646 * remote name entry on other MDT that
3647 * references this object with another name,
3648 * so we cannot know whether this linkEA is
3649 * valid or not. So keep it there and maybe
3650 * resolved when next LFSCK run. */
3651 if (ns->ln_flags & LF_INCOMPLETE) {
3652 lfsck_object_put(env, parent);
3657 /* Create the lost parent as an orphan. */
3658 rc = lfsck_namespace_create_orphan_dir(env, com,
3661 lfsck_object_put(env, parent);
3669 /* Add the missing name entry to the parent. */
3670 rc = lfsck_namespace_insert_normal(env, com,
3671 parent, child, cname->ln_name);
3672 if (unlikely(rc == -EEXIST))
3673 /* Unfortunately, someone reused the
3674 * name under the parent by race. So we
3675 * have to remove the linkEA entry from
3676 * current child object. It means that
3677 * the LFSCK cannot recover the system
3678 * totally back to its original status,
3679 * but it is necessary to make the
3680 * current system to be consistent. */
3681 rc = lfsck_namespace_shrink_linkea(env,
3685 linkea_next_entry(&ldata);
3688 lfsck_object_put(env, parent);
3696 } /* !dt_object_exists(parent) */
3698 /* The linkEA entry with bad parent will be removed. */
3699 if (unlikely(!dt_try_as_dir(env, parent))) {
3700 lfsck_object_put(env, parent);
3701 rc = lfsck_namespace_shrink_linkea(env, com, child,
3702 &ldata, cname, pfid, true);
3712 rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
3713 (const struct dt_key *)cname->ln_name);
3714 if (rc != 0 && rc != -ENOENT) {
3715 lfsck_object_put(env, parent);
3721 if (lu_fid_eq(cfid, lfsck_dto2fid(child))) {
3722 /* It is the most common case that we
3723 * find the name entry corresponding
3724 * to the linkEA entry. */
3725 lfsck_object_put(env, parent);
3726 linkea_next_entry(&ldata);
3728 /* The name entry references another
3729 * MDT-object that may be created by
3730 * the LFSCK for repairing dangling
3731 * name entry. Try to replace it. */
3732 rc = lfsck_namespace_replace_cond(env, com,
3733 parent, child, cfid, cname);
3734 lfsck_object_put(env, parent);
3740 linkea_next_entry(&ldata);
3742 rc = lfsck_namespace_shrink_linkea(env,
3756 /* The following handles -ENOENT case */
3758 rc = dt_attr_get(env, child, la);
3762 /* If there is no name entry in the parent dir and the object
3763 * link count is fewer than the linkea entries count, then the
3764 * linkea entry should be removed. */
3765 if (ldata.ld_leh->leh_reccount > la->la_nlink) {
3766 rc = lfsck_namespace_shrink_linkea_cond(env, com,
3767 parent, child, &ldata, cname, pfid);
3768 lfsck_object_put(env, parent);
3778 /* If the LFSCK is marked as LF_INCOMPLETE, then means some
3779 * MDT has ever tried to verify some remote MDT-object that
3780 * resides on this MDT, but this MDT failed to respond such
3781 * request. So means there may be some remote name entry on
3782 * other MDT that references this object with another name,
3783 * so we cannot know whether this linkEA is valid or not.
3784 * So keep it there and maybe resolved when next LFSCK run. */
3785 if (ns->ln_flags & LF_INCOMPLETE) {
3786 lfsck_object_put(env, parent);
3791 rc = lfsck_namespace_check_name(env, lfsck, parent, child,
3797 lfsck_object_put(env, parent);
3802 /* It is an invalid name entry, drop it. */
3803 if (unlikely(rc > 0)) {
3804 lfsck_object_put(env, parent);
3805 rc = lfsck_namespace_shrink_linkea(env, com, child,
3806 &ldata, cname, pfid, true);
3816 /* Add the missing name entry back to the namespace. */
3817 rc = lfsck_namespace_insert_normal(env, com, parent, child,
3819 if (unlikely(rc == -ESTALE))
3820 /* It may happen when the remote object has been
3821 * removed, but the local MDT is not aware of that. */
3824 if (unlikely(rc == -EEXIST))
3825 /* Unfortunately, someone reused the name under the
3826 * parent by race. So we have to remove the linkEA
3827 * entry from current child object. It means that the
3828 * LFSCK cannot recover the system totally back to
3829 * its original status, but it is necessary to make
3830 * the current system to be consistent.
3832 * It also may be because of the LFSCK found some
3833 * internal status of create operation. Under such
3834 * case, nothing to be done. */
3835 rc = lfsck_namespace_shrink_linkea_cond(env, com,
3836 parent, child, &ldata, cname, pfid);
3838 linkea_next_entry(&ldata);
3840 lfsck_object_put(env, parent);
3851 if (rc < 0 && rc != -ENODATA)
3854 if (rc == 0 && ldata.ld_leh != NULL)
3855 count = ldata.ld_leh->leh_reccount;
3858 /* If the LFSCK is marked as LF_INCOMPLETE, then means some
3859 * MDT has ever tried to verify some remote MDT-object that
3860 * resides on this MDT, but this MDT failed to respond such
3861 * request. So means there may be some remote name entry on
3862 * other MDT that references this object with another name,
3863 * so we cannot know whether this linkEA is valid or not.
3864 * So keep it there and maybe resolved when next LFSCK run. */
3865 if (!(ns->ln_flags & LF_INCOMPLETE) &&
3866 (ldata.ld_leh == NULL ||
3867 !ldata.ld_leh->leh_overflow_time)) {
3868 /* If the child becomes orphan, then insert it into
3869 * the global .lustre/lost+found/MDTxxxx directory. */
3870 rc = lfsck_namespace_insert_orphan(env, com, child,
3876 ns->ln_mul_ref_repaired++;
3881 rc = dt_attr_get(env, child, la);
3885 if (la->la_nlink != 0 && la->la_nlink != count) {
3886 if (unlikely(!S_ISREG(lfsck_object_type(child)) &&
3887 !S_ISLNK(lfsck_object_type(child)))) {
3888 CDEBUG(D_LFSCK, "%s: namespace LFSCK finds "
3889 "the object "DFID"'s nlink count %d "
3890 "does not match linkEA count %d, "
3891 "type %o, skip it.\n",
3892 lfsck_lfsck2name(lfsck),
3893 PFID(lfsck_dto2fid(child)),
3894 la->la_nlink, count,
3895 lfsck_object_type(child));
3896 } else if (la->la_nlink < count &&
3897 likely(!ldata.ld_leh->leh_overflow_time)) {
3898 rc = lfsck_namespace_repair_nlink(env, com,
3901 ns->ln_objs_nlink_repaired++;
3909 if (la->la_nlink > 1)
3910 ns->ln_mul_linked_repaired++;
3916 if (!rc && flags & LNTF_CHECK_AGENT_ENTRY)
3917 rc = lfsck_namespace_check_agent_entry(env, com, child);
3922 static void lfsck_namespace_dump_statistics(struct seq_file *m,
3923 struct lfsck_namespace *ns,
3924 __u64 checked_phase1,
3925 __u64 checked_phase2,
3926 time64_t time_phase1,
3927 time64_t time_phase2, bool dryrun)
3929 const char *postfix = dryrun ? "inconsistent" : "repaired";
3931 seq_printf(m, "checked_phase1: %llu\n"
3932 "checked_phase2: %llu\n"
3935 "failed_phase1: %llu\n"
3936 "failed_phase2: %llu\n"
3937 "directories: %llu\n"
3941 "multiple_linked_checked: %llu\n"
3942 "multiple_linked_%s: %llu\n"
3943 "unknown_inconsistency: %llu\n"
3944 "unmatched_pairs_%s: %llu\n"
3945 "dangling_%s: %llu\n"
3946 "multiple_referenced_%s: %llu\n"
3947 "bad_file_type_%s: %llu\n"
3948 "lost_dirent_%s: %llu\n"
3949 "local_lost_found_scanned: %llu\n"
3950 "local_lost_found_moved: %llu\n"
3951 "local_lost_found_skipped: %llu\n"
3952 "local_lost_found_failed: %llu\n"
3953 "striped_dirs_scanned: %llu\n"
3954 "striped_dirs_%s: %llu\n"
3955 "striped_dirs_failed: %llu\n"
3956 "striped_dirs_disabled: %llu\n"
3957 "striped_dirs_skipped: %llu\n"
3958 "striped_shards_scanned: %llu\n"
3959 "striped_shards_%s: %llu\n"
3960 "striped_shards_failed: %llu\n"
3961 "striped_shards_skipped: %llu\n"
3962 "name_hash_%s: %llu\n"
3963 "linkea_overflow_%s: %llu\n"
3964 "agent_entries_%s: %llu\n"
3965 "success_count: %u\n"
3966 "run_time_phase1: %lld seconds\n"
3967 "run_time_phase2: %lld seconds\n",
3970 dryrun ? "inconsistent" : "updated",
3971 ns->ln_items_repaired,
3972 dryrun ? "inconsistent" : "updated",
3973 ns->ln_objs_repaired_phase2,
3974 ns->ln_items_failed,
3975 ns->ln_objs_failed_phase2,
3976 ns->ln_dirs_checked,
3977 postfix, ns->ln_dirent_repaired,
3978 postfix, ns->ln_linkea_repaired,
3979 postfix, ns->ln_objs_nlink_repaired,
3980 ns->ln_mul_linked_checked,
3981 postfix, ns->ln_mul_linked_repaired,
3982 ns->ln_unknown_inconsistency,
3983 postfix, ns->ln_unmatched_pairs_repaired,
3984 postfix, ns->ln_dangling_repaired,
3985 postfix, ns->ln_mul_ref_repaired,
3986 postfix, ns->ln_bad_type_repaired,
3987 postfix, ns->ln_lost_dirent_repaired,
3988 ns->ln_local_lpf_scanned,
3989 ns->ln_local_lpf_moved,
3990 ns->ln_local_lpf_skipped,
3991 ns->ln_local_lpf_failed,
3992 ns->ln_striped_dirs_scanned,
3993 postfix, ns->ln_striped_dirs_repaired,
3994 ns->ln_striped_dirs_failed,
3995 ns->ln_striped_dirs_disabled,
3996 ns->ln_striped_dirs_skipped,
3997 ns->ln_striped_shards_scanned,
3998 postfix, ns->ln_striped_shards_repaired,
3999 ns->ln_striped_shards_failed,
4000 ns->ln_striped_shards_skipped,
4001 postfix, ns->ln_name_hash_repaired,
4002 dryrun ? "inconsistent" : "cleared",
4003 ns->ln_linkea_overflow_cleared,
4004 postfix, ns->ln_agent_entries_repaired,
4005 ns->ln_success_count,
4010 static void lfsck_namespace_release_lmv(const struct lu_env *env,
4011 struct lfsck_component *com)
4013 struct lfsck_instance *lfsck = com->lc_lfsck;
4014 struct lfsck_namespace *ns = com->lc_file_ram;
4016 while (!list_empty(&lfsck->li_list_lmv)) {
4017 struct lfsck_lmv_unit *llu;
4018 struct lfsck_lmv *llmv;
4020 llu = list_entry(lfsck->li_list_lmv.next,
4021 struct lfsck_lmv_unit, llu_link);
4022 llmv = &llu->llu_lmv;
4024 LASSERTF(atomic_read(&llmv->ll_ref) == 1,
4025 "still in using: %u\n",
4026 atomic_read(&llmv->ll_ref));
4028 ns->ln_striped_dirs_skipped++;
4029 lfsck_lmv_put(env, llmv);
4033 static int lfsck_namespace_check_for_double_scan(const struct lu_env *env,
4034 struct lfsck_component *com,
4035 struct dt_object *obj)
4037 struct lu_attr *la = &lfsck_env_info(env)->lti_la;
4040 rc = dt_attr_get(env, obj, la);
4044 /* zero-linkEA object may be orphan, but it also maybe because
4045 * of upgrading. Currently, we cannot record it for double scan.
