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 lad->lad_incomplete = 0;
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 lad->lad_incomplete = 0;
331 lad->lad_incomplete = 1;
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, 1);
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 int lfsck_namespace_unpack_linkea_entry(struct linkea_data *ldata,
764 struct lu_name *cname,
766 char *buf, const int buflen)
768 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, cname, pfid);
769 if (unlikely(ldata->ld_reclen <= 0 ||
770 ldata->ld_reclen + sizeof(struct link_ea_header) >
771 ldata->ld_leh->leh_len ||
772 cname->ln_namelen <= 0 ||
773 cname->ln_namelen > NAME_MAX ||
774 cname->ln_namelen >= buflen ||
778 /* To guarantee the 'name' is terminated with '0'. */
779 memcpy(buf, cname->ln_name, cname->ln_namelen);
780 buf[cname->ln_namelen] = 0;
781 cname->ln_name = buf;
786 static void lfsck_linkea_del_buf(struct linkea_data *ldata,
787 const struct lu_name *lname)
789 LASSERT(ldata->ld_leh != NULL && ldata->ld_lee != NULL);
791 /* If current record is corrupted, all the subsequent
792 * records will be dropped. */
793 if (unlikely(ldata->ld_reclen <= 0 ||
794 ldata->ld_reclen + sizeof(struct link_ea_header) >
795 ldata->ld_leh->leh_len)) {
796 void *ptr = ldata->ld_lee;
798 ldata->ld_leh->leh_len = sizeof(struct link_ea_header);
799 ldata->ld_leh->leh_reccount = 0;
800 linkea_first_entry(ldata);
801 while (ldata->ld_lee != NULL &&
802 (char *)ldata->ld_lee < (char *)ptr) {
803 int reclen = (ldata->ld_lee->lee_reclen[0] << 8) |
804 ldata->ld_lee->lee_reclen[1];
806 ldata->ld_leh->leh_len += reclen;
807 ldata->ld_leh->leh_reccount++;
808 ldata->ld_lee = (struct link_ea_entry *)
809 ((char *)ldata->ld_lee + reclen);
812 ldata->ld_lee = NULL;
814 linkea_del_buf(ldata, lname);
818 static int lfsck_namespace_filter_linkea_entry(struct linkea_data *ldata,
819 struct lu_name *cname,
823 struct link_ea_entry *oldlee;
827 oldlee = ldata->ld_lee;
828 oldlen = ldata->ld_reclen;
829 linkea_next_entry(ldata);
830 while (ldata->ld_lee != NULL) {
831 ldata->ld_reclen = (ldata->ld_lee->lee_reclen[0] << 8) |
832 ldata->ld_lee->lee_reclen[1];
833 if (unlikely(ldata->ld_reclen == oldlen &&
834 memcmp(ldata->ld_lee, oldlee, oldlen) == 0)) {
839 lfsck_linkea_del_buf(ldata, cname);
841 linkea_next_entry(ldata);
844 ldata->ld_lee = oldlee;
845 ldata->ld_reclen = oldlen;
851 * Insert orphan into .lustre/lost+found/MDTxxxx/ locally.
853 * Add the specified orphan MDT-object to the .lustre/lost+found/MDTxxxx/
854 * with the given type to generate the name, the detailed rules for name
855 * have been described as following.
857 * The function also generates the linkEA corresponding to the name entry
858 * under the .lustre/lost+found/MDTxxxx/ for the orphan MDT-object.
860 * \param[in] env pointer to the thread context
861 * \param[in] com pointer to the lfsck component
862 * \param[in] orphan pointer to the orphan MDT-object
863 * \param[in] infix additional information for the orphan name, such as
864 * the FID for original
865 * \param[in] type the type for describing why the orphan MDT-object is
866 * created. The rules are as following:
868 * type "D": The MDT-object is a directory, it may knows its parent
869 * but because there is no valid linkEA, the LFSCK cannot
870 * know where to put it back to the namespace.
871 * type "O": The MDT-object has no linkEA, and there is no name
872 * entry that references the MDT-object.
874 * type "S": The orphan MDT-object is a shard of a striped directory
876 * \see lfsck_layout_recreate_parent() for more types.
878 * The orphan name will be like:
879 * ${FID}-${infix}-${type}-${conflict_version}
881 * \param[out] count if some others inserted some linkEA entries by race,
882 * then return the linkEA entries count.
884 * \retval positive number for repaired cases
885 * \retval 0 if needs to repair nothing
886 * \retval negative error number on failure
888 static int lfsck_namespace_insert_orphan(const struct lu_env *env,
889 struct lfsck_component *com,
890 struct dt_object *orphan,
891 const char *infix, const char *type,
894 struct lfsck_thread_info *info = lfsck_env_info(env);
895 struct lu_name *cname = &info->lti_name;
896 struct dt_insert_rec *rec = &info->lti_dt_rec;
897 struct lu_attr *la = &info->lti_la2;
898 const struct lu_fid *cfid = lfsck_dto2fid(orphan);
899 const struct lu_fid *pfid;
901 struct lfsck_instance *lfsck = com->lc_lfsck;
902 struct dt_device *dev = lfsck_obj2dev(orphan);
903 struct dt_object *parent;
904 struct thandle *th = NULL;
905 struct lfsck_lock_handle *pllh = &info->lti_llh;
906 struct lustre_handle clh = { 0 };
907 struct linkea_data ldata2 = { NULL };
908 struct lu_buf linkea_buf;
915 cname->ln_name = NULL;
916 if (unlikely(lfsck->li_lpf_obj == NULL))
917 GOTO(log, rc = -ENXIO);
919 parent = lfsck->li_lpf_obj;
920 pfid = lfsck_dto2fid(parent);
924 namelen = snprintf(info->lti_key, NAME_MAX, DFID"%s-%s-%d",
925 PFID(cfid), infix, type, idx++);
926 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
927 (const struct dt_key *)info->lti_key);
928 if (rc != 0 && rc != -ENOENT)
931 if (unlikely(rc == 0 && lu_fid_eq(cfid, &tfid)))
933 } while (rc == 0 && !exist);
935 rc = lfsck_lock(env, lfsck, parent, info->lti_key, pllh,
936 MDS_INODELOCK_UPDATE, LCK_PW);
940 /* Re-check whether the name conflict with othrs after taken
942 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
943 (const struct dt_key *)info->lti_key);
945 if (!lu_fid_eq(cfid, &tfid)) {
952 } else if (rc != -ENOENT) {
958 cname->ln_name = info->lti_key;
959 cname->ln_namelen = namelen;
960 rc = linkea_links_new(&ldata2, &info->lti_linkea_buf2,
965 rc = lfsck_ibits_lock(env, lfsck, orphan, &clh,
966 MDS_INODELOCK_UPDATE | MDS_INODELOCK_LOOKUP |
967 MDS_INODELOCK_XATTR, LCK_EX);
971 lfsck_buf_init(&linkea_buf, ldata2.ld_buf->lb_buf,
972 ldata2.ld_leh->leh_len);
973 th = dt_trans_create(env, dev);
975 GOTO(log, rc = PTR_ERR(th));
977 if (S_ISDIR(lfsck_object_type(orphan))) {
978 rc = dt_declare_delete(env, orphan,
979 (const struct dt_key *)dotdot, th);
983 rec->rec_type = S_IFDIR;
985 rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
986 (const struct dt_key *)dotdot, th);
991 rc = dt_declare_xattr_set(env, orphan, &linkea_buf,
992 XATTR_NAME_LINK, 0, th);
997 rec->rec_type = lfsck_object_type(orphan) & S_IFMT;
999 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1000 (const struct dt_key *)cname->ln_name,
1005 if (S_ISDIR(rec->rec_type)) {
1006 rc = dt_declare_ref_add(env, parent, th);
1012 memset(la, 0, sizeof(*la));
1013 la->la_ctime = ktime_get_real_seconds();
1014 la->la_valid = LA_CTIME;
1015 rc = dt_declare_attr_set(env, orphan, la, th);
1019 rc = dt_trans_start_local(env, dev, th);
1023 dt_write_lock(env, orphan, 0);
1024 rc = lfsck_links_read2_with_rec(env, orphan, &ldata2);
1025 if (likely(rc == -ENODATA || rc == -EINVAL)) {
1026 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1027 GOTO(unlock, rc = 1);
1029 if (S_ISDIR(lfsck_object_type(orphan))) {
1030 rc = dt_delete(env, orphan,
1031 (const struct dt_key *)dotdot, th);
1035 rec->rec_type = S_IFDIR;
1036 rec->rec_fid = pfid;
1037 rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1038 (const struct dt_key *)dotdot, th, 1);
1043 rc = dt_xattr_set(env, orphan, &linkea_buf, XATTR_NAME_LINK, 0,
1046 if (rc == 0 && count != NULL)
1047 *count = ldata2.ld_leh->leh_reccount;
1051 dt_write_unlock(env, orphan);
1053 if (rc == 0 && !exist) {
1054 rec->rec_type = lfsck_object_type(orphan) & S_IFMT;
1055 rec->rec_fid = cfid;
1056 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1057 (const struct dt_key *)cname->ln_name, th, 1);
1058 if (rc == 0 && S_ISDIR(rec->rec_type)) {
1059 dt_write_lock(env, parent, 0);
1060 rc = dt_ref_add(env, parent, th);
1061 dt_write_unlock(env, parent);
1066 rc = dt_attr_set(env, orphan, la, th);
1068 GOTO(stop, rc = (rc == 0 ? 1 : rc));
1071 dt_write_unlock(env, orphan);
1074 dt_trans_stop(env, dev, th);
1077 lfsck_ibits_unlock(&clh, LCK_EX);
1079 CDEBUG(D_LFSCK, "%s: namespace LFSCK insert orphan for the "
1080 "object "DFID", name = %s: rc = %d\n",
1081 lfsck_lfsck2name(lfsck), PFID(cfid),
1082 cname->ln_name != NULL ? cname->ln_name : "<NULL>", rc);
1085 struct lfsck_namespace *ns = com->lc_file_ram;
1087 ns->ln_flags |= LF_INCONSISTENT;
1094 * Add the specified name entry back to namespace.
1096 * If there is a linkEA entry that back references a name entry under
1097 * some parent directory, but such parent directory does not have the
1098 * claimed name entry. On the other hand, the linkEA entries count is
1099 * not larger than the MDT-object's hard link count. Under such case,
1100 * it is quite possible that the name entry is lost. Then the LFSCK
1101 * should add the name entry back to the namespace.
1103 * \param[in] env pointer to the thread context
1104 * \param[in] com pointer to the lfsck component
1105 * \param[in] parent pointer to the directory under which the name entry
1106 * will be inserted into
1107 * \param[in] child pointer to the object referenced by the name entry
1108 * that to be inserted into the parent
1109 * \param[in] name the name for the child in the parent directory
1111 * \retval positive number for repaired cases
1112 * \retval 0 if nothing to be repaired
1113 * \retval negative error number on failure
1115 static int lfsck_namespace_insert_normal(const struct lu_env *env,
1116 struct lfsck_component *com,
1117 struct dt_object *parent,
1118 struct dt_object *child,
1121 struct lfsck_thread_info *info = lfsck_env_info(env);
1122 struct lu_attr *la = &info->lti_la;
1123 struct dt_insert_rec *rec = &info->lti_dt_rec;
1124 struct lfsck_instance *lfsck = com->lc_lfsck;
1125 /* The child and its name may be on different MDTs. */
1126 const struct lu_fid *pfid = lfsck_dto2fid(parent);
1127 const struct lu_fid *cfid = lfsck_dto2fid(child);
1128 struct dt_device *dev = lfsck->li_next;
1129 struct thandle *th = NULL;
1130 struct lfsck_lock_handle *llh = &info->lti_llh;
1134 /* @parent/@child may be based on lfsck->li_bottom,
1135 * but here we need the object based on the lfsck->li_next. */
1137 parent = lfsck_object_locate(dev, parent);
1139 GOTO(log, rc = PTR_ERR(parent));
1141 if (unlikely(!dt_try_as_dir(env, parent)))
1142 GOTO(log, rc = -ENOTDIR);
1144 child = lfsck_object_locate(dev, child);
1146 GOTO(log, rc = PTR_ERR(child));
1148 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1151 rc = lfsck_lock(env, lfsck, parent, name, llh,
1152 MDS_INODELOCK_UPDATE, LCK_PW);
1156 th = dt_trans_create(env, dev);
1158 GOTO(unlock, rc = PTR_ERR(th));
1160 rec->rec_type = lfsck_object_type(child) & S_IFMT;
1161 rec->rec_fid = cfid;
1162 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1163 (const struct dt_key *)name, th);
1167 if (S_ISDIR(rec->rec_type)) {
1168 rc = dt_declare_ref_add(env, parent, th);
1173 memset(la, 0, sizeof(*la));
1174 la->la_ctime = ktime_get_real_seconds();
1175 la->la_valid = LA_CTIME;
1176 rc = dt_declare_attr_set(env, parent, la, th);
1180 rc = dt_declare_attr_set(env, child, la, th);
1184 rc = dt_trans_start_local(env, dev, th);
1188 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1189 (const struct dt_key *)name, th, 1);
1193 if (S_ISDIR(rec->rec_type)) {
1194 dt_write_lock(env, parent, 0);
1195 rc = dt_ref_add(env, parent, th);
1196 dt_write_unlock(env, parent);
1201 la->la_ctime = ktime_get_real_seconds();
1202 rc = dt_attr_set(env, parent, la, th);
1206 rc = dt_attr_set(env, child, la, th);
1208 GOTO(stop, rc = (rc == 0 ? 1 : rc));
1211 dt_trans_stop(env, dev, th);
1217 CDEBUG(D_LFSCK, "%s: namespace LFSCK insert object "DFID" with "
1218 "the name %s and type %o to the parent "DFID": rc = %d\n",
1219 lfsck_lfsck2name(lfsck), PFID(cfid), name,
1220 lfsck_object_type(child) & S_IFMT, PFID(pfid), rc);
1223 struct lfsck_namespace *ns = com->lc_file_ram;
1225 ns->ln_flags |= LF_INCONSISTENT;
1227 ns->ln_lost_dirent_repaired++;
1234 * Create the specified orphan directory.