4046 * Because it may cause the LFSCK trace file to be too large. */
4048 /* "la_ctime" == 1 means that it has ever been removed from
4049 * backend /lost+found directory but not been added back to
4050 * the normal namespace yet. */
4052 if ((S_ISREG(lfsck_object_type(obj)) && la->la_nlink > 1) ||
4053 unlikely(la->la_ctime == 1))
4054 rc = lfsck_namespace_trace_update(env, com, lfsck_dto2fid(obj),
4055 LNTF_CHECK_LINKEA, true);
4060 /* namespace APIs */
4062 static int lfsck_namespace_reset(const struct lu_env *env,
4063 struct lfsck_component *com, bool init)
4065 struct lfsck_instance *lfsck = com->lc_lfsck;
4066 struct lfsck_namespace *ns = com->lc_file_ram;
4067 struct lfsck_assistant_data *lad = com->lc_data;
4068 struct dt_object *root;
4072 root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
4074 GOTO(log, rc = PTR_ERR(root));
4076 if (unlikely(!dt_try_as_dir(env, root)))
4077 GOTO(put, rc = -ENOTDIR);
4079 down_write(&com->lc_sem);
4081 memset(ns, 0, sizeof(*ns));
4083 __u32 count = ns->ln_success_count;
4084 time64_t last_time = ns->ln_time_last_complete;
4086 memset(ns, 0, sizeof(*ns));
4087 ns->ln_success_count = count;
4088 ns->ln_time_last_complete = last_time;
4090 ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
4091 ns->ln_status = LS_INIT;
4092 ns->ln_time_latest_reset = ktime_get_real_seconds();
4094 rc = lfsck_load_one_trace_file(env, com, root, &com->lc_obj,
4095 &dt_lfsck_namespace_features,
4096 LFSCK_NAMESPACE, true);
4100 rc = lfsck_load_sub_trace_files(env, com, &dt_lfsck_namespace_features,
4101 LFSCK_NAMESPACE, true);
4105 clear_bit(LAD_INCOMPLETE, &lad->lad_flags);
4106 CFS_RESET_BITMAP(lad->lad_bitmap);
4108 rc = lfsck_namespace_store(env, com);
4113 up_write(&com->lc_sem);
4116 lfsck_object_put(env, root);
4118 CDEBUG(D_LFSCK, "%s: namespace LFSCK reset: rc = %d\n",
4119 lfsck_lfsck2name(lfsck), rc);
4124 lfsck_namespace_fail(const struct lu_env *env, struct lfsck_component *com,
4127 struct lfsck_namespace *ns = com->lc_file_ram;
4129 down_write(&com->lc_sem);
4131 com->lc_new_checked++;
4132 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
4133 up_write(&com->lc_sem);
4136 static void lfsck_namespace_close_dir(const struct lu_env *env,
4137 struct lfsck_component *com)
4139 struct lfsck_namespace *ns = com->lc_file_ram;
4140 struct lfsck_assistant_data *lad = com->lc_data;
4141 struct lfsck_assistant_object *lso = NULL;
4142 struct lfsck_instance *lfsck = com->lc_lfsck;
4143 struct lfsck_lmv *llmv = lfsck->li_lmv;
4144 struct lfsck_namespace_req *lnr;
4145 struct lu_attr *la = &lfsck_env_info(env)->lti_la2;
4146 __u32 size = sizeof(*lnr) + LFSCK_TMPBUF_LEN;
4148 bool wakeup = false;
4154 rc = dt_attr_get(env, lfsck->li_obj_dir, la);
4158 OBD_ALLOC(lnr, size);
4160 ns->ln_striped_dirs_skipped++;
4165 lso = lfsck_assistant_object_init(env, lfsck_dto2fid(lfsck->li_obj_dir),
4166 la, lfsck->li_pos_current.lp_oit_cookie, true);
4168 OBD_FREE(lnr, size);
4169 ns->ln_striped_dirs_skipped++;
4174 /* Generate a dummy request to indicate that all shards' name entry
4175 * in this striped directory has been scanned for the first time. */
4176 INIT_LIST_HEAD(&lnr->lnr_lar.lar_list);
4177 lnr->lnr_lar.lar_parent = lso;
4178 lnr->lnr_lmv = lfsck_lmv_get(llmv);
4179 lnr->lnr_fid = *lfsck_dto2fid(lfsck->li_obj_dir);
4180 lnr->lnr_dir_cookie = MDS_DIR_END_OFF;
4181 lnr->lnr_size = size;
4182 lnr->lnr_type = lso->lso_attr.la_mode;
4184 spin_lock(&lad->lad_lock);
4185 if (lad->lad_assistant_status < 0 ||
4186 unlikely(!thread_is_running(&lfsck->li_thread) ||
4187 !thread_is_running(&lad->lad_thread))) {
4188 spin_unlock(&lad->lad_lock);
4189 lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar);
4190 ns->ln_striped_dirs_skipped++;
4195 list_add_tail(&lnr->lnr_lar.lar_list, &lad->lad_req_list);
4196 if (lad->lad_prefetched == 0)
4199 lad->lad_prefetched++;
4200 spin_unlock(&lad->lad_lock);
4202 wake_up_all(&lad->lad_thread.t_ctl_waitq);
4207 static int lfsck_namespace_open_dir(const struct lu_env *env,
4208 struct lfsck_component *com)
4210 struct lfsck_instance *lfsck = com->lc_lfsck;
4211 struct lfsck_namespace *ns = com->lc_file_ram;
4212 struct lfsck_lmv *llmv = lfsck->li_lmv;
4219 if (llmv->ll_lmv_master) {
4220 struct lmv_mds_md_v1 *lmv = &llmv->ll_lmv;
4222 if (lmv->lmv_master_mdt_index != lfsck_dev_idx(lfsck)) {
4223 lmv->lmv_master_mdt_index =
4224 lfsck_dev_idx(lfsck);
4225 ns->ln_flags |= LF_INCONSISTENT;
4226 llmv->ll_lmv_updated = 1;
4229 rc = lfsck_namespace_verify_stripe_slave(env, com,
4230 lfsck->li_obj_dir, llmv);
4233 RETURN(rc > 0 ? 0 : rc);
4236 static int lfsck_namespace_checkpoint(const struct lu_env *env,
4237 struct lfsck_component *com, bool init)
4239 struct lfsck_instance *lfsck = com->lc_lfsck;
4240 struct lfsck_namespace *ns = com->lc_file_ram;
4244 rc = lfsck_checkpoint_generic(env, com);
4249 down_write(&com->lc_sem);
4251 ns->ln_pos_latest_start = lfsck->li_pos_checkpoint;
4253 ns->ln_pos_last_checkpoint = lfsck->li_pos_checkpoint;
4254 ns->ln_run_time_phase1 += ktime_get_seconds() -
4255 lfsck->li_time_last_checkpoint;
4256 ns->ln_time_last_checkpoint = ktime_get_real_seconds();
4257 ns->ln_items_checked += com->lc_new_checked;
4258 com->lc_new_checked = 0;
4261 rc = lfsck_namespace_store(env, com);
4262 up_write(&com->lc_sem);
4265 CDEBUG(D_LFSCK, "%s: namespace LFSCK checkpoint at the pos [%llu"
4266 ", "DFID", %#llx], status = %d: rc = %d\n",
4267 lfsck_lfsck2name(lfsck), lfsck->li_pos_current.lp_oit_cookie,
4268 PFID(&lfsck->li_pos_current.lp_dir_parent),
4269 lfsck->li_pos_current.lp_dir_cookie, ns->ln_status, rc);
4271 return rc > 0 ? 0 : rc;
4274 static int lfsck_namespace_prep(const struct lu_env *env,
4275 struct lfsck_component *com,
4276 struct lfsck_start_param *lsp)
4278 struct lfsck_instance *lfsck = com->lc_lfsck;
4279 struct lfsck_namespace *ns = com->lc_file_ram;
4280 struct lfsck_position *pos = &com->lc_pos_start;
4283 rc = lfsck_namespace_load_bitmap(env, com);
4284 if (rc != 0 || ns->ln_status == LS_COMPLETED) {
4285 rc = lfsck_namespace_reset(env, com, false);
4287 rc = lfsck_set_param(env, lfsck, lsp->lsp_start, true);
4290 CDEBUG(D_LFSCK, "%s: namespace LFSCK prep failed: "
4291 "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
4297 down_write(&com->lc_sem);
4298 ns->ln_time_latest_start = ktime_get_real_seconds();
4299 spin_lock(&lfsck->li_lock);
4301 if (ns->ln_flags & LF_SCANNED_ONCE) {
4302 if (!lfsck->li_drop_dryrun ||
4303 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent)) {
4304 ns->ln_status = LS_SCANNING_PHASE2;
4305 list_move_tail(&com->lc_link,
4306 &lfsck->li_list_double_scan);
4307 if (!list_empty(&com->lc_link_dir))
4308 list_del_init(&com->lc_link_dir);
4309 lfsck_pos_set_zero(pos);
4311 ns->ln_status = LS_SCANNING_PHASE1;
4312 ns->ln_run_time_phase1 = 0;
4313 ns->ln_run_time_phase2 = 0;
4314 ns->ln_items_checked = 0;
4315 ns->ln_items_repaired = 0;
4316 ns->ln_items_failed = 0;
4317 ns->ln_dirs_checked = 0;
4318 ns->ln_objs_checked_phase2 = 0;
4319 ns->ln_objs_repaired_phase2 = 0;
4320 ns->ln_objs_failed_phase2 = 0;
4321 ns->ln_objs_nlink_repaired = 0;
4322 ns->ln_dirent_repaired = 0;
4323 ns->ln_linkea_repaired = 0;
4324 ns->ln_mul_linked_checked = 0;
4325 ns->ln_mul_linked_repaired = 0;
4326 ns->ln_unknown_inconsistency = 0;
4327 ns->ln_unmatched_pairs_repaired = 0;
4328 ns->ln_dangling_repaired = 0;
4329 ns->ln_mul_ref_repaired = 0;
4330 ns->ln_bad_type_repaired = 0;
4331 ns->ln_lost_dirent_repaired = 0;
4332 ns->ln_striped_dirs_scanned = 0;
4333 ns->ln_striped_dirs_repaired = 0;
4334 ns->ln_striped_dirs_failed = 0;
4335 ns->ln_striped_dirs_disabled = 0;
4336 ns->ln_striped_dirs_skipped = 0;
4337 ns->ln_striped_shards_scanned = 0;
4338 ns->ln_striped_shards_repaired = 0;
4339 ns->ln_striped_shards_failed = 0;
4340 ns->ln_striped_shards_skipped = 0;
4341 ns->ln_name_hash_repaired = 0;
4342 fid_zero(&ns->ln_fid_latest_scanned_phase2);
4343 if (list_empty(&com->lc_link_dir))
4344 list_add_tail(&com->lc_link_dir,
4345 &lfsck->li_list_dir);
4346 *pos = ns->ln_pos_first_inconsistent;
4349 ns->ln_status = LS_SCANNING_PHASE1;
4350 if (list_empty(&com->lc_link_dir))
4351 list_add_tail(&com->lc_link_dir,
4352 &lfsck->li_list_dir);
4353 if (!lfsck->li_drop_dryrun ||
4354 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent)) {
4355 *pos = ns->ln_pos_last_checkpoint;
4356 pos->lp_oit_cookie++;
4358 *pos = ns->ln_pos_first_inconsistent;
4362 spin_unlock(&lfsck->li_lock);
4363 up_write(&com->lc_sem);
4365 rc = lfsck_start_assistant(env, com, lsp);
4367 CDEBUG(D_LFSCK, "%s: namespace LFSCK prep done, start pos [%llu, "
4368 DFID", %#llx]: rc = %d\n",
4369 lfsck_lfsck2name(lfsck), pos->lp_oit_cookie,
4370 PFID(&pos->lp_dir_parent), pos->lp_dir_cookie, rc);
4375 static int lfsck_namespace_exec_oit(const struct lu_env *env,
4376 struct lfsck_component *com,
4377 struct dt_object *obj)
4379 struct lfsck_thread_info *info = lfsck_env_info(env);
4380 struct lfsck_namespace *ns = com->lc_file_ram;
4381 struct lfsck_instance *lfsck = com->lc_lfsck;
4382 const struct lu_fid *fid = lfsck_dto2fid(obj);
4383 struct lu_fid *pfid = &info->lti_fid2;
4384 struct lu_name *cname = &info->lti_name;
4385 struct lu_seq_range *range = &info->lti_range;
4386 struct seq_server_site *ss = lfsck_dev_site(lfsck);
4387 struct linkea_data ldata = { NULL };
4388 __u32 idx = lfsck_dev_idx(lfsck);
4390 bool remote = false;
4393 rc = lfsck_links_read(env, obj, &ldata);
4397 /* -EINVAL means crashed linkEA, should be verified. */
4398 if (rc == -EINVAL) {
4399 rc = lfsck_namespace_trace_update(env, com, fid,
4400 LNTF_CHECK_LINKEA, true);
4402 struct lustre_handle lh = { 0 };
4404 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
4405 MDS_INODELOCK_UPDATE |
4406 MDS_INODELOCK_XATTR, LCK_EX);
4408 rc = lfsck_namespace_links_remove(env, com,
4410 lfsck_ibits_unlock(&lh, LCK_EX);
4414 GOTO(out, rc = (rc == -ENOENT ? 0 : rc));
4417 if (rc && rc != -ENODATA)
4420 if (rc == -ENODATA || unlikely(!ldata.ld_leh->leh_reccount)) {
4421 rc = lfsck_namespace_check_for_double_scan(env, com, obj);
4426 linkea_first_entry(&ldata);
4427 while (ldata.ld_lee != NULL) {
4428 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
4430 if (!fid_is_sane(pfid)) {
4431 rc = lfsck_namespace_trace_update(env, com, fid,
4432 LNTF_CHECK_PARENT, true);
4433 } else if (!linkea_entry_is_valid(&ldata, cname, pfid)) {
4436 fld_range_set_mdt(range);
4437 rc = fld_server_lookup(env, ss->ss_server_fld,
4438 fid_seq(pfid), range);
4439 if ((rc == -ENOENT) ||
4440 (!rc && range->lsr_index != idx)) {
4448 linkea_next_entry(&ldata);
4451 if ((lu_object_has_agent_entry(&obj->do_lu) && !remote) ||
4452 (!lu_object_has_agent_entry(&obj->do_lu) && remote)) {
4453 rc = lfsck_namespace_trace_update(env, com, fid,
4454 LNTF_CHECK_AGENT_ENTRY, true);
4459 /* Record multiple-linked object. */
4460 if (ldata.ld_leh->leh_reccount > 1) {
4461 rc = lfsck_namespace_trace_update(env, com, fid,
4462 LNTF_CHECK_LINKEA, true);
4468 rc = lfsck_namespace_trace_update(env, com, fid,
4469 LNTF_CHECK_LINKEA, true);
4471 rc = lfsck_namespace_check_for_double_scan(env, com, obj);
4476 down_write(&com->lc_sem);
4477 if (S_ISDIR(lfsck_object_type(obj)))
4478 ns->ln_dirs_checked++;
4480 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
4481 up_write(&com->lc_sem);
4486 static int lfsck_namespace_exec_dir(const struct lu_env *env,
4487 struct lfsck_component *com,
4488 struct lfsck_assistant_object *lso,
4489 struct lu_dirent *ent, __u16 type)
4491 struct lfsck_assistant_data *lad = com->lc_data;
4492 struct lfsck_instance *lfsck = com->lc_lfsck;
4493 struct lfsck_namespace_req *lnr;
4494 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4495 struct ptlrpc_thread *mthread = &lfsck->li_thread;
4496 struct ptlrpc_thread *athread = &lad->lad_thread;
4497 struct l_wait_info lwi = { 0 };
4498 bool wakeup = false;
4500 l_wait_event(mthread->t_ctl_waitq,
4501 lad->lad_prefetched < bk->lb_async_windows ||
4502 !thread_is_running(mthread) ||
4503 !thread_is_running(athread),
4506 if (unlikely(!thread_is_running(mthread) ||
4507 !thread_is_running(athread)))
4510 if (unlikely(lfsck_is_dead_obj(lfsck->li_obj_dir)))
4513 lnr = lfsck_namespace_assistant_req_init(com->lc_lfsck, lso, ent, type);
4515 struct lfsck_namespace *ns = com->lc_file_ram;
4517 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
4518 return PTR_ERR(lnr);
4521 spin_lock(&lad->lad_lock);
4522 if (lad->lad_assistant_status < 0 ||
4523 unlikely(!thread_is_running(mthread) ||
4524 !thread_is_running(athread))) {
4525 spin_unlock(&lad->lad_lock);
4526 lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar);
4527 return lad->lad_assistant_status;
4530 list_add_tail(&lnr->lnr_lar.lar_list, &lad->lad_req_list);
4531 if (lad->lad_prefetched == 0)
4534 lad->lad_prefetched++;
4535 spin_unlock(&lad->lad_lock);
4537 wake_up_all(&lad->lad_thread.t_ctl_waitq);
4539 down_write(&com->lc_sem);
4540 com->lc_new_checked++;
4541 up_write(&com->lc_sem);
4546 static int lfsck_namespace_post(const struct lu_env *env,
4547 struct lfsck_component *com,
4548 int result, bool init)
4550 struct lfsck_instance *lfsck = com->lc_lfsck;
4551 struct lfsck_namespace *ns = com->lc_file_ram;
4555 lfsck_post_generic(env, com, &result);
4557 down_write(&com->lc_sem);
4558 lfsck_namespace_release_lmv(env, com);
4560 spin_lock(&lfsck->li_lock);
4562 ns->ln_pos_last_checkpoint = lfsck->li_pos_checkpoint;
4564 ns->ln_status = LS_SCANNING_PHASE2;
4565 ns->ln_flags |= LF_SCANNED_ONCE;
4566 ns->ln_flags &= ~LF_UPGRADE;
4567 list_del_init(&com->lc_link_dir);
4568 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
4569 } else if (result == 0) {
4570 if (lfsck->li_status != 0)
4571 ns->ln_status = lfsck->li_status;
4573 ns->ln_status = LS_STOPPED;
4574 if (ns->ln_status != LS_PAUSED) {
4575 list_del_init(&com->lc_link_dir);
4576 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4579 ns->ln_status = LS_FAILED;
4580 list_del_init(&com->lc_link_dir);
4581 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4583 spin_unlock(&lfsck->li_lock);
4586 ns->ln_run_time_phase1 += ktime_get_seconds() -
4587 lfsck->li_time_last_checkpoint;
4588 ns->ln_time_last_checkpoint = ktime_get_real_seconds();
4589 ns->ln_items_checked += com->lc_new_checked;
4590 com->lc_new_checked = 0;
4593 rc = lfsck_namespace_store(env, com);
4594 up_write(&com->lc_sem);
4596 CDEBUG(D_LFSCK, "%s: namespace LFSCK post done: rc = %d\n",
4597 lfsck_lfsck2name(lfsck), rc);
4603 lfsck_namespace_dump(const struct lu_env *env, struct lfsck_component *com,
4606 struct lfsck_instance *lfsck = com->lc_lfsck;
4607 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4608 struct lfsck_namespace *ns = com->lc_file_ram;
4610 down_read(&com->lc_sem);
4611 seq_printf(m, "name: lfsck_namespace\n"
4617 lfsck_status2name(ns->ln_status));
4619 lfsck_bits_dump(m, ns->ln_flags, lfsck_flags_names, "flags");
4621 lfsck_bits_dump(m, bk->lb_param, lfsck_param_names, "param");
4623 lfsck_time_dump(m, ns->ln_time_last_complete, "last_completed");
4625 lfsck_time_dump(m, ns->ln_time_latest_start, "latest_start");
4627 lfsck_time_dump(m, ns->ln_time_last_checkpoint, "last_checkpoint");
4629 lfsck_pos_dump(m, &ns->ln_pos_latest_start, "latest_start_position");
4631 lfsck_pos_dump(m, &ns->ln_pos_last_checkpoint,
4632 "last_checkpoint_position");
4634 lfsck_pos_dump(m, &ns->ln_pos_first_inconsistent,
4635 "first_failure_position");
4637 if (ns->ln_status == LS_SCANNING_PHASE1) {
4638 struct lfsck_position pos;
4639 time64_t duration = ktime_get_seconds() -
4640 lfsck->li_time_last_checkpoint;
4641 u64 checked = ns->ln_items_checked + com->lc_new_checked;
4642 u64 speed = checked;
4643 u64 new_checked = com->lc_new_checked;
4644 time64_t rtime = ns->ln_run_time_phase1 + duration;
4647 new_checked = div64_s64(new_checked, duration);
4650 speed = div64_s64(speed, rtime);
4652 lfsck_namespace_dump_statistics(m, ns, checked, 0, rtime, 0,
4653 bk->lb_param & LPF_DRYRUN);
4654 seq_printf(m, "average_speed_phase1: %llu items/sec\n"
4655 "average_speed_phase2: N/A\n"
4656 "average_speed_total: %llu items/sec\n"
4657 "real_time_speed_phase1: %llu items/sec\n"
4658 "real_time_speed_phase2: N/A\n",
4663 if (likely(lfsck->li_di_oit)) {
4664 const struct dt_it_ops *iops =
4665 &lfsck->li_obj_oit->do_index_ops->dio_it;
4667 /* The low layer otable-based iteration position may NOT
4668 * exactly match the namespace-based directory traversal
4669 * cookie. Generally, it is not a serious issue. But the
4670 * caller should NOT make assumption on that. */
4671 pos.lp_oit_cookie = iops->store(env, lfsck->li_di_oit);
4672 if (!lfsck->li_current_oit_processed)
4673 pos.lp_oit_cookie--;
4675 spin_lock(&lfsck->li_lock);
4676 if (lfsck->li_di_dir) {
4677 pos.lp_dir_cookie = lfsck->li_cookie_dir;
4678 if (pos.lp_dir_cookie >= MDS_DIR_END_OFF) {
4679 fid_zero(&pos.lp_dir_parent);
4680 pos.lp_dir_cookie = 0;
4683 *lfsck_dto2fid(lfsck->li_obj_dir);
4686 fid_zero(&pos.lp_dir_parent);
4687 pos.lp_dir_cookie = 0;
4689 spin_unlock(&lfsck->li_lock);
4691 pos = ns->ln_pos_last_checkpoint;
4694 lfsck_pos_dump(m, &pos, "current_position");
4695 } else if (ns->ln_status == LS_SCANNING_PHASE2) {
4696 time64_t duration = ktime_get_seconds() -
4697 com->lc_time_last_checkpoint;
4698 __u64 checked = ns->ln_objs_checked_phase2 +
4699 com->lc_new_checked;
4700 __u64 speed1 = ns->ln_items_checked;
4701 __u64 speed2 = checked;
4702 __u64 speed0 = speed1 + speed2;
4703 __u64 new_checked = com->lc_new_checked;
4704 time64_t rtime = ns->ln_run_time_phase2 + duration;
4705 time64_t time0 = ns->ln_run_time_phase1 + rtime;
4708 new_checked = div64_s64(new_checked, duration);
4710 if (ns->ln_run_time_phase1 != 0)
4711 speed1 = div64_s64(speed1, ns->ln_run_time_phase1);
4712 else if (ns->ln_items_checked != 0)
4716 speed2 = div64_s64(speed2, rtime);
4717 else if (checked != 0)
4721 speed0 = div64_s64(speed0, time0);
4723 lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked,
4725 ns->ln_run_time_phase1, rtime,
4726 bk->lb_param & LPF_DRYRUN);
4727 seq_printf(m, "average_speed_phase1: %llu items/sec\n"
4728 "average_speed_phase2: %llu objs/sec\n"
4729 "average_speed_total: %llu items/sec\n"
4730 "real_time_speed_phase1: N/A\n"
4731 "real_time_speed_phase2: %llu objs/sec\n"
4732 "current_position: "DFID"\n",
4737 PFID(&ns->ln_fid_latest_scanned_phase2));
4739 __u64 speed1 = ns->ln_items_checked;
4740 __u64 speed2 = ns->ln_objs_checked_phase2;
4741 __u64 speed0 = speed1 + speed2;
4742 time64_t time0 = ns->ln_run_time_phase1 + ns->ln_run_time_phase2;
4744 if (ns->ln_run_time_phase1 != 0)
4745 speed1 = div64_s64(speed1, ns->ln_run_time_phase1);
4746 else if (ns->ln_items_checked != 0)
4749 if (ns->ln_run_time_phase2 != 0)
4750 speed2 = div64_s64(speed2, ns->ln_run_time_phase2);
4751 else if (ns->ln_objs_checked_phase2 != 0)
4755 speed0 = div64_s64(speed0, time0);
4757 lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked,
4758 ns->ln_objs_checked_phase2,
4759 ns->ln_run_time_phase1,
4760 ns->ln_run_time_phase2,
4761 bk->lb_param & LPF_DRYRUN);