1236 * For the case that the parent MDT-object stored in some MDT-object's
1237 * linkEA entry is lost, the LFSCK will re-create the parent object as
1238 * an orphan and insert it into .lustre/lost+found/MDTxxxx/ directory
1239 * with the name ${FID}-P-${conflict_version}.
1241 * \param[in] env pointer to the thread context
1242 * \param[in] com pointer to the lfsck component
1243 * \param[in] orphan pointer to the orphan MDT-object to be created
1244 * \param[in] lmv pointer to master LMV EA that will be set to the orphan
1246 * \retval positive number for repaired cases
1247 * \retval negative error number on failure
1249 static int lfsck_namespace_create_orphan_dir(const struct lu_env *env,
1250 struct lfsck_component *com,
1251 struct dt_object *orphan,
1252 struct lmv_mds_md_v1 *lmv)
1254 struct lfsck_thread_info *info = lfsck_env_info(env);
1255 struct lu_attr *la = &info->lti_la;
1256 struct dt_allocation_hint *hint = &info->lti_hint;
1257 struct dt_object_format *dof = &info->lti_dof;
1258 struct lu_name *cname = &info->lti_name2;
1259 struct dt_insert_rec *rec = &info->lti_dt_rec;
1260 struct lmv_mds_md_v1 *lmv2 = &info->lti_lmv2;
1261 const struct lu_fid *cfid = lfsck_dto2fid(orphan);
1263 struct lfsck_instance *lfsck = com->lc_lfsck;
1264 struct lfsck_namespace *ns = com->lc_file_ram;
1265 struct dt_device *dev = lfsck_obj2dev(orphan);
1266 struct dt_object *parent = NULL;
1267 struct thandle *th = NULL;
1268 struct lfsck_lock_handle *llh = &info->lti_llh;
1269 struct linkea_data ldata = { NULL };
1270 struct lu_buf linkea_buf;
1271 struct lu_buf lmv_buf;
1279 LASSERT(!dt_object_exists(orphan));
1281 cname->ln_name = NULL;
1282 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1285 if (dt_object_remote(orphan)) {
1286 LASSERT(lfsck->li_lpf_root_obj != NULL);
1288 idx = lfsck_find_mdt_idx_by_fid(env, lfsck, cfid);
1290 GOTO(log, rc = idx);
1292 snprintf(name, 8, "MDT%04x", idx);
1293 rc = dt_lookup(env, lfsck->li_lpf_root_obj,
1294 (struct dt_rec *)&tfid,
1295 (const struct dt_key *)name);
1297 GOTO(log, rc = (rc == -ENOENT ? -ENXIO : rc));
1299 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
1301 GOTO(log, rc = PTR_ERR(parent));
1303 if (unlikely(!dt_try_as_dir(env, parent)))
1304 GOTO(log, rc = -ENOTDIR);
1306 if (unlikely(lfsck->li_lpf_obj == NULL))
1307 GOTO(log, rc = -ENXIO);
1309 parent = lfsck->li_lpf_obj;
1312 dev = lfsck_find_dev_by_fid(env, lfsck, cfid);
1314 GOTO(log, rc = PTR_ERR(dev));
1320 namelen = snprintf(name, 31, DFID"-P-%d",
1322 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
1323 (const struct dt_key *)name);
1324 if (rc != 0 && rc != -ENOENT)
1328 rc = lfsck_lock(env, lfsck, parent, name, llh,
1329 MDS_INODELOCK_UPDATE, LCK_PW);
1333 /* Re-check whether the name conflict with othrs after taken
1335 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
1336 (const struct dt_key *)name);
1337 if (unlikely(rc == 0)) {
1345 cname->ln_name = name;
1346 cname->ln_namelen = namelen;
1348 memset(la, 0, sizeof(*la));
1349 la->la_mode = S_IFDIR | 0700;
1350 la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
1351 LA_ATIME | LA_MTIME | LA_CTIME;
1353 orphan->do_ops->do_ah_init(env, hint, parent, orphan,
1354 la->la_mode & S_IFMT);
1356 memset(dof, 0, sizeof(*dof));
1357 dof->dof_type = dt_mode_to_dft(S_IFDIR);
1359 rc = linkea_links_new(&ldata, &info->lti_linkea_buf2,
1360 cname, lfsck_dto2fid(parent));
1364 th = dt_trans_create(env, dev);
1366 GOTO(unlock1, rc = PTR_ERR(th));
1368 /* Sync the remote transaction to guarantee that the subsequent
1369 * lock against the @orphan can find the @orphan in time. */
1370 if (dt_object_remote(orphan))
1373 rc = dt_declare_create(env, orphan, la, hint, dof, th);
1377 if (unlikely(!dt_try_as_dir(env, orphan)))
1378 GOTO(stop, rc = -ENOTDIR);
1380 rc = dt_declare_ref_add(env, orphan, th);
1384 rec->rec_type = S_IFDIR;
1385 rec->rec_fid = cfid;
1386 rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
1387 (const struct dt_key *)dot, th);
1391 rec->rec_fid = lfsck_dto2fid(parent);
1392 rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
1393 (const struct dt_key *)dotdot, th);
1398 lmv->lmv_magic = LMV_MAGIC;
1399 lmv->lmv_master_mdt_index = lfsck_dev_idx(lfsck);
1400 lfsck_lmv_header_cpu_to_le(lmv2, lmv);
1401 lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
1402 rc = dt_declare_xattr_set(env, orphan, &lmv_buf,
1403 XATTR_NAME_LMV, 0, th);
1408 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
1409 ldata.ld_leh->leh_len);
1410 rc = dt_declare_xattr_set(env, orphan, &linkea_buf,
1411 XATTR_NAME_LINK, 0, th);
1415 rec->rec_fid = cfid;
1416 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1417 (const struct dt_key *)name, th);
1419 rc = dt_declare_ref_add(env, parent, th);
1424 rc = dt_trans_start_local(env, dev, th);
1428 dt_write_lock(env, orphan, 0);
1429 rc = dt_create(env, orphan, la, hint, dof, th);
1433 rc = dt_ref_add(env, orphan, th);
1437 rec->rec_fid = cfid;
1438 rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1439 (const struct dt_key *)dot, th, 1);
1443 rec->rec_fid = lfsck_dto2fid(parent);
1444 rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1445 (const struct dt_key *)dotdot, th, 1);
1450 rc = dt_xattr_set(env, orphan, &lmv_buf, XATTR_NAME_LMV, 0, th);
1455 rc = dt_xattr_set(env, orphan, &linkea_buf,
1456 XATTR_NAME_LINK, 0, th);
1457 dt_write_unlock(env, orphan);
1461 rec->rec_fid = cfid;
1462 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1463 (const struct dt_key *)name, th, 1);
1465 dt_write_lock(env, parent, 0);
1466 rc = dt_ref_add(env, parent, th);
1467 dt_write_unlock(env, parent);
1470 GOTO(stop, rc = (rc == 0 ? 1 : rc));
1473 dt_write_unlock(env, orphan);
1476 rc1 = dt_trans_stop(env, dev, th);
1477 if (rc1 != 0 && rc > 0)
1484 CDEBUG(D_LFSCK, "%s: namespace LFSCK create orphan dir for "
1485 "the object "DFID", name = %s: rc = %d\n",
1486 lfsck_lfsck2name(lfsck), PFID(cfid),
1487 cname->ln_name != NULL ? cname->ln_name : "<NULL>", rc);
1489 if (parent != NULL && !IS_ERR(parent) && parent != lfsck->li_lpf_obj)
1490 lfsck_object_put(env, parent);
1493 ns->ln_flags |= LF_INCONSISTENT;
1499 * Remove the specified entry from the linkEA.
1501 * Locate the linkEA entry with the given @cname and @pfid, then
1502 * remove this entry or the other entries those are repeated with
1505 * \param[in] env pointer to the thread context
1506 * \param[in] com pointer to the lfsck component
1507 * \param[in] obj pointer to the dt_object to be handled
1508 * \param[in,out]ldata pointer to the buffer that holds the linkEA
1509 * \param[in] cname the name for the child in the parent directory
1510 * \param[in] pfid the parent directory's FID for the linkEA
1511 * \param[in] next if true, then remove the first found linkEA
1512 * entry, and move the ldata->ld_lee to next entry
1514 * \retval positive number for repaired cases
1515 * \retval 0 if nothing to be repaired
1516 * \retval negative error number on failure
1518 static int lfsck_namespace_shrink_linkea(const struct lu_env *env,
1519 struct lfsck_component *com,
1520 struct dt_object *obj,
1521 struct linkea_data *ldata,
1522 struct lu_name *cname,
1523 struct lu_fid *pfid,
1526 struct lfsck_instance *lfsck = com->lc_lfsck;
1527 struct dt_device *dev = lfsck_obj2dev(obj);
1528 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
1529 struct thandle *th = NULL;
1530 struct lustre_handle lh = { 0 };
1531 struct linkea_data ldata_new = { NULL };
1532 struct lu_buf linkea_buf;
1537 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
1538 MDS_INODELOCK_UPDATE | MDS_INODELOCK_XATTR,
1544 lfsck_linkea_del_buf(ldata, cname);
1546 lfsck_namespace_filter_linkea_entry(ldata, cname, pfid,
1548 if (ldata->ld_leh->leh_reccount > 0 ||
1549 unlikely(ldata->ld_leh->leh_overflow_time)) {
1550 lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
1551 ldata->ld_leh->leh_len);
1552 buflen = linkea_buf.lb_len;
1556 th = dt_trans_create(env, dev);
1558 GOTO(unlock1, rc = PTR_ERR(th));
1561 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
1562 XATTR_NAME_LINK, 0, th);
1564 rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, th);
1568 rc = dt_trans_start_local(env, dev, th);
1572 dt_write_lock(env, obj, 0);
1573 if (unlikely(lfsck_is_dead_obj(obj)))
1574 GOTO(unlock2, rc = -ENOENT);
1576 rc = lfsck_links_read2_with_rec(env, obj, &ldata_new);
1578 GOTO(unlock2, rc = (rc == -ENODATA ? 0 : rc));
1580 /* The specified linkEA entry has been removed by race. */
1581 rc = linkea_links_find(&ldata_new, cname, pfid);
1583 GOTO(unlock2, rc = 0);
1585 if (bk->lb_param & LPF_DRYRUN)
1586 GOTO(unlock2, rc = 1);
1589 lfsck_linkea_del_buf(&ldata_new, cname);
1591 lfsck_namespace_filter_linkea_entry(&ldata_new, cname, pfid,
1594 if (buflen < ldata_new.ld_leh->leh_len) {
1595 dt_write_unlock(env, obj);
1596 dt_trans_stop(env, dev, th);
1597 lfsck_buf_init(&linkea_buf, ldata_new.ld_buf->lb_buf,
1598 ldata_new.ld_leh->leh_len);
1602 if (ldata_new.ld_leh->leh_reccount > 0 ||
1603 unlikely(ldata->ld_leh->leh_overflow_time))
1604 rc = lfsck_links_write(env, obj, &ldata_new, th);
1606 rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, th);
1608 GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
1611 dt_write_unlock(env, obj);
1614 dt_trans_stop(env, dev, th);
1617 lfsck_ibits_unlock(&lh, LCK_EX);
1620 CDEBUG(D_LFSCK, "%s: namespace LFSCK remove %s linkEA entry "
1621 "for the object: "DFID", parent "DFID", name %.*s\n",
1622 lfsck_lfsck2name(lfsck), next ? "invalid" : "redundant",
1623 PFID(lfsck_dto2fid(obj)), PFID(pfid), cname->ln_namelen,
1627 struct lfsck_namespace *ns = com->lc_file_ram;
1629 ns->ln_flags |= LF_INCONSISTENT;
1636 * Conditionally remove the specified entry from the linkEA.