4762 seq_printf(m, "average_speed_phase1: %llu items/sec\n"
4763 "average_speed_phase2: %llu objs/sec\n"
4764 "average_speed_total: %llu items/sec\n"
4765 "real_time_speed_phase1: N/A\n"
4766 "real_time_speed_phase2: N/A\n"
4767 "current_position: N/A\n",
4773 up_read(&com->lc_sem);
4776 static int lfsck_namespace_double_scan(const struct lu_env *env,
4777 struct lfsck_component *com)
4779 struct lfsck_namespace *ns = com->lc_file_ram;
4780 struct lfsck_assistant_data *lad = com->lc_data;
4781 struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
4782 struct lfsck_tgt_desc *ltd;
4783 struct lfsck_tgt_desc *next;
4786 rc = lfsck_double_scan_generic(env, com, ns->ln_status);
4787 if (thread_is_stopped(&lad->lad_thread)) {
4788 LASSERT(list_empty(&lad->lad_req_list));
4789 LASSERT(list_empty(&lad->lad_mdt_phase1_list));
4791 spin_lock(<ds->ltd_lock);
4792 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4793 ltd_namespace_phase_list) {
4794 list_del_init(<d->ltd_namespace_phase_list);
4796 spin_unlock(<ds->ltd_lock);
4802 static void lfsck_namespace_data_release(const struct lu_env *env,
4803 struct lfsck_component *com)
4805 struct lfsck_assistant_data *lad = com->lc_data;
4806 struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
4807 struct lfsck_tgt_desc *ltd;
4808 struct lfsck_tgt_desc *next;
4810 LASSERT(lad != NULL);
4811 LASSERT(thread_is_init(&lad->lad_thread) ||
4812 thread_is_stopped(&lad->lad_thread));
4813 LASSERT(list_empty(&lad->lad_req_list));
4815 com->lc_data = NULL;
4816 lfsck_namespace_release_lmv(env, com);
4818 spin_lock(<ds->ltd_lock);
4819 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
4820 ltd_namespace_phase_list) {
4821 list_del_init(<d->ltd_namespace_phase_list);
4823 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4824 ltd_namespace_phase_list) {
4825 list_del_init(<d->ltd_namespace_phase_list);
4827 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_list,
4828 ltd_namespace_list) {
4829 list_del_init(<d->ltd_namespace_list);
4831 spin_unlock(<ds->ltd_lock);
4833 if (likely(lad->lad_bitmap != NULL))
4834 CFS_FREE_BITMAP(lad->lad_bitmap);
4839 static void lfsck_namespace_quit(const struct lu_env *env,
4840 struct lfsck_component *com)
4842 struct lfsck_assistant_data *lad = com->lc_data;
4843 struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
4844 struct lfsck_tgt_desc *ltd;
4845 struct lfsck_tgt_desc *next;
4847 LASSERT(lad != NULL);
4849 lfsck_quit_generic(env, com);
4851 LASSERT(thread_is_init(&lad->lad_thread) ||
4852 thread_is_stopped(&lad->lad_thread));
4853 LASSERT(list_empty(&lad->lad_req_list));
4855 lfsck_namespace_release_lmv(env, com);
4857 spin_lock(<ds->ltd_lock);
4858 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
4859 ltd_namespace_phase_list) {
4860 list_del_init(<d->ltd_namespace_phase_list);
4862 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4863 ltd_namespace_phase_list) {
4864 list_del_init(<d->ltd_namespace_phase_list);
4866 spin_unlock(<ds->ltd_lock);
4869 static int lfsck_namespace_in_notify(const struct lu_env *env,
4870 struct lfsck_component *com,
4871 struct lfsck_request *lr)
4873 struct lfsck_instance *lfsck = com->lc_lfsck;
4874 struct lfsck_namespace *ns = com->lc_file_ram;
4875 struct lfsck_assistant_data *lad = com->lc_data;
4876 struct lfsck_tgt_descs *ltds = &lfsck->li_mdt_descs;
4877 struct lfsck_tgt_desc *ltd;
4882 switch (lr->lr_event) {
4883 case LE_SET_LMV_MASTER: {
4884 struct dt_object *obj;
4886 obj = lfsck_object_find_bottom(env, lfsck, &lr->lr_fid);
4888 RETURN(PTR_ERR(obj));
4890 if (likely(dt_object_exists(obj)))
4891 rc = lfsck_namespace_notify_lmv_master_local(env, com,
4894 lfsck_object_put(env, obj);
4896 RETURN(rc > 0 ? 0 : rc);
4898 case LE_SET_LMV_SLAVE: {
4899 if (!(lr->lr_flags & LEF_RECHECK_NAME_HASH))
4900 ns->ln_striped_shards_repaired++;
4902 rc = lfsck_namespace_trace_update(env, com, &lr->lr_fid,
4903 LNTF_RECHECK_NAME_HASH, true);
4905 RETURN(rc > 0 ? 0 : rc);
4907 case LE_PHASE1_DONE:
4908 case LE_PHASE2_DONE:
4915 CDEBUG(D_LFSCK, "%s: namespace LFSCK handles notify %u from MDT %x, "
4916 "status %d, flags %x\n", lfsck_lfsck2name(lfsck), lr->lr_event,
4917 lr->lr_index, lr->lr_status, lr->lr_flags2);
4919 spin_lock(<ds->ltd_lock);
4920 ltd = lfsck_ltd2tgt(ltds, lr->lr_index);
4922 spin_unlock(<ds->ltd_lock);
4927 list_del_init(<d->ltd_namespace_phase_list);
4928 switch (lr->lr_event) {
4929 case LE_PHASE1_DONE:
4930 if (lr->lr_status <= 0) {
4931 ltd->ltd_namespace_done = 1;
4932 list_del_init(<d->ltd_namespace_list);
4933 CDEBUG(D_LFSCK, "%s: MDT %x failed/stopped at "
4934 "phase1 for namespace LFSCK: rc = %d.\n",
4935 lfsck_lfsck2name(lfsck),
4936 ltd->ltd_index, lr->lr_status);
4937 ns->ln_flags |= LF_INCOMPLETE;
4942 if (lr->lr_flags2 & LF_INCOMPLETE)
4943 ns->ln_flags |= LF_INCOMPLETE;
4945 if (list_empty(<d->ltd_namespace_list))
4946 list_add_tail(<d->ltd_namespace_list,
4947 &lad->lad_mdt_list);
4948 list_add_tail(<d->ltd_namespace_phase_list,
4949 &lad->lad_mdt_phase2_list);
4951 case LE_PHASE2_DONE:
4952 ltd->ltd_namespace_done = 1;
4953 list_del_init(<d->ltd_namespace_list);
4957 ltd->ltd_namespace_done = 1;
4958 list_del_init(<d->ltd_namespace_list);
4959 if (!(lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT)) {
4961 "%s: the peer MDT %x exit namespace LFSCK\n",
4962 lfsck_lfsck2name(lfsck), ltd->ltd_index);
4963 ns->ln_flags |= LF_INCOMPLETE;
4969 spin_unlock(<ds->ltd_lock);
4971 if (fail && lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) {
4972 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
4974 memset(stop, 0, sizeof(*stop));
4975 stop->ls_status = lr->lr_status;
4976 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
4977 lfsck_stop(env, lfsck->li_bottom, stop);
4978 } else if (lfsck_phase2_next_ready(lad)) {
4979 wake_up_all(&lad->lad_thread.t_ctl_waitq);
4985 static void lfsck_namespace_repaired(struct lfsck_namespace *ns, __u64 *count)
4987 *count += ns->ln_objs_nlink_repaired;
4988 *count += ns->ln_dirent_repaired;
4989 *count += ns->ln_linkea_repaired;
4990 *count += ns->ln_mul_linked_repaired;
4991 *count += ns->ln_unmatched_pairs_repaired;
4992 *count += ns->ln_dangling_repaired;
4993 *count += ns->ln_mul_ref_repaired;
4994 *count += ns->ln_bad_type_repaired;
4995 *count += ns->ln_lost_dirent_repaired;
4996 *count += ns->ln_striped_dirs_disabled;
4997 *count += ns->ln_striped_dirs_repaired;
4998 *count += ns->ln_striped_shards_repaired;
4999 *count += ns->ln_name_hash_repaired;
5000 *count += ns->ln_local_lpf_moved;
5003 static int lfsck_namespace_query_all(const struct lu_env *env,
5004 struct lfsck_component *com,
5005 __u32 *mdts_count, __u64 *repaired)
5007 struct lfsck_namespace *ns = com->lc_file_ram;
5008 struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
5009 struct lfsck_tgt_desc *ltd;
5014 rc = lfsck_query_all(env, com);
5018 down_read(<ds->ltd_rw_sem);
5019 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
5020 ltd = lfsck_ltd2tgt(ltds, idx);
5021 LASSERT(ltd != NULL);
5023 mdts_count[ltd->ltd_namespace_status]++;
5024 *repaired += ltd->ltd_namespace_repaired;
5026 up_read(<ds->ltd_rw_sem);
5028 down_read(&com->lc_sem);
5029 mdts_count[ns->ln_status]++;
5030 lfsck_namespace_repaired(ns, repaired);
5031 up_read(&com->lc_sem);
5036 static int lfsck_namespace_query(const struct lu_env *env,
5037 struct lfsck_component *com,
5038 struct lfsck_request *req,
5039 struct lfsck_reply *rep,
5040 struct lfsck_query *que, int idx)
5042 struct lfsck_namespace *ns = com->lc_file_ram;
5046 LASSERT(com->lc_lfsck->li_master);
5048 rc = lfsck_namespace_query_all(env, com,
5049 que->lu_mdts_count[idx],
5050 &que->lu_repaired[idx]);
5052 down_read(&com->lc_sem);
5053 rep->lr_status = ns->ln_status;
5054 if (req->lr_flags & LEF_QUERY_ALL)
5055 lfsck_namespace_repaired(ns, &rep->lr_repaired);
5056 up_read(&com->lc_sem);
5062 static struct lfsck_operations lfsck_namespace_ops = {
5063 .lfsck_reset = lfsck_namespace_reset,
5064 .lfsck_fail = lfsck_namespace_fail,
5065 .lfsck_close_dir = lfsck_namespace_close_dir,
5066 .lfsck_open_dir = lfsck_namespace_open_dir,
5067 .lfsck_checkpoint = lfsck_namespace_checkpoint,
5068 .lfsck_prep = lfsck_namespace_prep,
5069 .lfsck_exec_oit = lfsck_namespace_exec_oit,
5070 .lfsck_exec_dir = lfsck_namespace_exec_dir,
5071 .lfsck_post = lfsck_namespace_post,
5072 .lfsck_dump = lfsck_namespace_dump,
5073 .lfsck_double_scan = lfsck_namespace_double_scan,
5074 .lfsck_data_release = lfsck_namespace_data_release,
5075 .lfsck_quit = lfsck_namespace_quit,
5076 .lfsck_in_notify = lfsck_namespace_in_notify,
5077 .lfsck_query = lfsck_namespace_query,
5081 * Repair dangling name entry.
5083 * For the name entry with dangling reference, we need to repare the
5084 * inconsistency according to the LFSCK sponsor's requirement:
5086 * 1) Keep the inconsistency there and report the inconsistency case,
5087 * then give the chance to the application to find related issues,
5088 * and the users can make the decision about how to handle it with
5089 * more human knownledge. (by default)
5091 * 2) Re-create the missing MDT-object with the FID information.
5093 * \param[in] env pointer to the thread context
5094 * \param[in] com pointer to the lfsck component
5095 * \param[in] parent pointer to the dir object that contains the dangling
5097 * \param[in] child pointer to the object corresponding to the dangling
5099 * \param[in] lnr pointer to the namespace request that contains the
5100 * name's name, parent object, parent's LMV, and ect.
5102 * \retval positive number if no need to repair
5103 * \retval zero for repaired successfully
5104 * \retval negative error number on failure
5106 int lfsck_namespace_repair_dangling(const struct lu_env *env,
5107 struct lfsck_component *com,
5108 struct dt_object *parent,
5109 struct dt_object *child,
5110 struct lfsck_namespace_req *lnr)
5112 struct lfsck_thread_info *info = lfsck_env_info(env);
5113 struct lu_attr *la = &info->lti_la;
5114 struct dt_allocation_hint *hint = &info->lti_hint;
5115 struct dt_object_format *dof = &info->lti_dof;
5116 struct dt_insert_rec *rec = &info->lti_dt_rec;
5117 struct lmv_mds_md_v1 *lmv2 = &info->lti_lmv2;
5118 const struct lu_name *cname;
5119 const struct lu_fid *pfid = lfsck_dto2fid(parent);
5120 const struct lu_fid *cfid = lfsck_dto2fid(child);
5121 struct linkea_data ldata = { NULL };
5122 struct lfsck_lock_handle *llh = &info->lti_llh;
5123 struct lu_buf linkea_buf;
5124 struct lu_buf lmv_buf;
5125 struct lfsck_instance *lfsck = com->lc_lfsck;
5126 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5127 struct dt_device *dev = lfsck->li_next;
5128 struct thandle *th = NULL;
5130 __u16 type = lnr->lnr_type;
5134 cname = lfsck_name_get_const(env, lnr->lnr_name, lnr->lnr_namelen);
5135 if (bk->lb_param & LPF_CREATE_MDTOBJ)
5140 if (!create || bk->lb_param & LPF_DRYRUN)
5143 /* We may need to create the sub-objects of the @child via LOD,
5144 * so make the modification based on lfsck->li_next. */
5146 parent = lfsck_object_locate(dev, parent);
5148 GOTO(log, rc = PTR_ERR(parent));
5150 if (unlikely(!dt_try_as_dir(env, parent)))
5151 GOTO(log, rc = -ENOTDIR);
5153 child = lfsck_object_locate(dev, child);
5155 GOTO(log, rc = PTR_ERR(child));
5157 rc = linkea_links_new(&ldata, &info->lti_linkea_buf2,
5162 rc = lfsck_lock(env, lfsck, parent, lnr->lnr_name, llh,
5163 MDS_INODELOCK_UPDATE, LCK_PR);
5167 rc = lfsck_namespace_check_exist(env, parent, child, lnr->lnr_name);
5171 /* Set the ctime as zero, then others can know it is created for
5172 * repairing dangling name entry by LFSCK. And if the LFSCK made
5173 * wrong decision and the real MDT-object has been found later,
5174 * then the LFSCK has chance to fix the incosistency properly. */
5175 memset(la, 0, sizeof(*la));
5176 la->la_mode = (type & S_IFMT) | 0600;
5177 la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
5178 LA_ATIME | LA_MTIME | LA_CTIME;
5180 child->do_ops->do_ah_init(env, hint, parent, child,
5181 la->la_mode & S_IFMT);
5183 memset(dof, 0, sizeof(*dof));
5184 dof->dof_type = dt_mode_to_dft(type);
5185 /* If the target is a regular file, then the LFSCK will only create
5186 * the MDT-object without stripes (dof->dof_reg.striped = 0). related
5187 * OST-objects will be created when write open. */
5189 th = dt_trans_create(env, dev);
5191 GOTO(log, rc = PTR_ERR(th));
5193 /* 1a. create child. */
5194 rc = dt_declare_create(env, child, la, hint, dof, th);
5198 if (S_ISDIR(type)) {
5199 if (unlikely(!dt_try_as_dir(env, child)))
5200 GOTO(stop, rc = -ENOTDIR);
5202 /* 2a. increase child nlink */
5203 rc = dt_declare_ref_add(env, child, th);
5207 /* 3a. insert dot into child dir */
5208 rec->rec_type = S_IFDIR;
5209 rec->rec_fid = cfid;
5210 rc = dt_declare_insert(env, child,
5211 (const struct dt_rec *)rec,
5212 (const struct dt_key *)dot, th);
5216 /* 4a. insert dotdot into child dir */
5217 rec->rec_fid = pfid;
5218 rc = dt_declare_insert(env, child,
5219 (const struct dt_rec *)rec,
5220 (const struct dt_key *)dotdot, th);
5224 /* 5a. generate slave LMV EA. */
5225 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
5228 idx = lfsck_shard_name_to_index(env,
5229 lnr->lnr_name, lnr->lnr_namelen,
5231 if (unlikely(idx < 0))
5232 GOTO(stop, rc = idx);
5234 *lmv2 = lnr->lnr_lmv->ll_lmv;
5235 lmv2->lmv_magic = LMV_MAGIC_STRIPE;
5236 lmv2->lmv_master_mdt_index = idx;
5238 lfsck_lmv_header_cpu_to_le(lmv2, lmv2);
5239 lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
5240 rc = dt_declare_xattr_set(env, child, &lmv_buf,
5241 XATTR_NAME_LMV, 0, th);
5247 /* 6a. insert linkEA for child */
5248 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
5249 ldata.ld_leh->leh_len);
5250 rc = dt_declare_xattr_set(env, child, &linkea_buf,
5251 XATTR_NAME_LINK, 0, th);
5255 rc = dt_trans_start_local(env, dev, th);
5257 GOTO(stop, rc = (rc == -EEXIST ? 1 : rc));
5259 dt_write_lock(env, child, 0);
5260 /* 1b. create child */
5261 rc = dt_create(env, child, la, hint, dof, th);
5263 GOTO(unlock, rc = (rc == -EEXIST ? 1 : rc));
5265 if (S_ISDIR(type)) {
5266 /* 2b. increase child nlink */
5267 rc = dt_ref_add(env, child, th);
5271 /* 3b. insert dot into child dir */
5272 rec->rec_type = S_IFDIR;
5273 rec->rec_fid = cfid;
5274 rc = dt_insert(env, child, (const struct dt_rec *)rec,
5275 (const struct dt_key *)dot, th);
5279 /* 4b. insert dotdot into child dir */
5280 rec->rec_fid = pfid;
5281 rc = dt_insert(env, child, (const struct dt_rec *)rec,
5282 (const struct dt_key *)dotdot, th);
5286 /* 5b. generate slave LMV EA. */
5287 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
5288 rc = dt_xattr_set(env, child, &lmv_buf, XATTR_NAME_LMV,
5295 /* 6b. insert linkEA for child. */
5296 rc = dt_xattr_set(env, child, &linkea_buf,
5297 XATTR_NAME_LINK, 0, th);
5302 dt_write_unlock(env, child);
5305 dt_trans_stop(env, dev, th);
5309 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found dangling "
5310 "reference for: parent "DFID", child "DFID", type %u, "
5311 "name %s. %s: rc = %d\n", lfsck_lfsck2name(lfsck),
5312 PFID(pfid), PFID(cfid), type, cname->ln_name,
5313 create ? "Create the lost MDT-object as required" :
5314 "Keep the MDT-object there by default", rc);
5317 struct lfsck_namespace *ns = com->lc_file_ram;
5319 ns->ln_flags |= LF_INCONSISTENT;
5325 static int lfsck_namespace_assistant_handler_p1(const struct lu_env *env,
5326 struct lfsck_component *com,
5327 struct lfsck_assistant_req *lar)
5329 struct lfsck_thread_info *info = lfsck_env_info(env);
5330 struct lu_attr *la = &info->lti_la;
5331 struct lfsck_instance *lfsck = com->lc_lfsck;
5332 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5333 struct lfsck_namespace *ns = com->lc_file_ram;
5334 struct lfsck_assistant_data *lad = com->lc_data;
5335 struct linkea_data ldata = { NULL };
5336 const struct lu_name *cname;
5337 struct thandle *handle = NULL;
5338 struct lfsck_namespace_req *lnr =
5339 container_of0(lar, struct lfsck_namespace_req, lnr_lar);
5340 struct dt_object *dir = NULL;
5341 struct dt_object *obj = NULL;
5342 struct lfsck_assistant_object *lso = lar->lar_parent;
5343 const struct lu_fid *pfid = &lso->lso_fid;
5344 struct dt_device *dev = NULL;
5345 struct lustre_handle lh = { 0 };
5346 bool repaired = false;
5347 bool dtlocked = false;
5351 bool bad_hash = false;
5352 bool bad_linkea = false;
5356 enum lfsck_namespace_inconsistency_type type = LNIT_NONE;
5363 if (lnr->lnr_attr & LUDA_UPGRADE) {
5364 ns->ln_flags |= LF_UPGRADE;
5365 ns->ln_dirent_repaired++;
5367 } else if (lnr->lnr_attr & LUDA_REPAIR) {
5368 ns->ln_flags |= LF_INCONSISTENT;
5369 ns->ln_dirent_repaired++;
5373 if (unlikely(fid_is_zero(&lnr->lnr_fid))) {
5374 if (strcmp(lnr->lnr_name, dotdot) != 0)
5377 rc = lfsck_namespace_trace_update(env, com, pfid,
5378 LNTF_CHECK_PARENT, true);
5383 if (unlikely(!fid_is_sane(&lnr->lnr_fid))) {
5384 CDEBUG(D_LFSCK, "%s: dir scan find invalid FID "DFID
5385 " for the name entry %.*s under "DFID"\n",
5386 lfsck_lfsck2name(lfsck), PFID(&lnr->lnr_fid),
5387 lnr->lnr_namelen, lnr->lnr_name, PFID(pfid));
5389 if (strcmp(lnr->lnr_name, dotdot) != 0)
5390 /* invalid FID means bad name entry, remove it. */
5391 type = LNIT_BAD_DIRENT;
5393 /* If the parent FID is invalid, we cannot remove
5394 * the ".." entry directly. */
5395 rc = lfsck_namespace_trace_update(env, com, pfid,
5396 LNTF_CHECK_PARENT, true);
5401 if (unlikely(lnr->lnr_dir_cookie == MDS_DIR_END_OFF)) {
5402 rc = lfsck_namespace_striped_dir_rescan(env, com, lnr);
5407 if (fid_seq_is_dot(fid_seq(&lnr->lnr_fid)))
5410 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
5411 rc = lfsck_namespace_handle_striped_master(env, com, lnr);
5416 idx = lfsck_find_mdt_idx_by_fid(env, lfsck, &lnr->lnr_fid);
5418 GOTO(out, rc = idx);
5420 if (idx == lfsck_dev_idx(lfsck)) {
5421 if (unlikely(strcmp(lnr->lnr_name, dotdot) == 0))
5424 dev = lfsck->li_bottom;
5426 struct lfsck_tgt_desc *ltd;
5428 /* Usually, some local filesystem consistency verification
5429 * tools can guarantee the local namespace tree consistenct.
5430 * So the LFSCK will only verify the remote directory. */
5431 if (unlikely(strcmp(lnr->lnr_name, dotdot) == 0)) {
5432 rc = lfsck_namespace_trace_update(env, com, pfid,
5433 LNTF_CHECK_PARENT, true);
5438 ltd = lfsck_ltd2tgt(&lfsck->li_mdt_descs, idx);
5439 if (unlikely(ltd == NULL)) {
5440 CDEBUG(D_LFSCK, "%s: cannot talk with MDT %x which "
5441 "did not join the namespace LFSCK\n",
5442 lfsck_lfsck2name(lfsck), idx);
5443 lfsck_lad_set_bitmap(env, com, idx);
5445 GOTO(out, rc = -ENODEV);
5451 obj = lfsck_object_find_by_dev(env, dev, &lnr->lnr_fid);
5453 GOTO(out, rc = PTR_ERR(obj));
5455 cname = lfsck_name_get_const(env, lnr->lnr_name, lnr->lnr_namelen);
5456 if (dt_object_exists(obj) == 0) {
5460 dir = lfsck_assistant_object_load(env, lfsck, lso);
5464 GOTO(trace, rc == -ENOENT ? 0 : rc);
5468 rc = lfsck_namespace_check_exist(env, dir, obj, lnr->lnr_name);
5470 if (!lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5471 lnr->lnr_name, lnr->lnr_namelen)) {
5472 type = LNIT_BAD_DIRENT;
5477 type = LNIT_DANGLING;
5478 rc = lfsck_namespace_repair_dangling(env, com, dir,
5487 if (!(bk->lb_param & LPF_DRYRUN) && lad->lad_advance_lock) {
5490 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
5491 MDS_INODELOCK_UPDATE |
5492 MDS_INODELOCK_XATTR, LCK_EX);
5496 handle = dt_trans_create(env, dev);
5498 GOTO(out, rc = PTR_ERR(handle));
5500 rc = lfsck_declare_namespace_exec_dir(env, obj, handle);
5504 rc = dt_trans_start_local(env, dev, handle);
5508 dt_write_lock(env, obj, 0);
5512 rc = lfsck_links_read(env, obj, &ldata);
5513 if (unlikely(rc == -ENOENT)) {
5514 if (handle != NULL) {
5515 dt_write_unlock(env, obj);
5518 dt_trans_stop(env, dev, handle);
5521 lfsck_ibits_unlock(&lh, LCK_EX);
5524 /* It may happen when the remote object has been removed,
5525 * but the local MDT is not aware of that. */
5527 } else if (rc == 0) {
5528 count = ldata.ld_leh->leh_reccount;
5529 rc = linkea_links_find(&ldata, cname, pfid);
5531 (count == 1 || !S_ISDIR(lfsck_object_type(obj)))) {
5532 if ((lfsck_object_type(obj) & S_IFMT) !=
5534 ns->ln_flags |= LF_INCONSISTENT;
5535 type = LNIT_BAD_TYPE;
5541 /* If the name entry hash does not match the slave striped
5542 * directory, and the name entry does not match also, then
5543 * it is quite possible that name entry is corrupted. */
5544 if (!lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5545 lnr->lnr_name, lnr->lnr_namelen)) {
5546 ns->ln_flags |= LF_INCONSISTENT;
5547 type = LNIT_BAD_DIRENT;
5552 /* If the file type stored in the name entry does not match
5553 * the file type claimed by the object, and the object does
5554 * not recognize the name entry, then it is quite possible
5555 * that the name entry is corrupted. */
5556 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type) {
5557 ns->ln_flags |= LF_INCONSISTENT;
5558 type = LNIT_BAD_DIRENT;
5563 /* For sub-dir object, we cannot make sure whether the sub-dir
5564 * back references the parent via ".." name entry correctly or
5565 * not in the LFSCK first-stage scanning. It may be that the
5566 * (remote) sub-dir ".." name entry has no parent FID after
5567 * file-level backup/restore and its linkEA may be wrong.