1638 * Take the parent lock firstly, then check whether the specified
1639 * name entry exists or not: if yes, do nothing; otherwise, call
1640 * lfsck_namespace_shrink_linkea() to remove the linkea entry.
1642 * \param[in] env pointer to the thread context
1643 * \param[in] com pointer to the lfsck component
1644 * \param[in] parent pointer to the parent directory
1645 * \param[in] child pointer to the child object that holds the linkEA
1646 * \param[in,out]ldata pointer to the buffer that holds the linkEA
1647 * \param[in] cname the name for the child in the parent directory
1648 * \param[in] pfid the parent directory's FID for the linkEA
1650 * \retval positive number for repaired cases
1651 * \retval 0 if nothing to be repaired
1652 * \retval negative error number on failure
1654 static int lfsck_namespace_shrink_linkea_cond(const struct lu_env *env,
1655 struct lfsck_component *com,
1656 struct dt_object *parent,
1657 struct dt_object *child,
1658 struct linkea_data *ldata,
1659 struct lu_name *cname,
1660 struct lu_fid *pfid)
1662 struct lfsck_thread_info *info = lfsck_env_info(env);
1663 struct lu_fid *cfid = &info->lti_fid3;
1664 struct lfsck_lock_handle *llh = &info->lti_llh;
1668 rc = lfsck_lock(env, com->lc_lfsck, parent, cname->ln_name, llh,
1669 MDS_INODELOCK_UPDATE, LCK_PR);
1673 dt_read_lock(env, parent, 0);
1674 if (unlikely(lfsck_is_dead_obj(parent))) {
1675 dt_read_unlock(env, parent);
1677 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
1683 rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
1684 (const struct dt_key *)cname->ln_name);
1685 dt_read_unlock(env, parent);
1687 /* It is safe to release the ldlm lock, because when the logic come
1688 * here, we have got all the needed information above whether the
1689 * linkEA entry is valid or not. It is not important that others
1690 * may add new linkEA entry after the ldlm lock released. If other
1691 * has removed the specified linkEA entry by race, then it is OK,
1692 * because the subsequent lfsck_namespace_shrink_linkea() can handle
1695 if (rc == -ENOENT) {
1696 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
1705 /* The LFSCK just found some internal status of cross-MDTs
1706 * create operation. That is normal. */
1707 if (lu_fid_eq(cfid, lfsck_dto2fid(child))) {
1708 linkea_next_entry(ldata);
1713 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata, cname,
1720 * Conditionally replace name entry in the parent.
1722 * As required, the LFSCK may re-create the lost MDT-object for dangling
1723 * name entry, but such repairing may be wrong because of bad FID in the
1724 * name entry. As the LFSCK processing, the real MDT-object may be found,
1725 * then the LFSCK should check whether the former re-created MDT-object
1726 * has been modified or not, if not, then destroy it and update the name
1727 * entry in the parent to reference the real MDT-object.
1729 * \param[in] env pointer to the thread context
1730 * \param[in] com pointer to the lfsck component
1731 * \param[in] parent pointer to the parent directory
1732 * \param[in] child pointer to the MDT-object that may be the real
1733 * MDT-object corresponding to the name entry in parent
1734 * \param[in] cfid the current FID in the name entry
1735 * \param[in] cname contains the name of the child in the parent directory
1737 * \retval positive number for repaired cases
1738 * \retval 0 if nothing to be repaired
1739 * \retval negative error number on failure
1741 static int lfsck_namespace_replace_cond(const struct lu_env *env,
1742 struct lfsck_component *com,
1743 struct dt_object *parent,
1744 struct dt_object *child,
1745 const struct lu_fid *cfid,
1746 const struct lu_name *cname)
1748 struct lfsck_thread_info *info = lfsck_env_info(env);
1749 struct lu_attr *la = &info->lti_la;
1750 struct dt_insert_rec *rec = &info->lti_dt_rec;
1752 struct lfsck_instance *lfsck = com->lc_lfsck;
1753 /* The child and its name may be on different MDTs. */
1754 struct dt_device *dev = lfsck->li_next;
1755 const char *name = cname->ln_name;
1756 const struct lu_fid *pfid = lfsck_dto2fid(parent);
1757 struct dt_object *cobj = NULL;
1758 struct lfsck_lock_handle *pllh = &info->lti_llh;
1759 struct lustre_handle clh = { 0 };
1760 struct linkea_data ldata = { NULL };
1761 struct thandle *th = NULL;
1766 /* @parent/@child may be based on lfsck->li_bottom,
1767 * but here we need the object based on the lfsck->li_next. */
1769 parent = lfsck_object_locate(dev, parent);
1771 GOTO(log, rc = PTR_ERR(parent));
1773 if (unlikely(!dt_try_as_dir(env, parent)))
1774 GOTO(log, rc = -ENOTDIR);
1776 rc = lfsck_lock(env, lfsck, parent, name, pllh,
1777 MDS_INODELOCK_UPDATE, LCK_PW);
1781 if (!fid_is_sane(cfid)) {
1786 cobj = lfsck_object_find_by_dev(env, dev, cfid);
1789 if (rc == -ENOENT) {
1797 if (!dt_object_exists(cobj)) {
1802 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
1803 (const struct dt_key *)name);
1804 if (rc == -ENOENT) {
1812 /* Someone changed the name entry, cannot replace it. */
1813 if (!lu_fid_eq(cfid, &tfid))
1816 /* lock the object to be destroyed. */
1817 rc = lfsck_ibits_lock(env, lfsck, cobj, &clh,
1818 MDS_INODELOCK_UPDATE |
1819 MDS_INODELOCK_UPDATE | MDS_INODELOCK_XATTR,
1824 if (unlikely(lfsck_is_dead_obj(cobj))) {
1829 rc = dt_attr_get(env, cobj, la);
1833 /* The object has been modified by other(s), or it is not created by
1834 * LFSCK, the two cases are indistinguishable. So cannot replace it. */
1835 if (la->la_ctime != 0)
1838 if (S_ISREG(la->la_mode)) {
1839 rc = dt_xattr_get(env, cobj, &LU_BUF_NULL, XATTR_NAME_LOV);
1840 /* If someone has created related OST-object(s),
1842 if ((rc > 0) || (rc < 0 && rc != -ENODATA))
1843 GOTO(log, rc = (rc > 0 ? 0 : rc));
1847 dt_read_lock(env, child, 0);
1848 rc = lfsck_links_read2_with_rec(env, child, &ldata);
1849 dt_read_unlock(env, child);
1851 /* Someone changed the child, no need to replace. */
1858 rc = linkea_links_find(&ldata, cname, pfid);
1859 /* Someone moved the child, no need to replace. */
1863 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1866 th = dt_trans_create(env, dev);
1868 GOTO(log, rc = PTR_ERR(th));
1871 rc = dt_declare_destroy(env, cobj, th);
1876 rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
1880 rec->rec_type = S_IFDIR;
1881 rec->rec_fid = lfsck_dto2fid(child);
1882 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1883 (const struct dt_key *)name, th);
1887 rc = dt_trans_start_local(env, dev, th);
1892 rc = dt_destroy(env, cobj, th);
1897 /* The old name entry maybe not exist. */
1898 rc = dt_delete(env, parent, (const struct dt_key *)name, th);
1899 if (rc != 0 && rc != -ENOENT)
1902 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1903 (const struct dt_key *)name, th, 1);
1905 GOTO(stop, rc = (rc == 0 ? 1 : rc));
1908 dt_trans_stop(env, dev, th);
1911 lfsck_ibits_unlock(&clh, LCK_EX);
1914 if (cobj != NULL && !IS_ERR(cobj))
1915 lfsck_object_put(env, cobj);
1917 CDEBUG(D_LFSCK, "%s: namespace LFSCK conditionally destroy the "
1918 "object "DFID" because of conflict with the object "DFID
1919 " under the parent "DFID" with name %s: rc = %d\n",
1920 lfsck_lfsck2name(lfsck), PFID(cfid),
1921 PFID(lfsck_dto2fid(child)), PFID(pfid), name, rc);
1927 * Overwrite the linkEA for the object with the given ldata.
1929 * The caller should take the ldlm lock before the calling.
1931 * \param[in] env pointer to the thread context
1932 * \param[in] com pointer to the lfsck component
1933 * \param[in] obj pointer to the dt_object to be handled
1934 * \param[in] ldata pointer to the new linkEA data
1936 * \retval positive number for repaired cases
1937 * \retval 0 if nothing to be repaired
1938 * \retval negative error number on failure
1940 int lfsck_namespace_rebuild_linkea(const struct lu_env *env,
1941 struct lfsck_component *com,
1942 struct dt_object *obj,
1943 struct linkea_data *ldata)
1945 struct lfsck_instance *lfsck = com->lc_lfsck;
1946 struct dt_device *dev = lfsck_obj2dev(obj);
1947 struct thandle *th = NULL;
1948 struct lu_buf linkea_buf;
1952 th = dt_trans_create(env, dev);
1954 GOTO(log, rc = PTR_ERR(th));
1956 lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
1957 ldata->ld_leh->leh_len);
1958 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
1959 XATTR_NAME_LINK, 0, th);
1963 rc = dt_trans_start_local(env, dev, th);
1967 dt_write_lock(env, obj, 0);
1968 if (unlikely(lfsck_is_dead_obj(obj)))
1969 GOTO(unlock, rc = 0);
1971 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1972 GOTO(unlock, rc = 1);
1974 rc = dt_xattr_set(env, obj, &linkea_buf,
1975 XATTR_NAME_LINK, 0, th);
1977 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
1980 dt_write_unlock(env, obj);
1983 dt_trans_stop(env, dev, th);
1986 CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild linkEA for the "
1987 "object "DFID": rc = %d\n",
1988 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
1991 struct lfsck_namespace *ns = com->lc_file_ram;
1993 ns->ln_flags |= LF_INCONSISTENT;
2000 * Repair invalid name entry.
2002 * If the name entry contains invalid information, such as bad file type
2003 * or (and) corrupted object FID, then either remove the name entry or
2004 * udpate the name entry with the given (right) information.
2006 * \param[in] env pointer to the thread context
2007 * \param[in] com pointer to the lfsck component
2008 * \param[in] parent pointer to the parent directory
2009 * \param[in] child pointer to the object referenced by the name entry
2010 * \param[in] name the old name of the child under the parent directory
2011 * \param[in] name2 the new name of the child under the parent directory
2012 * \param[in] type the type claimed by the name entry
2013 * \param[in] update update the name entry if true; otherwise, remove it
2014 * \param[in] dec decrease the parent nlink count if true
2016 * \retval positive number for repaired successfully
2017 * \retval 0 if nothing to be repaired
2018 * \retval negative error number on failure
2020 int lfsck_namespace_repair_dirent(const struct lu_env *env,
2021 struct lfsck_component *com,
2022 struct dt_object *parent,
2023 struct dt_object *child,
2024 const char *name, const char *name2,
2025 __u16 type, bool update, bool dec)
2027 struct lfsck_thread_info *info = lfsck_env_info(env);
2028 struct dt_insert_rec *rec = &info->lti_dt_rec;
2029 const struct lu_fid *pfid = lfsck_dto2fid(parent);
2030 struct lu_fid cfid = {0};
2032 struct lfsck_instance *lfsck = com->lc_lfsck;
2033 struct dt_device *dev = lfsck->li_next;
2034 struct thandle *th = NULL;
2035 struct lfsck_lock_handle *llh = &info->lti_llh;
2036 struct lustre_handle lh = { 0 };
2041 cfid = *lfsck_dto2fid(child);
2042 parent = lfsck_object_locate(dev, parent);
2044 GOTO(log, rc = PTR_ERR(parent));
2046 if (unlikely(!dt_try_as_dir(env, parent)))
2047 GOTO(log, rc = -ENOTDIR);
2049 if (!update || strcmp(name, name2) == 0)
2050 rc = lfsck_lock(env, lfsck, parent, name, llh,
2051 MDS_INODELOCK_UPDATE, LCK_PW);
2053 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
2054 MDS_INODELOCK_UPDATE, LCK_PW);
2058 th = dt_trans_create(env, dev);
2060 GOTO(unlock1, rc = PTR_ERR(th));
2062 rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
2067 rec->rec_type = lfsck_object_type(child) & S_IFMT;
2068 LASSERT(!fid_is_zero(&cfid));
2069 rec->rec_fid = &cfid;
2070 rc = dt_declare_insert(env, parent,
2071 (const struct dt_rec *)rec,
2072 (const struct dt_key *)name2, th);
2077 if (dec && S_ISDIR(type)) {
2078 rc = dt_declare_ref_del(env, parent, th);
2083 rc = dt_trans_start_local(env, dev, th);
2088 dt_write_lock(env, parent, 0);
2089 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
2090 (const struct dt_key *)name);
2091 /* Someone has removed the bad name entry by race. */
2093 GOTO(unlock2, rc = 0);
2098 /* Someone has removed the bad name entry and reused it for other
2099 * object by race. */
2100 if (!lu_fid_eq(&tfid, &cfid))
2101 GOTO(unlock2, rc = 0);
2103 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2104 GOTO(unlock2, rc = 1);
2106 rc = dt_delete(env, parent, (const struct dt_key *)name, th);
2111 rc = dt_insert(env, parent,
2112 (const struct dt_rec *)rec,
2113 (const struct dt_key *)name2, th, 1);
2118 if (dec && S_ISDIR(type)) {
2119 rc = dt_ref_del(env, parent, th);
2124 GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
2127 dt_write_unlock(env, parent);
2130 dt_trans_stop(env, dev, th);
2132 /* We are not sure whether the child will become orphan or not.