5568 * So under such case, we should replace the linkEA according
5569 * to current name entry. But this needs to be done during the
5570 * LFSCK second-stage scanning. The LFSCK will record the name
5571 * entry for further possible using. */
5575 } else if (unlikely(rc == -EINVAL)) {
5576 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type)
5577 type = LNIT_BAD_TYPE;
5580 /* The magic crashed, we are not sure whether there are more
5581 * corrupt data in the linkea, so remove all linkea entries. */
5585 } else if (rc == -ENODATA) {
5586 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type)
5587 type = LNIT_BAD_TYPE;
5594 if (bk->lb_param & LPF_DRYRUN) {
5596 ns->ln_flags |= LF_UPGRADE;
5598 ns->ln_flags |= LF_INCONSISTENT;
5599 ns->ln_linkea_repaired++;
5605 if (!lustre_handle_is_used(&lh)) {
5613 LASSERT(handle != NULL);
5616 dir = lfsck_assistant_object_load(env, lfsck, lso);
5620 GOTO(stop, rc == -ENOENT ? 0 : rc);
5624 rc = lfsck_namespace_check_exist(env, dir, obj, lnr->lnr_name);
5629 if (!remove && newdata)
5630 ns->ln_flags |= LF_UPGRADE;
5631 else if (remove || !(ns->ln_flags & LF_UPGRADE))
5632 ns->ln_flags |= LF_INCONSISTENT;
5637 rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, handle);
5638 if (rc != 0 && rc != -ENOENT && rc != -ENODATA)
5643 rc = linkea_data_new(&ldata,
5644 &lfsck_env_info(env)->lti_linkea_buf);
5649 rc = linkea_add_buf(&ldata, cname, pfid);
5651 rc = lfsck_links_write(env, obj, &ldata, handle);
5655 count = ldata.ld_leh->leh_reccount;
5656 if (!S_ISDIR(lfsck_object_type(obj)) ||
5657 !dt_object_remote(obj)) {
5658 ns->ln_linkea_repaired++;
5668 dt_write_unlock(env, obj);
5670 if (handle != NULL && !IS_ERR(handle))
5671 dt_trans_stop(env, dev, handle);
5674 lfsck_ibits_unlock(&lh, LCK_EX);
5676 if (!name_is_dot_or_dotdot(lnr->lnr_name, lnr->lnr_namelen) &&
5677 !lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5678 lnr->lnr_name, lnr->lnr_namelen) &&
5679 type != LNIT_BAD_DIRENT) {
5680 ns->ln_flags |= LF_INCONSISTENT;
5684 dir = lfsck_assistant_object_load(env, lfsck, lso);
5688 GOTO(trace, rc == -ENOENT ? 0 : rc);
5692 rc = lfsck_namespace_repair_bad_name_hash(env, com, dir,
5693 lnr->lnr_lmv, lnr->lnr_name);
5699 if (type != LNIT_NONE && dir == NULL) {
5700 dir = lfsck_assistant_object_load(env, lfsck, lso);
5704 GOTO(trace, rc == -ENOENT ? 0 : rc);
5711 rc = lfsck_namespace_repair_dirent(env, com, dir,
5712 obj, lnr->lnr_name, lnr->lnr_name,
5713 lnr->lnr_type, true, false);
5717 case LNIT_BAD_DIRENT:
5719 /* XXX: This is a bad dirent, we do not know whether
5720 * the original name entry reference a regular
5721 * file or a directory, then keep the parent's
5722 * nlink count unchanged here. */
5723 rc = lfsck_namespace_repair_dirent(env, com, dir,
5724 obj, lnr->lnr_name, lnr->lnr_name,
5725 lnr->lnr_type, false, false);
5733 if (obj != NULL && count == 1 &&
5734 S_ISREG(lfsck_object_type(obj)))
5735 dt_attr_get(env, obj, la);
5739 down_write(&com->lc_sem);
5741 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail to handle "
5742 "the entry: "DFID", parent "DFID", name %.*s: rc = %d\n",
5743 lfsck_lfsck2name(lfsck), PFID(&lnr->lnr_fid), PFID(pfid),
5744 lnr->lnr_namelen, lnr->lnr_name, rc);
5746 lfsck_namespace_record_failure(env, lfsck, ns);
5747 if ((rc == -ENOTCONN || rc == -ESHUTDOWN || rc == -EREMCHG ||
5748 rc == -ETIMEDOUT || rc == -EHOSTDOWN ||
5749 rc == -EHOSTUNREACH || rc == -EINPROGRESS) &&
5750 dev != NULL && dev != lfsck->li_bottom)
5751 lfsck_lad_set_bitmap(env, com, idx);
5753 if (!(bk->lb_param & LPF_FAILOUT))
5757 ns->ln_items_repaired++;
5759 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant "
5760 "repaired the entry: "DFID", parent "DFID
5761 ", name %.*s, type %d\n",
5762 lfsck_lfsck2name(lfsck),
5763 PFID(&lnr->lnr_fid), PFID(pfid),
5764 lnr->lnr_namelen, lnr->lnr_name, type);
5768 ns->ln_dangling_repaired++;
5771 ns->ln_bad_type_repaired++;
5773 case LNIT_BAD_DIRENT:
5774 ns->ln_dirent_repaired++;
5780 if (bk->lb_param & LPF_DRYRUN &&
5781 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent))
5782 lfsck_pos_fill(env, lfsck,
5783 &ns->ln_pos_first_inconsistent,
5788 ns->ln_name_hash_repaired++;
5790 /* Not count repeatedly. */
5792 ns->ln_items_repaired++;
5794 CDEBUG(D_LFSCK, "%s: namespace LFSCK "
5795 "assistant repaired the entry: "
5798 lfsck_lfsck2name(lfsck),
5799 PFID(&lnr->lnr_fid), PFID(pfid),
5800 lnr->lnr_namelen, lnr->lnr_name);
5803 if (bk->lb_param & LPF_DRYRUN &&
5804 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent))
5805 lfsck_pos_fill(env, lfsck,
5806 &ns->ln_pos_first_inconsistent,
5813 if (count > 1 || la->la_nlink > 1)
5814 ns->ln_mul_linked_checked++;
5816 up_write(&com->lc_sem);
5818 if (obj != NULL && !IS_ERR(obj))
5819 lfsck_object_put(env, obj);
5821 if (dir != NULL && !IS_ERR(dir))
5822 lfsck_object_put(env, dir);
5824 lad->lad_advance_lock = bad_linkea;
5830 * Handle one orphan under the backend /lost+found directory
5832 * Insert the orphan FID into the namespace LFSCK trace file for further
5833 * processing (via the subsequent namespace LFSCK second-stage scanning).
5834 * At the same time, remove the orphan name entry from backend /lost+found
5835 * directory. There is an interval between the orphan name entry removed
5836 * from the backend /lost+found directory and the orphan FID in the LFSCK
5837 * trace file handled. In such interval, the LFSCK can be reset, then
5838 * all the FIDs recorded in the namespace LFSCK trace file will be dropped.
5839 * To guarantee that the orphans can be found when LFSCK run next time
5840 * without e2fsck again, when remove the orphan name entry, the LFSCK
5841 * will set the orphan's ctime attribute as 1. Since normal applications
5842 * cannot change the object's ctime attribute as 1. Then when LFSCK run
5843 * next time, it can record the object (that ctime is 1) in the namespace
5844 * LFSCK trace file during the first-stage scanning.
5846 * \param[in] env pointer to the thread context
5847 * \param[in] com pointer to the lfsck component
5848 * \param[in] parent pointer to the object for the backend /lost+found
5849 * \param[in] ent pointer to the name entry for the target under the
5850 * backend /lost+found
5852 * \retval positive for repaired
5853 * \retval 0 if needs to repair nothing
5854 * \retval negative error number on failure
5856 static int lfsck_namespace_scan_local_lpf_one(const struct lu_env *env,
5857 struct lfsck_component *com,
5858 struct dt_object *parent,
5859 struct lu_dirent *ent)
5861 struct lfsck_thread_info *info = lfsck_env_info(env);
5862 struct lu_fid *key = &info->lti_fid;
5863 struct lu_attr *la = &info->lti_la;
5864 struct lfsck_instance *lfsck = com->lc_lfsck;
5865 struct dt_object *obj;
5866 struct dt_device *dev = lfsck->li_bottom;
5867 struct dt_object *child = NULL;
5868 struct thandle *th = NULL;
5875 child = lfsck_object_find_by_dev(env, dev, &ent->lde_fid);
5877 RETURN(PTR_ERR(child));
5879 LASSERT(dt_object_exists(child));
5880 LASSERT(!dt_object_remote(child));
5882 idx = lfsck_sub_trace_file_fid2idx(&ent->lde_fid);
5883 obj = com->lc_sub_trace_objs[idx].lsto_obj;
5884 fid_cpu_to_be(key, &ent->lde_fid);
5885 rc = dt_lookup(env, obj, (struct dt_rec *)&flags,
5886 (const struct dt_key *)key);
5889 flags |= LNTF_CHECK_ORPHAN;
5890 } else if (rc == -ENOENT) {
5891 flags = LNTF_CHECK_ORPHAN;
5896 th = dt_trans_create(env, dev);
5898 GOTO(out, rc = PTR_ERR(th));
5900 /* a1. remove name entry from backend /lost+found */
5901 rc = dt_declare_delete(env, parent,
5902 (const struct dt_key *)ent->lde_name, th);
5906 if (S_ISDIR(lfsck_object_type(child))) {
5907 /* a2. decrease parent's nlink */
5908 rc = dt_declare_ref_del(env, parent, th);
5914 /* a3. remove child's FID from the LFSCK trace file. */
5915 rc = dt_declare_delete(env, obj,
5916 (const struct dt_key *)key, th);
5920 /* a4. set child's ctime as 1 */
5921 memset(la, 0, sizeof(*la));
5923 la->la_valid = LA_CTIME;
5924 rc = dt_declare_attr_set(env, child, la, th);
5929 /* a5. insert child's FID into the LFSCK trace file. */
5930 rc = dt_declare_insert(env, obj, (const struct dt_rec *)&flags,
5931 (const struct dt_key *)key, th);
5935 rc = dt_trans_start_local(env, dev, th);
5939 /* b1. remove name entry from backend /lost+found */
5940 rc = dt_delete(env, parent, (const struct dt_key *)ent->lde_name, th);
5944 if (S_ISDIR(lfsck_object_type(child))) {
5945 /* b2. decrease parent's nlink */
5946 dt_write_lock(env, parent, 0);
5947 rc = dt_ref_del(env, parent, th);
5948 dt_write_unlock(env, parent);
5954 /* a3. remove child's FID from the LFSCK trace file. */
5955 rc = dt_delete(env, obj, (const struct dt_key *)key, th);
5959 /* b4. set child's ctime as 1 */
5960 rc = dt_attr_set(env, child, la, th);
5965 /* b5. insert child's FID into the LFSCK trace file. */
5966 rc = dt_insert(env, obj, (const struct dt_rec *)&flags,
5967 (const struct dt_key *)key, th);
5969 GOTO(stop, rc = (rc == 0 ? 1 : rc));
5972 dt_trans_stop(env, dev, th);
5975 lfsck_object_put(env, child);
5981 * Handle orphans under the backend /lost+found directory
5983 * Some backend checker, such as e2fsck for ldiskfs may find some orphans
5984 * and put them under the backend /lost+found directory that is invisible
5985 * to client. The LFSCK will scan such directory, for the original client
5986 * visible orphans, add their fids into the namespace LFSCK trace file,
5987 * then the subsenquent namespace LFSCK second-stage scanning can handle
5988 * them as other objects to be double scanned: either move back to normal
5989 * namespace, or to the global visible orphan directory:
5990 * /ROOT/.lustre/lost+found/MDTxxxx/
5992 * \param[in] env pointer to the thread context
5993 * \param[in] com pointer to the lfsck component
5995 static void lfsck_namespace_scan_local_lpf(const struct lu_env *env,
5996 struct lfsck_component *com)
5998 struct lfsck_thread_info *info = lfsck_env_info(env);
5999 struct lu_dirent *ent =
6000 (struct lu_dirent *)info->lti_key;
6001 struct lu_seq_range *range = &info->lti_range;
6002 struct lfsck_instance *lfsck = com->lc_lfsck;
6003 struct ptlrpc_thread *thread = &lfsck->li_thread;
6004 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
6005 struct lfsck_namespace *ns = com->lc_file_ram;
6006 struct dt_object *parent;
6007 const struct dt_it_ops *iops;
6009 struct seq_server_site *ss = lfsck_dev_site(lfsck);
6011 __u32 idx = lfsck_dev_idx(lfsck);
6016 parent = lfsck_object_find_by_dev(env, lfsck->li_bottom,
6017 &LU_BACKEND_LPF_FID);
6018 if (IS_ERR(parent)) {
6019 CERROR("%s: fail to find backend /lost+found: rc = %ld\n",
6020 lfsck_lfsck2name(lfsck), PTR_ERR(parent));
6024 /* It is normal that the /lost+found does not exist for ZFS backend. */
6025 if (!dt_object_exists(parent))
6028 if (unlikely(!dt_try_as_dir(env, parent)))
6029 GOTO(out, rc = -ENOTDIR);
6031 CDEBUG(D_LFSCK, "%s: start to scan backend /lost+found\n",
6032 lfsck_lfsck2name(lfsck));
6034 com->lc_new_scanned = 0;
6035 iops = &parent->do_index_ops->dio_it;
6036 di = iops->init(env, parent, LUDA_64BITHASH | LUDA_TYPE);
6038 GOTO(out, rc = PTR_ERR(di));
6040 rc = iops->load(env, di, 0);
6042 rc = iops->next(env, di);
6047 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
6048 unlikely(!thread_is_running(thread)))
6051 rc = iops->rec(env, di, (struct dt_rec *)ent,
6052 LUDA_64BITHASH | LUDA_TYPE);
6054 rc = lfsck_unpack_ent(ent, &cookie, &type);
6056 if (unlikely(rc != 0)) {
6057 CDEBUG(D_LFSCK, "%s: fail to iterate backend "
6058 "/lost+found: rc = %d\n",
6059 lfsck_lfsck2name(lfsck), rc);
6064 /* skip dot and dotdot entries */
6065 if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen))
6068 if (!fid_seq_in_fldb(fid_seq(&ent->lde_fid)))
6071 if (fid_is_norm(&ent->lde_fid)) {
6072 fld_range_set_mdt(range);
6073 rc = fld_local_lookup(env, ss->ss_server_fld,
6074 fid_seq(&ent->lde_fid), range);
6077 } else if (idx != 0) {
6078 /* If the returned FID is IGIF, then there are three
6081 * 1) The object is upgraded from old Lustre-1.8 with
6082 * IGIF assigned to such object.