2133 * Record it in the LFSCK trace file for further checking in
2134 * the second-stage scanning. */
2135 if (!update && !dec && child && rc == 0)
2136 lfsck_namespace_trace_update(env, com, &cfid,
2137 LNTF_CHECK_LINKEA, true);
2140 /* It is harmless even if unlock the unused lock_handle */
2141 lfsck_ibits_unlock(&lh, LCK_PW);
2145 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found bad name "
2146 "entry for: parent "DFID", child "DFID", name %s, type "
2147 "in name entry %o, type claimed by child %o. repair it "
2148 "by %s with new name2 %s: rc = %d\n",
2149 lfsck_lfsck2name(lfsck), PFID(pfid), PFID(&cfid),
2150 name, type, update ? lfsck_object_type(child) : 0,
2151 update ? "updating" : "removing", name2, rc);
2154 struct lfsck_namespace *ns = com->lc_file_ram;
2156 ns->ln_flags |= LF_INCONSISTENT;
2163 * Update the ".." name entry for the given object.
2165 * The object's ".." is corrupted, this function will update the ".." name
2166 * entry with the given pfid, and the linkEA with the given ldata.
2168 * The caller should take the ldlm lock before the calling.
2170 * \param[in] env pointer to the thread context
2171 * \param[in] com pointer to the lfsck component
2172 * \param[in] obj pointer to the dt_object to be handled
2173 * \param[in] pfid the new fid for the object's ".." name entry
2174 * \param[in] cname the name for the @obj in the parent directory
2176 * \retval positive number for repaired cases
2177 * \retval 0 if nothing to be repaired
2178 * \retval negative error number on failure
2180 static int lfsck_namespace_repair_unmatched_pairs(const struct lu_env *env,
2181 struct lfsck_component *com,
2182 struct dt_object *obj,
2183 const struct lu_fid *pfid,
2184 struct lu_name *cname)
2186 struct lfsck_thread_info *info = lfsck_env_info(env);
2187 struct dt_insert_rec *rec = &info->lti_dt_rec;
2188 struct lfsck_instance *lfsck = com->lc_lfsck;
2189 struct dt_device *dev = lfsck_obj2dev(obj);
2190 struct thandle *th = NULL;
2191 struct linkea_data ldata = { NULL };
2192 struct lu_buf linkea_buf;
2196 LASSERT(!dt_object_remote(obj));
2197 LASSERT(S_ISDIR(lfsck_object_type(obj)));
2199 rc = linkea_links_new(&ldata, &info->lti_big_buf, cname, pfid);
2203 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
2204 ldata.ld_leh->leh_len);
2206 th = dt_trans_create(env, dev);
2208 GOTO(log, rc = PTR_ERR(th));
2210 rc = dt_declare_delete(env, obj, (const struct dt_key *)dotdot, th);
2214 rec->rec_type = S_IFDIR;
2215 rec->rec_fid = pfid;
2216 rc = dt_declare_insert(env, obj, (const struct dt_rec *)rec,
2217 (const struct dt_key *)dotdot, th);
2221 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
2222 XATTR_NAME_LINK, 0, th);
2226 rc = dt_trans_start_local(env, dev, th);
2230 dt_write_lock(env, obj, 0);
2231 if (unlikely(lfsck_is_dead_obj(obj)))
2232 GOTO(unlock, rc = 0);
2234 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2235 GOTO(unlock, rc = 1);
2237 /* The old ".." name entry maybe not exist. */
2238 dt_delete(env, obj, (const struct dt_key *)dotdot, th);
2240 rc = dt_insert(env, obj, (const struct dt_rec *)rec,
2241 (const struct dt_key *)dotdot, th, 1);
2245 rc = lfsck_links_write(env, obj, &ldata, th);
2247 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2250 dt_write_unlock(env, obj);
2253 dt_trans_stop(env, dev, th);
2256 CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild dotdot name entry for "
2257 "the object "DFID", new parent "DFID": rc = %d\n",
2258 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)),
2262 struct lfsck_namespace *ns = com->lc_file_ram;
2264 ns->ln_flags |= LF_INCONSISTENT;
2271 * Handle orphan @obj during Double Scan Directory.
2273 * Remove the @obj's current (invalid) linkEA entries, and insert
2274 * it in the directory .lustre/lost+found/MDTxxxx/ with the name:
2275 * ${FID}-${PFID}-D-${conflict_version}
2277 * The caller should take the ldlm lock before the calling.
2279 * \param[in] env pointer to the thread context
2280 * \param[in] com pointer to the lfsck component
2281 * \param[in] obj pointer to the orphan object to be handled
2282 * \param[in] pfid the new fid for the object's ".." name entry
2283 * \param[in,out] lh ldlm lock handler for the given @obj
2284 * \param[out] type to tell the caller what the inconsistency is
2286 * \retval positive number for repaired cases
2287 * \retval 0 if nothing to be repaired
2288 * \retval negative error number on failure
2291 lfsck_namespace_dsd_orphan(const struct lu_env *env,
2292 struct lfsck_component *com,
2293 struct dt_object *obj,
2294 const struct lu_fid *pfid,
2295 struct lustre_handle *lh,
2296 enum lfsck_namespace_inconsistency_type *type)
2298 struct lfsck_thread_info *info = lfsck_env_info(env);
2299 struct lfsck_namespace *ns = com->lc_file_ram;
2303 /* Remove the unrecognized linkEA. */
2304 rc = lfsck_namespace_links_remove(env, com, obj);
2305 lfsck_ibits_unlock(lh, LCK_EX);
2306 if (rc < 0 && rc != -ENODATA)
2309 *type = LNIT_MUL_REF;
2311 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
2312 * ever tried to verify some remote MDT-object that resides on this
2313 * MDT, but this MDT failed to respond such request. So means there
2314 * may be some remote name entry on other MDT that references this
2315 * object with another name, so we cannot know whether this linkEA
2316 * is valid or not. So keep it there and maybe resolved when next
2318 if (ns->ln_flags & LF_INCOMPLETE)
2321 /* The unique linkEA is invalid, even if the ".." name entry may be
2322 * valid, we still cannot know via which name entry this directory
2323 * will be referenced. Then handle it as pure orphan. */
2324 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2325 "-"DFID, PFID(pfid));
2326 rc = lfsck_namespace_insert_orphan(env, com, obj,
2327 info->lti_tmpbuf, "D", NULL);
2333 * Double Scan Directory object for single linkEA entry case.
2335 * The given @child has unique linkEA entry. If the linkEA entry is valid,
2336 * then check whether the name is in the namespace or not, if not, add the
2337 * missing name entry back to namespace. If the linkEA entry is invalid,
2338 * then remove it and insert the @child in the .lustre/lost+found/MDTxxxx/
2341 * \param[in] env pointer to the thread context
2342 * \param[in] com pointer to the lfsck component
2343 * \param[in] child pointer to the directory to be double scanned
2344 * \param[in] pfid the FID corresponding to the ".." entry
2345 * \param[in] ldata pointer to the linkEA data for the given @child
2346 * \param[in,out] lh ldlm lock handler for the given @child
2347 * \param[out] type to tell the caller what the inconsistency is
2348 * \param[in] retry if found inconsistency, but the caller does not hold
2349 * ldlm lock on the @child, then set @retry as true
2350 * \param[in] unknown set if does not know how to repair the inconsistency
2352 * \retval positive number for repaired cases
2353 * \retval 0 if nothing to be repaired
2354 * \retval negative error number on failure
2357 lfsck_namespace_dsd_single(const struct lu_env *env,
2358 struct lfsck_component *com,
2359 struct dt_object *child,
2360 const struct lu_fid *pfid,
2361 struct linkea_data *ldata,
2362 struct lustre_handle *lh,
2363 enum lfsck_namespace_inconsistency_type *type,
2364 bool *retry, bool *unknown)
2366 struct lfsck_thread_info *info = lfsck_env_info(env);
2367 struct lu_name *cname = &info->lti_name;
2368 const struct lu_fid *cfid = lfsck_dto2fid(child);
2370 struct lfsck_namespace *ns = com->lc_file_ram;
2371 struct lfsck_instance *lfsck = com->lc_lfsck;
2372 struct dt_object *parent = NULL;
2373 struct lmv_mds_md_v1 *lmv;
2377 rc = lfsck_namespace_unpack_linkea_entry(ldata, cname, &tfid,
2379 sizeof(info->lti_key));
2380 /* The unique linkEA entry with bad parent will be handled as orphan. */
2382 if (!lustre_handle_is_used(lh) && retry != NULL)
2385 rc = lfsck_namespace_dsd_orphan(env, com, child,
2391 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
2393 GOTO(out, rc = PTR_ERR(parent));
2395 /* We trust the unique linkEA entry in spite of whether it matches the
2396 * ".." name entry or not. Because even if the linkEA entry is wrong
2397 * and the ".." name entry is right, we still cannot know via which
2398 * name entry the child will be referenced, since all known entries
2399 * have been verified during the first-stage scanning. */
2400 if (!dt_object_exists(parent)) {
2401 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2402 * has ever tried to verify some remote MDT-object that resides
2403 * on this MDT, but this MDT failed to respond such request. So
2404 * means there may be some remote name entry on other MDT that
2405 * references this object with another name, so we cannot know
2406 * whether this linkEA is valid or not. So keep it there and
2407 * maybe resolved when next LFSCK run. */
2408 if (ns->ln_flags & LF_INCOMPLETE)
2411 if (!lustre_handle_is_used(lh) && retry != NULL) {
2417 lfsck_ibits_unlock(lh, LCK_EX);
2420 lmv = &info->lti_lmv;
2421 rc = lfsck_read_stripe_lmv(env, child, lmv);
2422 if (rc != 0 && rc != -ENODATA)
2425 if (rc == -ENODATA || lmv->lmv_magic != LMV_MAGIC_STRIPE) {
2427 } else if (lfsck_shard_name_to_index(env,
2428 cname->ln_name, cname->ln_namelen,
2429 S_IFDIR, cfid) < 0) {
2430 /* It is an invalid name entry, we
2431 * cannot trust the parent also. */
2432 rc = lfsck_namespace_shrink_linkea(env, com, child,
2433 ldata, cname, &tfid, true);
2437 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2438 "-"DFID, PFID(pfid));
2439 rc = lfsck_namespace_insert_orphan(env, com, child,
2440 info->lti_tmpbuf, "S", NULL);
2445 /* Create the lost parent as an orphan. */
2446 rc = lfsck_namespace_create_orphan_dir(env, com, parent, lmv);
2448 /* Add the missing name entry to the parent. */
2449 rc = lfsck_namespace_insert_normal(env, com, parent,
2450 child, cname->ln_name);
2451 if (unlikely(rc == -EEXIST)) {
2452 /* Unfortunately, someone reused the name
2453 * under the parent by race. So we have
2454 * to remove the linkEA entry from
2455 * current child object. It means that the
2456 * LFSCK cannot recover the system
2457 * totally back to its original status,
2458 * but it is necessary to make the
2459 * current system to be consistent. */
2460 rc = lfsck_namespace_shrink_linkea(env,
2462 cname, &tfid, true);
2464 snprintf(info->lti_tmpbuf,
2465 sizeof(info->lti_tmpbuf),
2466 "-"DFID, PFID(pfid));
2467 rc = lfsck_namespace_insert_orphan(env,
2468 com, child, info->lti_tmpbuf,
2475 } /* !dt_object_exists(parent) */
2477 /* The unique linkEA entry with bad parent will be handled as orphan. */
2478 if (unlikely(!dt_try_as_dir(env, parent))) {
2479 if (!lustre_handle_is_used(lh) && retry != NULL)
2482 rc = lfsck_namespace_dsd_orphan(env, com, child,
2488 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
2489 (const struct dt_key *)cname->ln_name);
2490 if (rc == -ENOENT) {
2491 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2492 * has ever tried to verify some remote MDT-object that resides
2493 * on this MDT, but this MDT failed to respond such request. So
2494 * means there may be some remote name entry on other MDT that
2495 * references this object with another name, so we cannot know
2496 * whether this linkEA is valid or not. So keep it there and
2497 * maybe resolved when next LFSCK run. */
2498 if (ns->ln_flags & LF_INCOMPLETE)
2501 if (!lustre_handle_is_used(lh) && retry != NULL) {
2507 lfsck_ibits_unlock(lh, LCK_EX);
2508 rc = lfsck_namespace_check_name(env, parent, child, cname);
2515 /* It is an invalid name entry, drop it. */
2516 if (unlikely(rc > 0)) {
2517 rc = lfsck_namespace_shrink_linkea(env, com, child,
2518 ldata, cname, &tfid, true);
2520 snprintf(info->lti_tmpbuf,
2521 sizeof(info->lti_tmpbuf),
2522 "-"DFID, PFID(pfid));
2523 rc = lfsck_namespace_insert_orphan(env, com,
2524 child, info->lti_tmpbuf, "D", NULL);
2530 /* Add the missing name entry back to the namespace. */
2531 rc = lfsck_namespace_insert_normal(env, com, parent, child,
2533 if (unlikely(rc == -ESTALE))
2534 /* It may happen when the remote object has been
2535 * removed, but the local MDT is not aware of that. */
2538 if (unlikely(rc == -EEXIST)) {
2539 /* Unfortunately, someone reused the name under the
2540 * parent by race. So we have to remove the linkEA
2541 * entry from current child object. It means that the
2542 * LFSCK cannot recover the system totally back to
2543 * its original status, but it is necessary to make
2544 * the current system to be consistent.