6083 * 2) The object is a backend local object and is
6084 * invisible to client.
6085 * 3) The object lost its LMV EA, and since there is
6086 * no FID-in-dirent for the orphan in the backend
6087 * /lost+found directory, then the low layer will
6088 * return IGIF for such object.
6090 * For MDTx (x != 0), it is either case 2) or case 3),
6091 * but from the LFSCK view, they are indistinguishable.
6092 * To be safe, the LFSCK will keep it there and report
6093 * some message, then the adminstrator can handle that
6096 * For MDT0, it is more possible the case 1). The LFSCK
6097 * will handle the orphan as an upgraded object. */
6098 CDEBUG(D_LFSCK, "%s: the orphan %.*s with IGIF "DFID
6099 "in the backend /lost+found on the MDT %04x, "
6100 "to be safe, skip it.\n",
6101 lfsck_lfsck2name(lfsck), ent->lde_namelen,
6102 ent->lde_name, PFID(&ent->lde_fid), idx);
6106 rc = lfsck_namespace_scan_local_lpf_one(env, com, parent, ent);
6109 down_write(&com->lc_sem);
6110 com->lc_new_scanned++;
6111 ns->ln_local_lpf_scanned++;
6113 ns->ln_local_lpf_moved++;
6115 ns->ln_local_lpf_skipped++;
6117 ns->ln_local_lpf_failed++;
6118 up_write(&com->lc_sem);
6120 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
6124 lfsck_control_speed_by_self(com);
6125 if (unlikely(!thread_is_running(thread))) {
6130 rc = iops->next(env, di);
6134 iops->fini(env, di);
6139 CDEBUG(D_LFSCK, "%s: stop to scan backend /lost+found: rc = %d\n",
6140 lfsck_lfsck2name(lfsck), rc);
6142 lfsck_object_put(env, parent);
6146 * Rescan the striped directory after the master LMV EA reset.
6148 * Sometimes, the master LMV EA of the striped directory maybe lost, so when
6149 * the namespace LFSCK engine scan the striped directory for the first time,
6150 * it will be regarded as a normal directory. As the LFSCK processing, some
6151 * other LFSCK instance on other MDT will find the shard of this striped dir,
6152 * and find that the master MDT-object of the striped directory lost its LMV
6153 * EA, then such remote LFSCK instance will regenerate the master LMV EA and
6154 * notify the LFSCK instance on this MDT to rescan the striped directory.
6156 * \param[in] env pointer to the thread context
6157 * \param[in] com pointer to the lfsck component
6158 * \param[in] llu the lfsck_lmv_unit that contains the striped directory
6161 * \retval positive number for success
6162 * \retval 0 for LFSCK stopped/paused
6163 * \retval negative error number on failure
6165 static int lfsck_namespace_rescan_striped_dir(const struct lu_env *env,
6166 struct lfsck_component *com,
6167 struct lfsck_lmv_unit *llu)
6169 struct lfsck_thread_info *info = lfsck_env_info(env);
6170 struct lfsck_instance *lfsck = com->lc_lfsck;
6171 struct lfsck_assistant_data *lad = com->lc_data;
6172 struct dt_object *dir;
6173 const struct dt_it_ops *iops;
6175 struct lu_dirent *ent =
6176 (struct lu_dirent *)info->lti_key;
6177 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
6178 struct ptlrpc_thread *thread = &lfsck->li_thread;
6179 struct lfsck_assistant_object *lso = NULL;
6180 struct lfsck_namespace_req *lnr;
6181 struct lfsck_assistant_req *lar;
6186 LASSERT(list_empty(&lad->lad_req_list));
6188 lfsck->li_lmv = &llu->llu_lmv;
6189 lfsck->li_obj_dir = lfsck_object_get(llu->llu_obj);
6190 rc = lfsck_open_dir(env, lfsck, 0);
6194 dir = lfsck->li_obj_dir;
6195 di = lfsck->li_di_dir;
6196 iops = &dir->do_index_ops->dio_it;
6198 rc = iops->rec(env, di, (struct dt_rec *)ent,
6199 lfsck->li_args_dir);
6201 rc = lfsck_unpack_ent(ent, &lfsck->li_cookie_dir,
6205 if (bk->lb_param & LPF_FAILOUT)
6211 if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen))
6215 lso = lfsck_assistant_object_init(env,
6216 lfsck_dto2fid(dir), NULL,
6217 lfsck->li_pos_current.lp_oit_cookie, true);
6219 if (bk->lb_param & LPF_FAILOUT)
6220 GOTO(out, rc = PTR_ERR(lso));
6227 lnr = lfsck_namespace_assistant_req_init(lfsck, lso, ent, type);
6229 if (bk->lb_param & LPF_FAILOUT)
6230 GOTO(out, rc = PTR_ERR(lnr));
6235 lar = &lnr->lnr_lar;
6236 rc = lfsck_namespace_assistant_handler_p1(env, com, lar);
6237 lfsck_namespace_assistant_req_fini(env, lar);
6238 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
6241 if (unlikely(!thread_is_running(thread)))
6245 rc = iops->next(env, di);
6249 if (lso != NULL && !IS_ERR(lso))
6250 lfsck_assistant_object_put(env, lso);
6252 lfsck_close_dir(env, lfsck, rc);
6256 /* The close_dir() may insert a dummy lnr in the lad->lad_req_list. */
6257 if (list_empty(&lad->lad_req_list))
6260 spin_lock(&lad->lad_lock);
6261 lar = list_entry(lad->lad_req_list.next, struct lfsck_assistant_req,
6263 list_del_init(&lar->lar_list);
6264 spin_unlock(&lad->lad_lock);
6266 rc = lfsck_namespace_assistant_handler_p1(env, com, lar);
6267 lfsck_namespace_assistant_req_fini(env, lar);
6269 RETURN(rc == 0 ? 1 : rc);
6273 lfsck_namespace_double_scan_one_trace_file(const struct lu_env *env,
6274 struct lfsck_component *com,
6275 struct dt_object *obj, bool first)
6277 struct lfsck_instance *lfsck = com->lc_lfsck;
6278 struct ptlrpc_thread *thread = &lfsck->li_thread;
6279 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
6280 struct lfsck_namespace *ns = com->lc_file_ram;
6281 const struct dt_it_ops *iops = &obj->do_index_ops->dio_it;
6282 struct dt_object *target;
6290 di = iops->init(env, obj, 0);
6292 RETURN(PTR_ERR(di));
6295 fid_cpu_to_be(&fid, &ns->ln_fid_latest_scanned_phase2);
6298 rc = iops->get(env, di, (const struct dt_key *)&fid);
6303 /* The start one either has been processed or does not exist,
6305 rc = iops->next(env, di);
6311 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
6312 unlikely(!thread_is_running(thread)))
6315 key = iops->key(env, di);
6324 fid_be_to_cpu(&fid, (const struct lu_fid *)key);
6325 if (!fid_is_sane(&fid)) {
6330 target = lfsck_object_find_bottom(env, lfsck, &fid);
6331 if (IS_ERR(target)) {
6332 rc = PTR_ERR(target);
6336 if (dt_object_exists(target)) {
6337 rc = iops->rec(env, di, (struct dt_rec *)&flags, 0);
6339 rc = lfsck_namespace_double_scan_one(env, com,
6346 lfsck_object_put(env, target);
6349 down_write(&com->lc_sem);
6350 com->lc_new_checked++;
6351 com->lc_new_scanned++;
6353 ns->ln_fid_latest_scanned_phase2 = fid;
6356 ns->ln_objs_repaired_phase2++;
6358 ns->ln_objs_failed_phase2++;
6359 up_write(&com->lc_sem);
6361 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
6364 if (unlikely(com->lc_time_next_checkpoint <=
6365 ktime_get_seconds()) &&
6366 com->lc_new_checked != 0) {
6367 down_write(&com->lc_sem);
6368 ns->ln_run_time_phase2 += ktime_get_seconds() -
6369 com->lc_time_last_checkpoint;
6370 ns->ln_time_last_checkpoint = ktime_get_real_seconds();
6371 ns->ln_objs_checked_phase2 += com->lc_new_checked;
6372 com->lc_new_checked = 0;
6373 lfsck_namespace_store(env, com);
6374 up_write(&com->lc_sem);
6376 com->lc_time_last_checkpoint = ktime_get_seconds();
6377 com->lc_time_next_checkpoint =
6378 com->lc_time_last_checkpoint +
6379 LFSCK_CHECKPOINT_INTERVAL;
6382 lfsck_control_speed_by_self(com);
6383 if (unlikely(!thread_is_running(thread)))
6386 rc = iops->next(env, di);
6395 iops->fini(env, di);
6400 static int lfsck_namespace_assistant_handler_p2(const struct lu_env *env,
6401 struct lfsck_component *com)
6403 struct lfsck_instance *lfsck = com->lc_lfsck;
6404 struct lfsck_namespace *ns = com->lc_file_ram;
6409 while (!list_empty(&lfsck->li_list_lmv)) {
6410 struct lfsck_lmv_unit *llu;
6412 spin_lock(&lfsck->li_lock);
6413 llu = list_entry(lfsck->li_list_lmv.next,
6414 struct lfsck_lmv_unit, llu_link);
6415 list_del_init(&llu->llu_link);
6416 spin_unlock(&lfsck->li_lock);
6418 rc = lfsck_namespace_rescan_striped_dir(env, com, llu);
6423 CDEBUG(D_LFSCK, "%s: namespace LFSCK phase2 scan start\n",
6424 lfsck_lfsck2name(lfsck));
6426 lfsck_namespace_scan_local_lpf(env, com);
6428 com->lc_new_checked = 0;
6429 com->lc_new_scanned = 0;
6430 com->lc_time_last_checkpoint = ktime_get_seconds();
6431 com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
6432 LFSCK_CHECKPOINT_INTERVAL;
6434 i = lfsck_sub_trace_file_fid2idx(&ns->ln_fid_latest_scanned_phase2);
6435 rc = lfsck_namespace_double_scan_one_trace_file(env, com,
6436 com->lc_sub_trace_objs[i].lsto_obj, true);
6437 while (rc > 0 && ++i < LFSCK_STF_COUNT)
6438 rc = lfsck_namespace_double_scan_one_trace_file(env, com,
6439 com->lc_sub_trace_objs[i].lsto_obj, false);
6441 CDEBUG(D_LFSCK, "%s: namespace LFSCK phase2 scan stop at the No. %d "
6442 "trace file: rc = %d\n", lfsck_lfsck2name(lfsck), i, rc);
6447 static void lfsck_namespace_assistant_fill_pos(const struct lu_env *env,
6448 struct lfsck_component *com,
6449 struct lfsck_position *pos)
6451 struct lfsck_assistant_data *lad = com->lc_data;
6452 struct lfsck_namespace_req *lnr;
6454 if (((struct lfsck_namespace *)(com->lc_file_ram))->ln_status !=
6458 if (list_empty(&lad->lad_req_list))
6461 lnr = list_entry(lad->lad_req_list.next,
6462 struct lfsck_namespace_req,
6464 pos->lp_oit_cookie = lnr->lnr_lar.lar_parent->lso_oit_cookie;
6465 pos->lp_dir_cookie = lnr->lnr_dir_cookie - 1;
6466 pos->lp_dir_parent = lnr->lnr_lar.lar_parent->lso_fid;
6469 static int lfsck_namespace_double_scan_result(const struct lu_env *env,
6470 struct lfsck_component *com,
6473 struct lfsck_instance *lfsck = com->lc_lfsck;
6474 struct lfsck_namespace *ns = com->lc_file_ram;
6476 down_write(&com->lc_sem);
6477 ns->ln_run_time_phase2 += ktime_get_seconds() -
6478 com->lc_time_last_checkpoint;
6479 ns->ln_time_last_checkpoint = ktime_get_real_seconds();
6480 ns->ln_objs_checked_phase2 += com->lc_new_checked;
6481 com->lc_new_checked = 0;
6484 if (ns->ln_flags & LF_INCOMPLETE)
6485 ns->ln_status = LS_PARTIAL;
6487 ns->ln_status = LS_COMPLETED;
6488 ns->ln_flags &= ~LF_SCANNED_ONCE;
6489 if (!(lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN))
6490 ns->ln_flags &= ~LF_INCONSISTENT;
6491 ns->ln_time_last_complete = ns->ln_time_last_checkpoint;
6492 ns->ln_success_count++;
6493 } else if (rc == 0) {
6494 if (lfsck->li_status != 0)
6495 ns->ln_status = lfsck->li_status;
6497 ns->ln_status = LS_STOPPED;
6499 ns->ln_status = LS_FAILED;
6502 rc = lfsck_namespace_store(env, com);
6503 up_write(&com->lc_sem);
6509 lfsck_namespace_assistant_sync_failures_interpret(const struct lu_env *env,
6510 struct ptlrpc_request *req,
6514 struct lfsck_async_interpret_args *laia = args;
6515 struct lfsck_tgt_desc *ltd = laia->laia_ltd;
6517 ltd->ltd_synced_failures = 1;
6524 * Notify remote LFSCK instances about former failures.
6526 * The local LFSCK instance has recorded which MDTs have ever failed to respond
6527 * some LFSCK verification requests (maybe because of network issues or the MDT
6528 * itself trouble). During the respond gap the MDT may missed some name entries
6529 * verification, then the MDT cannot know whether related MDT-objects have been
6530 * referenced by related name entries or not, then in the second-stage scanning,
6531 * these MDT-objects will be regarded as orphan, if the MDT-object contains bad
6532 * linkEA for back reference, then it will misguide the LFSCK to generate wrong
6533 * name entry for repairing the orphan.
6535 * To avoid above trouble, when layout LFSCK finishes the first-stage scanning,
6536 * it will scan the bitmap for the ever failed MDTs, and notify them that they
6537 * have ever missed some name entries verification and should skip the handling
6538 * for orphan MDT-objects.
6540 * \param[in] env pointer to the thread context
6541 * \param[in] com pointer to the lfsck component
6542 * \param[in] lr pointer to the lfsck request
6544 static void lfsck_namespace_assistant_sync_failures(const struct lu_env *env,
6545 struct lfsck_component *com,
6546 struct lfsck_request *lr)
6548 struct lfsck_async_interpret_args *laia =
6549 &lfsck_env_info(env)->lti_laia2;
6550 struct lfsck_assistant_data *lad = com->lc_data;
6551 struct lfsck_namespace *ns = com->lc_file_ram;
6552 struct lfsck_instance *lfsck = com->lc_lfsck;
6553 struct lfsck_tgt_descs *ltds = &lfsck->li_mdt_descs;
6554 struct lfsck_tgt_desc *ltd;
6555 struct ptlrpc_request_set *set;
6560 if (!test_bit(LAD_INCOMPLETE, &lad->lad_flags))
6563 set = ptlrpc_prep_set();
6565 GOTO(out, rc = -ENOMEM);
6567 lr->lr_flags2 = ns->ln_flags | LF_INCOMPLETE;
6568 memset(laia, 0, sizeof(*laia));
6569 lad->lad_touch_gen++;
6571 down_read(<ds->ltd_rw_sem);
6572 cfs_foreach_bit(lad->lad_bitmap, idx) {
6573 ltd = lfsck_ltd2tgt(ltds, idx);
6577 laia->laia_ltd = ltd;
6578 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
6579 lfsck_namespace_assistant_sync_failures_interpret,
6580 laia, LFSCK_NOTIFY);
6582 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail "
6583 "to sync failure with MDT %x: rc = %d\n",
6584 lfsck_lfsck2name(lfsck), ltd->ltd_index, rc);
6586 up_read(<ds->ltd_rw_sem);
6588 rc = ptlrpc_set_wait(env, set);
6589 ptlrpc_set_destroy(set);
6595 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail "
6596 "to sync failure with MDTs, and related MDTs "
6597 "may handle orphan improperly: rc = %d\n",
6598 lfsck_lfsck2name(lfsck), rc);
6603 struct lfsck_assistant_operations lfsck_namespace_assistant_ops = {
6604 .la_handler_p1 = lfsck_namespace_assistant_handler_p1,
6605 .la_handler_p2 = lfsck_namespace_assistant_handler_p2,
6606 .la_fill_pos = lfsck_namespace_assistant_fill_pos,
6607 .la_double_scan_result = lfsck_namespace_double_scan_result,
6608 .la_req_fini = lfsck_namespace_assistant_req_fini,
6609 .la_sync_failures = lfsck_namespace_assistant_sync_failures,
6613 * Verify the specified linkEA entry for the given directory object.
6614 * If the object has no such linkEA entry or it has more other linkEA
6615 * entries, then re-generate the linkEA with the given information.
6617 * \param[in] env pointer to the thread context
6618 * \param[in] obj pointer to the dt_object to be handled
6619 * \param[in] cname the name for the child in the parent directory
6620 * \param[in] pfid the parent directory's FID for the linkEA
6622 * \retval 0 for success
6623 * \retval negative error number on failure
6625 int lfsck_verify_linkea(const struct lu_env *env, struct dt_object *obj,
6626 const struct lu_name *cname, const struct lu_fid *pfid)
6628 struct dt_device *dev = lfsck_obj2dev(obj);
6629 struct linkea_data ldata = { NULL };
6630 struct lu_buf linkea_buf;
6633 int fl = LU_XATTR_CREATE;
6637 LASSERT(S_ISDIR(lfsck_object_type(obj)));
6639 rc = lfsck_links_read_with_rec(env, obj, &ldata);
6640 if (rc == -ENODATA) {
6642 } else if (rc == 0) {
6643 fl = LU_XATTR_REPLACE;
6644 if (ldata.ld_leh->leh_reccount != 1) {
6647 rc = linkea_links_find(&ldata, cname, pfid);
6656 rc = linkea_links_new(&ldata, &lfsck_env_info(env)->lti_linkea_buf,
6661 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
6662 ldata.ld_leh->leh_len);
6663 th = dt_trans_create(env, dev);
6665 RETURN(PTR_ERR(th));
6667 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
6668 XATTR_NAME_LINK, fl, th);
6672 rc = dt_trans_start_local(env, dev, th);
6676 dt_write_lock(env, obj, 0);
6677 rc = dt_xattr_set(env, obj, &linkea_buf,
6678 XATTR_NAME_LINK, fl, th);
6679 dt_write_unlock(env, obj);
6684 dt_trans_stop(env, dev, th);
6689 * Get the name and parent directory's FID from the first linkEA entry.
6691 * \param[in] env pointer to the thread context
6692 * \param[in] obj pointer to the object which get linkEA from
6693 * \param[out] name pointer to the buffer to hold the name
6694 * in the first linkEA entry
6695 * \param[out] pfid pointer to the buffer to hold the parent
6696 * directory's FID in the first linkEA entry
6698 * \retval 0 for success
6699 * \retval negative error number on failure
6701 int lfsck_links_get_first(const struct lu_env *env, struct dt_object *obj,
6702 char *name, struct lu_fid *pfid)
6704 struct lu_name *cname = &lfsck_env_info(env)->lti_name;
6705 struct linkea_data ldata = { NULL };
6708 rc = lfsck_links_read_with_rec(env, obj, &ldata);
6712 linkea_first_entry(&ldata);
6713 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, cname, pfid);
6714 if (!linkea_entry_is_valid(&ldata, cname, pfid))
6717 /* To guarantee the 'name' is terminated with '0'. */
6718 memcpy(name, cname->ln_name, cname->ln_namelen);
6719 name[cname->ln_namelen] = 0;
6725 * Update the object's name entry with the given FID.
6727 * \param[in] env pointer to the thread context
6728 * \param[in] lfsck pointer to the lfsck instance
6729 * \param[in] dir pointer to the directory that holds
6731 * \param[in] name the name for the entry to be updated
6732 * \param[in] fid the new FID for the name entry referenced
6733 * \param[in] type the type for the name entry to be updated
6735 * \retval 0 for success
6736 * \retval negative error number on failure
6738 int lfsck_update_name_entry(const struct lu_env *env,
6739 struct lfsck_instance *lfsck,
6740 struct dt_object *dir, const char *name,
6741 const struct lu_fid *fid, __u32 type)
6743 struct lfsck_thread_info *info = lfsck_env_info(env);
6744 struct dt_insert_rec *rec = &info->lti_dt_rec;
6745 struct lfsck_lock_handle *llh = &info->lti_llh;
6746 struct dt_device *dev = lfsck_obj2dev(dir);
6752 rc = lfsck_lock(env, lfsck, dir, name, llh,
6753 MDS_INODELOCK_UPDATE, LCK_PW);
6757 th = dt_trans_create(env, dev);
6759 GOTO(unlock, rc = PTR_ERR(th));
6761 rc = dt_declare_delete(env, dir, (const struct dt_key *)name, th);
6765 rec->rec_type = type;
6767 rc = dt_declare_insert(env, dir, (const struct dt_rec *)rec,
6768 (const struct dt_key *)name, th);
6772 rc = dt_declare_ref_add(env, dir, th);
6776 rc = dt_trans_start_local(env, dev, th);
6780 rc = dt_delete(env, dir, (const struct dt_key *)name, th);
6781 if (rc == -ENOENT) {
6789 rc = dt_insert(env, dir, (const struct dt_rec *)rec,
6790 (const struct dt_key *)name, th);
6791 if (rc == 0 && S_ISDIR(type) && !exists) {
6792 dt_write_lock(env, dir, 0);
6793 rc = dt_ref_add(env, dir, th);
6794 dt_write_unlock(env, dir);
6800 dt_trans_stop(env, dev, th);
6804 CDEBUG(D_LFSCK, "%s: update name entry "DFID"/%s with the FID "DFID
6805 " and the type %o: rc = %d\n", lfsck_lfsck2name(lfsck),
6806 PFID(lfsck_dto2fid(dir)), name, PFID(fid), type, rc);
6811 int lfsck_namespace_setup(const struct lu_env *env,
6812 struct lfsck_instance *lfsck)
6814 struct lfsck_component *com;
6815 struct lfsck_namespace *ns;
6816 struct dt_object *root = NULL;
6817 struct dt_object *obj;
6822 LASSERT(lfsck->li_master);
6828 INIT_LIST_HEAD(&com->lc_link);
6829 INIT_LIST_HEAD(&com->lc_link_dir);
6830 init_rwsem(&com->lc_sem);
6831 atomic_set(&com->lc_ref, 1);
6832 com->lc_lfsck = lfsck;
6833 com->lc_type = LFSCK_TYPE_NAMESPACE;
6834 com->lc_ops = &lfsck_namespace_ops;
6835 com->lc_data = lfsck_assistant_data_init(
6836 &lfsck_namespace_assistant_ops,
6838 if (com->lc_data == NULL)
6839 GOTO(out, rc = -ENOMEM);
6841 com->lc_file_size = sizeof(struct lfsck_namespace);
6842 OBD_ALLOC(com->lc_file_ram, com->lc_file_size);
6843 if (com->lc_file_ram == NULL)
6844 GOTO(out, rc = -ENOMEM);
6846 OBD_ALLOC(com->lc_file_disk, com->lc_file_size);
6847 if (com->lc_file_disk == NULL)
6848 GOTO(out, rc = -ENOMEM);
6850 for (i = 0; i < LFSCK_STF_COUNT; i++)
6851 mutex_init(&com->lc_sub_trace_objs[i].lsto_mutex);
6853 root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
6855 GOTO(out, rc = PTR_ERR(root));
6857 if (unlikely(!dt_try_as_dir(env, root)))
6858 GOTO(out, rc = -ENOTDIR);
6860 obj = local_index_find_or_create(env, lfsck->li_los, root,
6862 S_IFREG | S_IRUGO | S_IWUSR,
6863 &dt_lfsck_namespace_features);
6865 GOTO(out, rc = PTR_ERR(obj));
6868 rc = lfsck_namespace_load(env, com);
6869 if (rc == -ENODATA) {
6870 rc = lfsck_namespace_init(env, com);
6871 } else if (rc < 0) {
6872 rc = lfsck_namespace_reset(env, com, true);
6874 rc = lfsck_load_sub_trace_files(env, com,
6875 &dt_lfsck_namespace_features, LFSCK_NAMESPACE, false);
6877 rc = lfsck_namespace_reset(env, com, true);
6882 ns = com->lc_file_ram;
6883 switch (ns->ln_status) {
6888 spin_lock(&lfsck->li_lock);
6889 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
6890 spin_unlock(&lfsck->li_lock);
6893 CERROR("%s: unknown lfsck_namespace status %d\n",
6894 lfsck_lfsck2name(lfsck), ns->ln_status);
6896 case LS_SCANNING_PHASE1:
6897 case LS_SCANNING_PHASE2:
6898 /* No need to store the status to disk right now.
6899 * If the system crashed before the status stored,
6900 * it will be loaded back when next time. */
6901 ns->ln_status = LS_CRASHED;
6905 spin_lock(&lfsck->li_lock);
6906 list_add_tail(&com->lc_link, &lfsck->li_list_scan);
6907 list_add_tail(&com->lc_link_dir, &lfsck->li_list_dir);
6908 spin_unlock(&lfsck->li_lock);
6915 if (root != NULL && !IS_ERR(root))
6916 lfsck_object_put(env, root);
6918 lfsck_component_cleanup(env, com);
6919 CERROR("%s: fail to init namespace LFSCK component: rc = %d\n",
6920 lfsck_lfsck2name(lfsck), rc);