2546 * It also may be because of the LFSCK found some
2547 * internal status of create operation. Under such
2548 * case, nothing to be done. */
2549 rc = lfsck_namespace_shrink_linkea_cond(env, com,
2550 parent, child, ldata, cname, &tfid);
2552 snprintf(info->lti_tmpbuf,
2553 sizeof(info->lti_tmpbuf),
2554 "-"DFID, PFID(pfid));
2555 rc = lfsck_namespace_insert_orphan(env, com,
2556 child, info->lti_tmpbuf, "D", NULL);
2561 } /* rc == -ENOENT */
2566 if (!lu_fid_eq(&tfid, cfid)) {
2567 if (!lustre_handle_is_used(lh) && retry != NULL) {
2573 lfsck_ibits_unlock(lh, LCK_EX);
2574 /* The name entry references another MDT-object that
2575 * may be created by the LFSCK for repairing dangling
2576 * name entry. Try to replace it. */
2577 rc = lfsck_namespace_replace_cond(env, com, parent, child,
2580 rc = lfsck_namespace_dsd_orphan(env, com, child,
2586 /* Zero FID may because the remote directroy object has invalid linkEA,
2587 * or lost linkEA. Under such case, the LFSCK on this MDT does not know
2588 * how to repair the inconsistency, but the namespace LFSCK on the MDT
2589 * where its name entry resides may has more information (name, FID) to
2590 * repair such inconsistency. So here, keep the inconsistency to avoid
2591 * some imporper repairing. */
2592 if (fid_is_zero(pfid)) {
2599 /* The ".." name entry is wrong, update it. */
2600 if (!lu_fid_eq(pfid, lfsck_dto2fid(parent))) {
2601 if (!lustre_handle_is_used(lh) && retry != NULL) {
2607 *type = LNIT_UNMATCHED_PAIRS;
2608 rc = lfsck_namespace_repair_unmatched_pairs(env, com, child,
2609 lfsck_dto2fid(parent), cname);
2615 if (parent != NULL && !IS_ERR(parent))
2616 lfsck_object_put(env, parent);
2622 * Double Scan Directory object for multiple linkEA entries case.
2624 * The given @child has multiple linkEA entries. There is at most one linkEA
2625 * entry will be valid, all the others will be removed. Firstly, the function
2626 * will try to find out the linkEA entry for which the name entry exists under
2627 * the given parent (@pfid). If there is no linkEA entry that matches the given
2628 * ".." name entry, then tries to find out the first linkEA entry that both the
2629 * parent and the name entry exist to rebuild a new ".." name entry.
2631 * \param[in] env pointer to the thread context
2632 * \param[in] com pointer to the lfsck component
2633 * \param[in] child pointer to the directory to be double scanned
2634 * \param[in] pfid the FID corresponding to the ".." entry
2635 * \param[in] ldata pointer to the linkEA data for the given @child
2636 * \param[in,out] lh ldlm lock handler for the given @child
2637 * \param[out] type to tell the caller what the inconsistency is
2638 * \param[in] lpf true if the ".." entry is under lost+found/MDTxxxx/
2639 * \param[in] unknown set if does not know how to repair the inconsistency
2641 * \retval positive number for repaired cases
2642 * \retval 0 if nothing to be repaired
2643 * \retval negative error number on failure
2646 lfsck_namespace_dsd_multiple(const struct lu_env *env,
2647 struct lfsck_component *com,
2648 struct dt_object *child,
2649 const struct lu_fid *pfid,
2650 struct linkea_data *ldata,
2651 struct lustre_handle *lh,
2652 enum lfsck_namespace_inconsistency_type *type,
2653 bool lpf, bool *unknown)
2655 struct lfsck_thread_info *info = lfsck_env_info(env);
2656 struct lu_name *cname = &info->lti_name;
2657 const struct lu_fid *cfid = lfsck_dto2fid(child);
2658 struct lu_fid *pfid2 = &info->lti_fid3;
2660 struct lfsck_namespace *ns = com->lc_file_ram;
2661 struct lfsck_instance *lfsck = com->lc_lfsck;
2662 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
2663 struct dt_object *parent = NULL;
2664 struct linkea_data ldata_new = { NULL };
2665 int dirent_count = 0;
2671 while (ldata->ld_lee != NULL) {
2672 rc = lfsck_namespace_unpack_linkea_entry(ldata, cname, &tfid,
2674 sizeof(info->lti_key));
2675 /* Drop invalid linkEA entry. */
2677 lfsck_linkea_del_buf(ldata, cname);
2681 /* Drop repeated linkEA entries. */
2682 lfsck_namespace_filter_linkea_entry(ldata, cname, &tfid, true);
2684 /* If current dotdot is the .lustre/lost+found/MDTxxxx/,
2685 * then it is possible that: the directry object has ever
2686 * been lost, but its name entry was there. In the former
2687 * LFSCK run, during the first-stage scanning, the LFSCK
2688 * found the dangling name entry, but it did not recreate
2689 * the lost object, and when moved to the second-stage
2690 * scanning, some children objects of the lost directory
2691 * object were found, then the LFSCK recreated such lost
2692 * directory object as an orphan.
2694 * When the LFSCK runs again, if the dangling name is still
2695 * there, the LFSCK should move the orphan directory object
2696 * back to the normal namespace. */
2697 if (!lpf && !fid_is_zero(pfid) &&
2698 !lu_fid_eq(pfid, &tfid) && once) {
2699 linkea_next_entry(ldata);
2703 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
2705 RETURN(PTR_ERR(parent));
2707 if (!dt_object_exists(parent)) {
2708 lfsck_object_put(env, parent);
2709 if (ldata->ld_leh->leh_reccount > 1) {
2710 /* If it is NOT the last linkEA entry, then
2711 * there is still other chance to make the
2712 * child to be visible via other parent, then
2713 * remove this linkEA entry. */
2714 lfsck_linkea_del_buf(ldata, cname);
2721 /* The linkEA entry with bad parent will be removed. */
2722 if (unlikely(!dt_try_as_dir(env, parent))) {
2723 lfsck_object_put(env, parent);
2724 lfsck_linkea_del_buf(ldata, cname);
2728 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
2729 (const struct dt_key *)cname->ln_name);
2730 *pfid2 = *lfsck_dto2fid(parent);
2731 if (rc == -ENOENT) {
2732 lfsck_object_put(env, parent);
2733 linkea_next_entry(ldata);
2738 lfsck_object_put(env, parent);
2743 if (lu_fid_eq(&tfid, cfid)) {
2744 lfsck_object_put(env, parent);
2745 /* If the parent (that is declared via linkEA entry)
2746 * directory contains the specified child, but such
2747 * parent does not match the dotdot name entry, then
2748 * trust the linkEA. */
2749 if (!fid_is_zero(pfid) && !lu_fid_eq(pfid, pfid2)) {
2750 *type = LNIT_UNMATCHED_PAIRS;
2751 rc = lfsck_namespace_repair_unmatched_pairs(env,
2752 com, child, pfid2, cname);
2758 /* It is the most common case that we find the
2759 * name entry corresponding to the linkEA entry
2760 * that matches the ".." name entry. */
2761 rc = linkea_links_new(&ldata_new, &info->lti_big_buf,
2766 rc = lfsck_namespace_rebuild_linkea(env, com, child,
2771 lfsck_linkea_del_buf(ldata, cname);
2772 linkea_first_entry(ldata);
2773 /* There may be some invalid dangling name entries under
2774 * other parent directories, remove all of them. */
2775 while (ldata->ld_lee != NULL) {
2776 rc = lfsck_namespace_unpack_linkea_entry(ldata,
2777 cname, &tfid, info->lti_key,
2778 sizeof(info->lti_key));
2782 parent = lfsck_object_find_bottom(env, lfsck,
2784 if (IS_ERR(parent)) {
2785 rc = PTR_ERR(parent);
2786 if (rc != -ENOENT &&
2787 bk->lb_param & LPF_FAILOUT)
2793 if (!dt_object_exists(parent)) {
2794 lfsck_object_put(env, parent);
2798 rc = lfsck_namespace_repair_dirent(env, com,
2799 parent, child, cname->ln_name,
2800 cname->ln_name, S_IFDIR, false, true);
2801 lfsck_object_put(env, parent);
2803 if (bk->lb_param & LPF_FAILOUT)
2812 lfsck_linkea_del_buf(ldata, cname);
2815 ns->ln_dirent_repaired += dirent_count;
2818 } /* lu_fid_eq(&tfid, lfsck_dto2fid(child)) */
2820 lfsck_ibits_unlock(lh, LCK_EX);
2821 /* The name entry references another MDT-object that may be
2822 * created by the LFSCK for repairing dangling name entry.
2823 * Try to replace it. */
2824 rc = lfsck_namespace_replace_cond(env, com, parent, child,
2826 lfsck_object_put(env, parent);
2833 lfsck_linkea_del_buf(ldata, cname);
2834 } /* while (ldata->ld_lee != NULL) */
2836 /* If there is still linkEA overflow, return. */
2837 if (unlikely(ldata->ld_leh->leh_overflow_time))
2840 linkea_first_entry(ldata);
2841 if (ldata->ld_leh->leh_reccount == 1) {
2842 rc = lfsck_namespace_dsd_single(env, com, child, pfid, ldata,
2843 lh, type, NULL, unknown);
2848 /* All linkEA entries are invalid and removed, then handle the @child
2850 if (ldata->ld_leh->leh_reccount == 0) {
2851 rc = lfsck_namespace_dsd_orphan(env, com, child, pfid, lh,
2857 /* If the dangling name entry for the orphan directory object has
2858 * been remvoed, then just check whether the directory object is
2859 * still under the .lustre/lost+found/MDTxxxx/ or not. */
2865 /* There is no linkEA entry that matches the ".." name entry. Find
2866 * the first linkEA entry that both parent and name entry exist to
2867 * rebuild a new ".." name entry. */
2877 * Repair the object's nlink attribute.
2879 * If all the known name entries have been verified, then the object's hard
2880 * link attribute should match the object's linkEA entries count unless the
2881 * object's has too many hard link to be recorded in the linkEA. Such cases
2882 * should have been marked in the LFSCK trace file. Otherwise, trust the
2883 * linkEA to update the object's nlink attribute.
2885 * \param[in] env pointer to the thread context
2886 * \param[in] com pointer to the lfsck component
2887 * \param[in] obj pointer to the dt_object to be handled
2888 * \param[in,out] la pointer to buffer to object's attribute before
2889 * and after the repairing
2891 * \retval positive number for repaired cases
2892 * \retval 0 if nothing to be repaired
2893 * \retval negative error number on failure
2895 static int lfsck_namespace_repair_nlink(const struct lu_env *env,
2896 struct lfsck_component *com,
2897 struct dt_object *obj,
2900 struct lfsck_namespace *ns = com->lc_file_ram;
2901 struct lfsck_instance *lfsck = com->lc_lfsck;
2902 struct dt_device *dev = lfsck_obj2dev(obj);
2903 const struct lu_fid *cfid = lfsck_dto2fid(obj);
2904 struct thandle *th = NULL;
2905 struct linkea_data ldata = { NULL };
2906 struct lustre_handle lh = { 0 };
2907 __u32 old = la->la_nlink;
2911 LASSERT(!dt_object_remote(obj));
2913 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
2914 MDS_INODELOCK_UPDATE, LCK_PW);
2918 th = dt_trans_create(env, dev);
2920 GOTO(log, rc = PTR_ERR(th));
2922 la->la_valid = LA_NLINK;
2923 rc = dt_declare_attr_set(env, obj, la, th);
2927 rc = dt_trans_start_local(env, dev, th);
2931 dt_write_lock(env, obj, 0);
2932 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
2933 * ever tried to verify some remote MDT-object that resides on this
2934 * MDT, but this MDT failed to respond such request. So means there
2935 * may be some remote name entry on other MDT that references this
2936 * object with another name, so we cannot know whether this linkEA
2937 * is valid or not. So keep it there and maybe resolved when next
2939 if (ns->ln_flags & LF_INCOMPLETE)
2940 GOTO(unlock, rc = 0);
2942 rc = dt_attr_get(env, obj, la);
2944 GOTO(unlock, rc = (rc == -ENOENT ? 0 : rc));
2946 rc = lfsck_links_read2_with_rec(env, obj, &ldata);
2948 GOTO(unlock, rc = (rc == -ENODATA ? 0 : rc));
2950 /* XXX: Currently, we only update the nlink attribute if the known
2951 * linkEA entries is larger than the nlink attribute. That is
2953 if (la->la_nlink >= ldata.ld_leh->leh_reccount ||
2954 unlikely(la->la_nlink == 0 ||
2955 ldata.ld_leh->leh_overflow_time))
2956 GOTO(unlock, rc = 0);
2958 la->la_nlink = ldata.ld_leh->leh_reccount;
2959 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2960 GOTO(unlock, rc = 1);
2962 rc = dt_attr_set(env, obj, la, th);
2964 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2967 dt_write_unlock(env, obj);
2970 dt_trans_stop(env, dev, th);
2973 lfsck_ibits_unlock(&lh, LCK_PW);
2974 CDEBUG(D_LFSCK, "%s: namespace LFSCK repaired the object "DFID"'s "
2975 "nlink count from %u to %u: rc = %d\n",
2976 lfsck_lfsck2name(lfsck), PFID(cfid), old, la->la_nlink, rc);
2979 ns->ln_flags |= LF_INCONSISTENT;
2985 * Double scan the directory object for namespace LFSCK.
2987 * This function will verify the <parent, child> pairs in the namespace tree:
2988 * the parent references the child via some name entry that should be in the
2989 * child's linkEA entry, the child should back references the parent via its
2992 * The LFSCK will scan every linkEA entry in turn until find out the first
2993 * matched pairs. If found, then all other linkEA entries will be dropped.
2994 * If all the linkEA entries cannot match the ".." name entry, then there
2995 * are serveral possible cases:
2997 * 1) If there is only one linkEA entry, then trust it as long as the PFID
2998 * in the linkEA entry is valid.
3000 * 2) If there are multiple linkEA entries, then try to find the linkEA
3001 * that matches the ".." name entry. If found, then all other entries
3002 * are invalid; otherwise, it is quite possible that the ".." name entry
3003 * is corrupted. Under such case, the LFSCK will rebuild the ".." name
3004 * entry according to the first valid linkEA entry (both the parent and
3005 * the name entry should exist).
3007 * 3) If the directory object has no (valid) linkEA entry, then the
3008 * directory object will be handled as pure orphan and inserted
3009 * in the .lustre/lost+found/MDTxxxx/ with the name:
3010 * ${self_FID}-${PFID}-D-${conflict_version}
3012 * \param[in] env pointer to the thread context
3013 * \param[in] com pointer to the lfsck component
3014 * \param[in] child pointer to the directory object to be handled
3015 * \param[in] flags to indicate the specical checking on the @child
3017 * \retval positive number for repaired cases
3018 * \retval 0 if nothing to be repaired
3019 * \retval negative error number on failure
3021 static int lfsck_namespace_double_scan_dir(const struct lu_env *env,
3022 struct lfsck_component *com,
3023 struct dt_object *child, __u8 flags)
3025 struct lfsck_thread_info *info = lfsck_env_info(env);
3026 const struct lu_fid *cfid = lfsck_dto2fid(child);
3027 struct lu_fid *pfid = &info->lti_fid2;
3028 struct lfsck_namespace *ns = com->lc_file_ram;
3029 struct lfsck_instance *lfsck = com->lc_lfsck;
3030 struct lustre_handle lh = { 0 };
3031 struct linkea_data ldata = { NULL };
3032 bool unknown = false;
3035 enum lfsck_namespace_inconsistency_type type = LNIT_BAD_LINKEA;
3039 LASSERT(!dt_object_remote(child));
3041 if (flags & LNTF_UNCERTAIN_LMV) {
3042 if (flags & LNTF_RECHECK_NAME_HASH) {
3043 rc = lfsck_namespace_scan_shard(env, com, child);
3047 ns->ln_striped_shards_scanned++;
3049 ns->ln_striped_shards_skipped++;
3053 flags &= ~(LNTF_RECHECK_NAME_HASH | LNTF_UNCERTAIN_LMV);
3057 if (flags & (LNTF_CHECK_LINKEA | LNTF_CHECK_PARENT) &&
3058 !(lfsck->li_bookmark_ram.lb_param & LPF_ALL_TGT)) {
3059 CDEBUG(D_LFSCK, "%s: some MDT(s) maybe NOT take part in the"
3060 "the namespace LFSCK, then the LFSCK cannot guarantee"
3061 "all the name entries have been verified in first-stage"
3062 "scanning. So have to skip orphan related handling for"
3063 "the directory object "DFID" with remote name entry\n",
3064 lfsck_lfsck2name(lfsck), PFID(cfid));
3069 if (unlikely(!dt_try_as_dir(env, child)))
3070 GOTO(out, rc = -ENOTDIR);
3072 /* We only take ldlm lock on the @child when required. When the
3073 * logic comes here for the first time, it is always false. */
3077 rc = lfsck_ibits_lock(env, lfsck, child, &lh,
3078 MDS_INODELOCK_UPDATE |
3079 MDS_INODELOCK_XATTR, LCK_EX);
3084 dt_read_lock(env, child, 0);
3085 if (unlikely(lfsck_is_dead_obj(child))) {
3086 dt_read_unlock(env, child);
3091 rc = dt_lookup(env, child, (struct dt_rec *)pfid,
3092 (const struct dt_key *)dotdot);
3094 if (rc != -ENOENT && rc != -ENODATA && rc != -EINVAL) {
3095 dt_read_unlock(env, child);
3100 if (!lustre_handle_is_used(&lh)) {
3101 dt_read_unlock(env, child);
3106 } else if (lfsck->li_lpf_obj != NULL &&
3107 lu_fid_eq(pfid, lfsck_dto2fid(lfsck->li_lpf_obj))) {
3109 } else if (unlikely(!fid_is_sane(pfid))) {
3113 rc = lfsck_links_read(env, child, &ldata);
3114 dt_read_unlock(env, child);
3116 if (rc != -ENODATA && rc != -EINVAL)
3119 if (!lustre_handle_is_used(&lh))
3122 if (rc == -EINVAL && !fid_is_zero(pfid)) {
3123 /* Remove the corrupted linkEA. */
3124 rc = lfsck_namespace_links_remove(env, com, child);
3126 /* Here, because of the crashed linkEA, we
3127 * cannot know whether there is some parent
3128 * that references the child directory via
3129 * some name entry or not. So keep it there,
3130 * when the LFSCK run next time, if there is
3131 * some parent that references this object,
3132 * then the LFSCK can rebuild the linkEA;
3133 * otherwise, this object will be handled
3134 * as orphan as above. */
3137 /* 1. If we have neither ".." nor linkEA,
3138 * then it is an orphan.
3140 * 2. If we only have the ".." name entry,
3141 * but no parent references this child
3142 * directory, then handle it as orphan. */
3143 lfsck_ibits_unlock(&lh, LCK_EX);
3144 type = LNIT_MUL_REF;
3146 /* If the LFSCK is marked as LF_INCOMPLETE,
3147 * then means some MDT has ever tried to
3148 * verify some remote MDT-object that resides
3149 * on this MDT, but this MDT failed to respond
3150 * such request. So means there may be some
3151 * remote name entry on other MDT that
3152 * references this object with another name,
3153 * so we cannot know whether this linkEA is
3154 * valid or not. So keep it there and maybe
3155 * resolved when next LFSCK run. */
3156 if (ns->ln_flags & LF_INCOMPLETE)
3159 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
3160 "-"DFID, PFID(pfid));
3161 rc = lfsck_namespace_insert_orphan(env, com, child,
3162 info->lti_tmpbuf, "D", NULL);
3168 linkea_first_entry(&ldata);
3169 /* This is the most common case: the object has unique linkEA entry. */
3170 if (ldata.ld_leh->leh_reccount == 1) {
3171 rc = lfsck_namespace_dsd_single(env, com, child, pfid, &ldata,
3172 &lh, &type, &retry, &unknown);
3174 LASSERT(!lustre_handle_is_used(&lh));
3183 if (!lustre_handle_is_used(&lh))
3186 if (unlikely(ldata.ld_leh->leh_reccount == 0)) {
3187 rc = lfsck_namespace_dsd_orphan(env, com, child, pfid, &lh,
3193 /* When we come here, the cases usually like that:
3194 * 1) The directory object has a corrupted linkEA entry. During the
3195 * first-stage scanning, the LFSCK cannot know such corruption,
3196 * then it appends the right linkEA entry according to the found
3197 * name entry after the bad one.
3199 * 2) The directory object has a right linkEA entry. During the
3200 * first-stage scanning, the LFSCK finds some bad name entry,
3201 * but the LFSCK cannot aware that at that time, then it adds
3202 * the bad linkEA entry for further processing. */
3203 rc = lfsck_namespace_dsd_multiple(env, com, child, pfid, &ldata,
3204 &lh, &type, lpf, &unknown);
3209 lfsck_ibits_unlock(&lh, LCK_EX);
3212 case LNIT_BAD_LINKEA:
3213 ns->ln_linkea_repaired++;
3215 case LNIT_UNMATCHED_PAIRS:
3216 ns->ln_unmatched_pairs_repaired++;
3219 ns->ln_mul_ref_repaired++;
3227 ns->ln_unknown_inconsistency++;
3233 lfsck_namespace_linkea_stale_overflow(struct linkea_data *ldata,
3234 struct lfsck_namespace *ns)
3236 /* Both the leh_overflow_time and ln_time_latest_reset are
3237 * local time based, so need NOT to care about clock drift
3238 * among the servers. */
3239 return ldata->ld_leh->leh_overflow_time &&
3240 ldata->ld_leh->leh_overflow_time < ns->ln_time_latest_reset;
3244 * Clear the object's linkEA overflow timestamp.
3246 * If the MDT-object has too many hard links as to the linkEA cannot hold
3247 * all of them, then overflow timestamp will be set in the linkEA header.
3248 * If some hard links are removed after that, then it is possible to hold
3249 * other missed linkEA entries. If the namespace LFSCK have added all the
3250 * related linkEA entries, then it will remove the overflow timestamp.
3252 * \param[in] env pointer to the thread context
3253 * \param[in] com pointer to the lfsck component
3254 * \param[in] ldata pointer to the linkEA data for the given @obj
3255 * \param[in] obj pointer to the dt_object to be handled
3257 * \retval positive number for repaired cases
3258 * \retval 0 if nothing to be repaired
3259 * \retval negative error number on failure
3261 static int lfsck_namespace_linkea_clear_overflow(const struct lu_env *env,
3262 struct lfsck_component *com,
3263 struct linkea_data *ldata,
3264 struct dt_object *obj)
3266 struct lfsck_namespace *ns = com->lc_file_ram;
3267 struct lfsck_instance *lfsck = com->lc_lfsck;
3268 struct dt_device *dev = lfsck_obj2dev(obj);
3269 struct thandle *th = NULL;
3270 struct lustre_handle lh = { 0 };
3271 struct lu_buf linkea_buf;
3275 LASSERT(!dt_object_remote(obj));
3277 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
3278 MDS_INODELOCK_UPDATE, LCK_PW);
3282 th = dt_trans_create(env, dev);
3284 GOTO(log, rc = PTR_ERR(th));
3286 rc = dt_declare_xattr_set(env, obj,
3287 lfsck_buf_get_const(env, NULL, MAX_LINKEA_SIZE),
3288 XATTR_NAME_LINK, 0, th);
3292 rc = dt_trans_start_local(env, dev, th);
3296 dt_write_lock(env, obj, 0);
3297 rc = lfsck_links_read(env, obj, ldata);
3301 if (unlikely(!lfsck_namespace_linkea_stale_overflow(ldata, ns)))
3302 GOTO(unlock, rc = 0);
3304 ldata->ld_leh->leh_overflow_time = 0;
3305 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
3306 GOTO(unlock, rc = 1);
3308 /* If all known entries are in the linkEA, then the 'leh_reccount'
3309 * should NOT be zero. */
3310 LASSERT(ldata->ld_leh->leh_reccount > 0);
3312 lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
3313 ldata->ld_leh->leh_len);
3314 rc = dt_xattr_set(env, obj, &linkea_buf, XATTR_NAME_LINK, 0, th);
3315 if (unlikely(rc == -ENOSPC))
3323 dt_write_unlock(env, obj);
3326 dt_trans_stop(env, dev, th);
3329 lfsck_ibits_unlock(&lh, LCK_PW);
3330 CDEBUG(D_LFSCK, "%s: clear linkea overflow timestamp for the object "
3332 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
3338 * Verify the object's agent entry.
3340 * If the object claims to have agent entry but the linkEA does not contain
3341 * remote parent, then remove the agent entry. Otherwise, if the object has
3342 * no agent entry but its linkEA contains remote parent, then will generate
3343 * agent entry for it.
3345 * \param[in] env pointer to the thread context
3346 * \param[in] com pointer to the lfsck component
3347 * \param[in] obj pointer to the dt_object to be handled
3349 * \retval positive number for repaired cases
3350 * \retval 0 if nothing to be repaired
3351 * \retval negative error number on failure
3353 static int lfsck_namespace_check_agent_entry(const struct lu_env *env,
3354 struct lfsck_component *com,
3355 struct dt_object *obj)
3357 struct linkea_data ldata = { NULL };
3358 struct lfsck_thread_info *info = lfsck_env_info(env);
3359 struct lfsck_namespace *ns = com->lc_file_ram;
3360 struct lfsck_instance *lfsck = com->lc_lfsck;
3361 struct lu_fid *pfid = &info->lti_fid2;
3362 struct lu_name *cname = &info->lti_name;
3363 struct lu_seq_range *range = &info->lti_range;
3364 struct seq_server_site *ss = lfsck_dev_site(lfsck);
3365 __u32 idx = lfsck_dev_idx(lfsck);
3367 bool remote = false;
3370 if (!(lfsck->li_bookmark_ram.lb_param & LPF_ALL_TGT))
3373 rc = lfsck_links_read_with_rec(env, obj, &ldata);
3374 if (rc == -ENOENT || rc == -ENODATA)
3377 if (rc && rc != -EINVAL)
3380 /* We check the agent entry again after verifying the linkEA
3381 * successfully. So invalid linkEA should be dryrun mode. */
3382 if (rc == -EINVAL || unlikely(!ldata.ld_leh->leh_reccount))
3385 linkea_first_entry(&ldata);
3386 while (ldata.ld_lee != NULL && !remote) {
3387 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
3389 /* If parent FID is unknown, not verify agent entry. */
3390 if (!fid_is_sane(pfid))
3393 fld_range_set_mdt(range);
3394 rc = fld_server_lookup(env, ss->ss_server_fld,
3395 fid_seq(pfid), range);
3397 GOTO(out, rc = (rc == -ENOENT ? 0 : rc));
3399 if (range->lsr_index != idx)
3402 linkea_next_entry(&ldata);
3405 if ((lu_object_has_agent_entry(&obj->do_lu) && !remote) ||
3406 (!lu_object_has_agent_entry(&obj->do_lu) && remote)) {
3407 struct dt_device *dev = lfsck_obj2dev(obj);
3408 struct linkea_data ldata2 = { NULL };
3409 struct lustre_handle lh = { 0 };
3410 struct lu_buf linkea_buf;
3411 struct thandle *handle;
3413 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
3416 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
3417 MDS_INODELOCK_UPDATE |
3418 MDS_INODELOCK_XATTR, LCK_EX);
3422 handle = dt_trans_create(env, dev);
3424 GOTO(unlock, rc = PTR_ERR(handle));
3426 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
3427 ldata.ld_leh->leh_len);
3428 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
3429 XATTR_NAME_LINK, LU_XATTR_REPLACE, handle);
3433 rc = dt_trans_start_local(env, dev, handle);
3437 dt_write_lock(env, obj, 0);
3438 rc = lfsck_links_read2_with_rec(env, obj, &ldata2);
3440 if (rc == -ENOENT || rc == -ENODATA)
3445 /* If someone changed linkEA by race, then the agent
3446 * entry will be updated by lower layer automatically. */
3447 if (ldata.ld_leh->leh_len != ldata2.ld_leh->leh_len ||
3448 memcmp(ldata.ld_buf->lb_buf, ldata2.ld_buf->lb_buf,
3449 ldata.ld_leh->leh_len) != 0)
3450 GOTO(unlock2, rc = 0);
3452 rc = dt_xattr_set(env, obj, &linkea_buf, XATTR_NAME_LINK,
3453 LU_XATTR_REPLACE, handle);
3460 dt_write_unlock(env, obj);
3462 dt_trans_stop(env, dev, handle);
3464 lfsck_ibits_unlock(&lh, LCK_EX);
3471 ns->ln_agent_entries_repaired++;
3473 CDEBUG(D_LFSCK, "%s: repair agent entry for "DFID": rc = %d\n",
3474 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
3479 * Double scan the MDT-object for namespace LFSCK.
3481 * If the MDT-object contains invalid or repeated linkEA entries, then drop
3482 * those entries from the linkEA; if the linkEA becomes empty or the object
3483 * has no linkEA, then it is an orphan and will be added into the directory
3484 * .lustre/lost+found/MDTxxxx/; if the remote parent is lost, then recreate
3485 * the remote parent; if the name entry corresponding to some linkEA entry
3486 * is lost, then add the name entry back to the namespace.
3488 * \param[in] env pointer to the thread context
3489 * \param[in] com pointer to the lfsck component
3490 * \param[in] child pointer to the dt_object to be handled
3491 * \param[in] flags some hints to indicate how the @child should be handled
3493 * \retval positive number for repaired cases
3494 * \retval 0 if nothing to be repaired
3495 * \retval negative error number on failure
3497 static int lfsck_namespace_double_scan_one(const struct lu_env *env,
3498 struct lfsck_component *com,
3499 struct dt_object *child, __u8 flags)
3501 struct lfsck_thread_info *info = lfsck_env_info(env);
3502 struct lu_attr *la = &info->lti_la;
3503 struct lu_name *cname = &info->lti_name;
3504 struct lu_fid *pfid = &info->lti_fid;
3505 struct lu_fid *cfid = &info->lti_fid2;
3506 struct lfsck_instance *lfsck = com->lc_lfsck;
3507 struct lfsck_namespace *ns = com->lc_file_ram;
3508 struct dt_object *parent = NULL;
3509 struct linkea_data ldata = { NULL };
3510 bool repaired = false;
3515 dt_read_lock(env, child, 0);
3516 if (unlikely(lfsck_is_dead_obj(child))) {
3517 dt_read_unlock(env, child);
3522 if (S_ISDIR(lfsck_object_type(child))) {
3523 dt_read_unlock(env, child);
3524 rc = lfsck_namespace_double_scan_dir(env, com, child, flags);
3525 if (!rc && flags & LNTF_CHECK_AGENT_ENTRY)
3526 rc = lfsck_namespace_check_agent_entry(env, com, child);
3531 rc = lfsck_links_read(env, child, &ldata);
3532 dt_read_unlock(env, child);
3534 if (rc == -EINVAL) {
3535 struct lustre_handle lh = { 0 };
3537 rc = lfsck_ibits_lock(env, com->lc_lfsck, child, &lh,
3538 MDS_INODELOCK_UPDATE |
3539 MDS_INODELOCK_XATTR, LCK_EX);
3541 rc = lfsck_namespace_links_remove(env, com, child);
3542 lfsck_ibits_unlock(&lh, LCK_EX);
3551 if (!(ns->ln_flags & LF_INCOMPLETE) &&
3552 unlikely(lfsck_namespace_linkea_stale_overflow(&ldata, ns))) {
3553 rc = lfsck_namespace_linkea_clear_overflow(env, com, &ldata,
3559 ns->ln_linkea_overflow_cleared++;
3562 linkea_first_entry(&ldata);
3563 while (ldata.ld_lee != NULL) {
3564 rc = lfsck_namespace_unpack_linkea_entry(&ldata, cname, pfid,
3566 sizeof(info->lti_key));
3567 /* Invalid PFID in the linkEA entry. */
3569 rc = lfsck_namespace_shrink_linkea(env, com, child,
3570 &ldata, cname, pfid, true);
3580 rc = lfsck_namespace_filter_linkea_entry(&ldata, cname, pfid,
3582 /* Found repeated linkEA entries */
3584 rc = lfsck_namespace_shrink_linkea(env, com, child,
3585 &ldata, cname, pfid, false);
3597 parent = lfsck_object_find_bottom(env, lfsck, pfid);
3599 GOTO(out, rc = PTR_ERR(parent));
3601 if (!dt_object_exists(parent)) {
3604 if (ldata.ld_leh->leh_reccount > 1) {
3605 /* If it is NOT the last linkEA entry, then
3606 * there is still other chance to make the
3607 * child to be visible via other parent, then
3608 * remove this linkEA entry. */
3609 rc = lfsck_namespace_shrink_linkea(env, com,
3610 child, &ldata, cname, pfid, true);
3612 /* If the LFSCK is marked as LF_INCOMPLETE,
3613 * then means some MDT has ever tried to
3614 * verify some remote MDT-object that resides
3615 * on this MDT, but this MDT failed to respond
3616 * such request. So means there may be some
3617 * remote name entry on other MDT that
3618 * references this object with another name,
3619 * so we cannot know whether this linkEA is
3620 * valid or not. So keep it there and maybe
3621 * resolved when next LFSCK run. */
3622 if (ns->ln_flags & LF_INCOMPLETE) {
3623 lfsck_object_put(env, parent);
3628 /* Create the lost parent as an orphan. */
3629 rc = lfsck_namespace_create_orphan_dir(env, com,
3632 lfsck_object_put(env, parent);
3640 /* Add the missing name entry to the parent. */
3641 rc = lfsck_namespace_insert_normal(env, com,
3642 parent, child, cname->ln_name);
3643 if (unlikely(rc == -EEXIST))
3644 /* Unfortunately, someone reused the
3645 * name under the parent by race. So we
3646 * have to remove the linkEA entry from
3647 * current child object. It means that
3648 * the LFSCK cannot recover the system
3649 * totally back to its original status,
3650 * but it is necessary to make the
3651 * current system to be consistent. */
3652 rc = lfsck_namespace_shrink_linkea(env,
3656 linkea_next_entry(&ldata);
3659 lfsck_object_put(env, parent);
3667 } /* !dt_object_exists(parent) */
3669 /* The linkEA entry with bad parent will be removed. */
3670 if (unlikely(!dt_try_as_dir(env, parent))) {
3671 lfsck_object_put(env, parent);
3672 rc = lfsck_namespace_shrink_linkea(env, com, child,
3673 &ldata, cname, pfid, true);
3683 rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
3684 (const struct dt_key *)cname->ln_name);
3685 if (rc != 0 && rc != -ENOENT) {
3686 lfsck_object_put(env, parent);
3692 if (lu_fid_eq(cfid, lfsck_dto2fid(child))) {
3693 /* It is the most common case that we
3694 * find the name entry corresponding
3695 * to the linkEA entry. */
3696 lfsck_object_put(env, parent);
3697 linkea_next_entry(&ldata);
3699 /* The name entry references another
3700 * MDT-object that may be created by
3701 * the LFSCK for repairing dangling
3702 * name entry. Try to replace it. */
3703 rc = lfsck_namespace_replace_cond(env, com,
3704 parent, child, cfid, cname);
3705 lfsck_object_put(env, parent);
3711 linkea_next_entry(&ldata);
3713 rc = lfsck_namespace_shrink_linkea(env,
3727 /* The following handles -ENOENT case */
3729 rc = dt_attr_get(env, child, la);
3733 /* If there is no name entry in the parent dir and the object
3734 * link count is fewer than the linkea entries count, then the
3735 * linkea entry should be removed. */
3736 if (ldata.ld_leh->leh_reccount > la->la_nlink) {
3737 rc = lfsck_namespace_shrink_linkea_cond(env, com,
3738 parent, child, &ldata, cname, pfid);
3739 lfsck_object_put(env, parent);
3749 /* If the LFSCK is marked as LF_INCOMPLETE, then means some
3750 * MDT has ever tried to verify some remote MDT-object that
3751 * resides on this MDT, but this MDT failed to respond such
3752 * request. So means there may be some remote name entry on
3753 * other MDT that references this object with another name,
3754 * so we cannot know whether this linkEA is valid or not.
3755 * So keep it there and maybe resolved when next LFSCK run. */
3756 if (ns->ln_flags & LF_INCOMPLETE) {
3757 lfsck_object_put(env, parent);
3762 rc = lfsck_namespace_check_name(env, parent, child, cname);
3767 lfsck_object_put(env, parent);
3772 /* It is an invalid name entry, drop it. */
3773 if (unlikely(rc > 0)) {
3774 lfsck_object_put(env, parent);
3775 rc = lfsck_namespace_shrink_linkea(env, com, child,
3776 &ldata, cname, pfid, true);
3786 /* Add the missing name entry back to the namespace. */
3787 rc = lfsck_namespace_insert_normal(env, com, parent, child,
3789 if (unlikely(rc == -ESTALE))
3790 /* It may happen when the remote object has been
3791 * removed, but the local MDT is not aware of that. */
3794 if (unlikely(rc == -EEXIST))
3795 /* Unfortunately, someone reused the name under the
3796 * parent by race. So we have to remove the linkEA
3797 * entry from current child object. It means that the
3798 * LFSCK cannot recover the system totally back to
3799 * its original status, but it is necessary to make
3800 * the current system to be consistent.
3802 * It also may be because of the LFSCK found some
3803 * internal status of create operation. Under such
3804 * case, nothing to be done. */
3805 rc = lfsck_namespace_shrink_linkea_cond(env, com,
3806 parent, child, &ldata, cname, pfid);
3808 linkea_next_entry(&ldata);
3810 lfsck_object_put(env, parent);
3821 if (rc < 0 && rc != -ENODATA)
3824 if (rc == 0 && ldata.ld_leh != NULL)
3825 count = ldata.ld_leh->leh_reccount;
3828 /* If the LFSCK is marked as LF_INCOMPLETE, then means some
3829 * MDT has ever tried to verify some remote MDT-object that
3830 * resides on this MDT, but this MDT failed to respond such
3831 * request. So means there may be some remote name entry on
3832 * other MDT that references this object with another name,
3833 * so we cannot know whether this linkEA is valid or not.
3834 * So keep it there and maybe resolved when next LFSCK run. */
3835 if (!(ns->ln_flags & LF_INCOMPLETE) &&
3836 (ldata.ld_leh == NULL ||
3837 !ldata.ld_leh->leh_overflow_time)) {
3838 /* If the child becomes orphan, then insert it into
3839 * the global .lustre/lost+found/MDTxxxx directory. */
3840 rc = lfsck_namespace_insert_orphan(env, com, child,
3846 ns->ln_mul_ref_repaired++;
3851 rc = dt_attr_get(env, child, la);
3855 if (la->la_nlink != 0 && la->la_nlink != count) {
3856 if (unlikely(!S_ISREG(lfsck_object_type(child)) &&
3857 !S_ISLNK(lfsck_object_type(child)))) {
3858 CDEBUG(D_LFSCK, "%s: namespace LFSCK finds "
3859 "the object "DFID"'s nlink count %d "
3860 "does not match linkEA count %d, "
3861 "type %o, skip it.\n",
3862 lfsck_lfsck2name(lfsck),
3863 PFID(lfsck_dto2fid(child)),
3864 la->la_nlink, count,
3865 lfsck_object_type(child));
3866 } else if (la->la_nlink < count &&
3867 likely(!ldata.ld_leh->leh_overflow_time)) {
3868 rc = lfsck_namespace_repair_nlink(env, com,
3871 ns->ln_objs_nlink_repaired++;
3879 if (la->la_nlink > 1)
3880 ns->ln_mul_linked_repaired++;
3886 if (!rc && flags & LNTF_CHECK_AGENT_ENTRY)
3887 rc = lfsck_namespace_check_agent_entry(env, com, child);
3892 static void lfsck_namespace_dump_statistics(struct seq_file *m,
3893 struct lfsck_namespace *ns,
3894 __u64 checked_phase1,
3895 __u64 checked_phase2,
3896 time64_t time_phase1,
3897 time64_t time_phase2, bool dryrun)
3899 const char *postfix = dryrun ? "inconsistent" : "repaired";
3901 seq_printf(m, "checked_phase1: %llu\n"
3902 "checked_phase2: %llu\n"
3905 "failed_phase1: %llu\n"
3906 "failed_phase2: %llu\n"
3907 "directories: %llu\n"
3911 "multiple_linked_checked: %llu\n"
3912 "multiple_linked_%s: %llu\n"
3913 "unknown_inconsistency: %llu\n"
3914 "unmatched_pairs_%s: %llu\n"
3915 "dangling_%s: %llu\n"
3916 "multiple_referenced_%s: %llu\n"
3917 "bad_file_type_%s: %llu\n"
3918 "lost_dirent_%s: %llu\n"
3919 "local_lost_found_scanned: %llu\n"
3920 "local_lost_found_moved: %llu\n"
3921 "local_lost_found_skipped: %llu\n"
3922 "local_lost_found_failed: %llu\n"
3923 "striped_dirs_scanned: %llu\n"
3924 "striped_dirs_%s: %llu\n"
3925 "striped_dirs_failed: %llu\n"
3926 "striped_dirs_disabled: %llu\n"
3927 "striped_dirs_skipped: %llu\n"
3928 "striped_shards_scanned: %llu\n"
3929 "striped_shards_%s: %llu\n"
3930 "striped_shards_failed: %llu\n"
3931 "striped_shards_skipped: %llu\n"
3932 "name_hash_%s: %llu\n"
3933 "linkea_overflow_%s: %llu\n"
3934 "agent_entries_%s: %llu\n"
3935 "success_count: %u\n"
3936 "run_time_phase1: %lld seconds\n"
3937 "run_time_phase2: %lld seconds\n",
3940 dryrun ? "inconsistent" : "updated",
3941 ns->ln_items_repaired,
3942 dryrun ? "inconsistent" : "updated",
3943 ns->ln_objs_repaired_phase2,
3944 ns->ln_items_failed,
3945 ns->ln_objs_failed_phase2,
3946 ns->ln_dirs_checked,
3947 postfix, ns->ln_dirent_repaired,
3948 postfix, ns->ln_linkea_repaired,
3949 postfix, ns->ln_objs_nlink_repaired,
3950 ns->ln_mul_linked_checked,
3951 postfix, ns->ln_mul_linked_repaired,
3952 ns->ln_unknown_inconsistency,
3953 postfix, ns->ln_unmatched_pairs_repaired,
3954 postfix, ns->ln_dangling_repaired,
3955 postfix, ns->ln_mul_ref_repaired,
3956 postfix, ns->ln_bad_type_repaired,
3957 postfix, ns->ln_lost_dirent_repaired,
3958 ns->ln_local_lpf_scanned,
3959 ns->ln_local_lpf_moved,
3960 ns->ln_local_lpf_skipped,
3961 ns->ln_local_lpf_failed,
3962 ns->ln_striped_dirs_scanned,
3963 postfix, ns->ln_striped_dirs_repaired,
3964 ns->ln_striped_dirs_failed,
3965 ns->ln_striped_dirs_disabled,
3966 ns->ln_striped_dirs_skipped,
3967 ns->ln_striped_shards_scanned,
3968 postfix, ns->ln_striped_shards_repaired,
3969 ns->ln_striped_shards_failed,
3970 ns->ln_striped_shards_skipped,
3971 postfix, ns->ln_name_hash_repaired,
3972 dryrun ? "inconsistent" : "cleared",
3973 ns->ln_linkea_overflow_cleared,
3974 postfix, ns->ln_agent_entries_repaired,
3975 ns->ln_success_count,
3980 static void lfsck_namespace_release_lmv(const struct lu_env *env,
3981 struct lfsck_component *com)
3983 struct lfsck_instance *lfsck = com->lc_lfsck;
3984 struct lfsck_namespace *ns = com->lc_file_ram;
3986 while (!list_empty(&lfsck->li_list_lmv)) {
3987 struct lfsck_lmv_unit *llu;
3988 struct lfsck_lmv *llmv;
3990 llu = list_entry(lfsck->li_list_lmv.next,
3991 struct lfsck_lmv_unit, llu_link);
3992 llmv = &llu->llu_lmv;
3994 LASSERTF(atomic_read(&llmv->ll_ref) == 1,
3995 "still in using: %u\n",
3996 atomic_read(&llmv->ll_ref));
3998 ns->ln_striped_dirs_skipped++;
3999 lfsck_lmv_put(env, llmv);
4003 static int lfsck_namespace_check_for_double_scan(const struct lu_env *env,
4004 struct lfsck_component *com,
4005 struct dt_object *obj)
4007 struct lu_attr *la = &lfsck_env_info(env)->lti_la;
4010 rc = dt_attr_get(env, obj, la);
4014 /* zero-linkEA object may be orphan, but it also maybe because
4015 * of upgrading. Currently, we cannot record it for double scan.
4016 * Because it may cause the LFSCK trace file to be too large. */
4018 /* "la_ctime" == 1 means that it has ever been removed from
4019 * backend /lost+found directory but not been added back to
4020 * the normal namespace yet. */
4022 if ((S_ISREG(lfsck_object_type(obj)) && la->la_nlink > 1) ||
4023 unlikely(la->la_ctime == 1))
4024 rc = lfsck_namespace_trace_update(env, com, lfsck_dto2fid(obj),
4025 LNTF_CHECK_LINKEA, true);
4030 /* namespace APIs */
4032 static int lfsck_namespace_reset(const struct lu_env *env,
4033 struct lfsck_component *com, bool init)
4035 struct lfsck_instance *lfsck = com->lc_lfsck;
4036 struct lfsck_namespace *ns = com->lc_file_ram;
4037 struct lfsck_assistant_data *lad = com->lc_data;
4038 struct dt_object *root;
4042 root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
4044 GOTO(log, rc = PTR_ERR(root));
4046 if (unlikely(!dt_try_as_dir(env, root)))
4047 GOTO(put, rc = -ENOTDIR);
4049 down_write(&com->lc_sem);
4051 memset(ns, 0, sizeof(*ns));
4053 __u32 count = ns->ln_success_count;
4054 time64_t last_time = ns->ln_time_last_complete;
4056 memset(ns, 0, sizeof(*ns));
4057 ns->ln_success_count = count;
4058 ns->ln_time_last_complete = last_time;
4060 ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
4061 ns->ln_status = LS_INIT;
4062 ns->ln_time_latest_reset = ktime_get_real_seconds();
4064 rc = lfsck_load_one_trace_file(env, com, root, &com->lc_obj,
4065 &dt_lfsck_namespace_features,
4066 LFSCK_NAMESPACE, true);
4070 rc = lfsck_load_sub_trace_files(env, com, &dt_lfsck_namespace_features,
4071 LFSCK_NAMESPACE, true);
4075 lad->lad_incomplete = 0;
4076 CFS_RESET_BITMAP(lad->lad_bitmap);
4078 rc = lfsck_namespace_store(env, com);
4083 up_write(&com->lc_sem);
4086 lfsck_object_put(env, root);
4088 CDEBUG(D_LFSCK, "%s: namespace LFSCK reset: rc = %d\n",
4089 lfsck_lfsck2name(lfsck), rc);
4094 lfsck_namespace_fail(const struct lu_env *env, struct lfsck_component *com,
4097 struct lfsck_namespace *ns = com->lc_file_ram;
4099 down_write(&com->lc_sem);
4101 com->lc_new_checked++;
4102 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
4103 up_write(&com->lc_sem);
4106 static void lfsck_namespace_close_dir(const struct lu_env *env,
4107 struct lfsck_component *com)
4109 struct lfsck_namespace *ns = com->lc_file_ram;
4110 struct lfsck_assistant_data *lad = com->lc_data;
4111 struct lfsck_assistant_object *lso = NULL;
4112 struct lfsck_instance *lfsck = com->lc_lfsck;
4113 struct lfsck_lmv *llmv = lfsck->li_lmv;
4114 struct lfsck_namespace_req *lnr;
4115 struct lu_attr *la = &lfsck_env_info(env)->lti_la2;
4116 __u32 size = sizeof(*lnr) + LFSCK_TMPBUF_LEN;
4118 bool wakeup = false;
4124 rc = dt_attr_get(env, lfsck->li_obj_dir, la);
4128 OBD_ALLOC(lnr, size);
4130 ns->ln_striped_dirs_skipped++;
4135 lso = lfsck_assistant_object_init(env, lfsck_dto2fid(lfsck->li_obj_dir),
4136 la, lfsck->li_pos_current.lp_oit_cookie, true);
4138 OBD_FREE(lnr, size);
4139 ns->ln_striped_dirs_skipped++;
4144 /* Generate a dummy request to indicate that all shards' name entry
4145 * in this striped directory has been scanned for the first time. */
4146 INIT_LIST_HEAD(&lnr->lnr_lar.lar_list);
4147 lnr->lnr_lar.lar_parent = lso;
4148 lnr->lnr_lmv = lfsck_lmv_get(llmv);
4149 lnr->lnr_fid = *lfsck_dto2fid(lfsck->li_obj_dir);
4150 lnr->lnr_dir_cookie = MDS_DIR_END_OFF;
4151 lnr->lnr_size = size;
4152 lnr->lnr_type = lso->lso_attr.la_mode;
4154 spin_lock(&lad->lad_lock);
4155 if (lad->lad_assistant_status < 0 ||
4156 unlikely(!thread_is_running(&lfsck->li_thread) ||
4157 !thread_is_running(&lad->lad_thread))) {
4158 spin_unlock(&lad->lad_lock);
4159 lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar);
4160 ns->ln_striped_dirs_skipped++;
4165 list_add_tail(&lnr->lnr_lar.lar_list, &lad->lad_req_list);
4166 if (lad->lad_prefetched == 0)
4169 lad->lad_prefetched++;
4170 spin_unlock(&lad->lad_lock);
4172 wake_up_all(&lad->lad_thread.t_ctl_waitq);
4177 static int lfsck_namespace_open_dir(const struct lu_env *env,
4178 struct lfsck_component *com)
4180 struct lfsck_instance *lfsck = com->lc_lfsck;
4181 struct lfsck_namespace *ns = com->lc_file_ram;
4182 struct lfsck_lmv *llmv = lfsck->li_lmv;
4189 if (llmv->ll_lmv_master) {
4190 struct lmv_mds_md_v1 *lmv = &llmv->ll_lmv;
4192 if (lmv->lmv_master_mdt_index != lfsck_dev_idx(lfsck)) {
4193 lmv->lmv_master_mdt_index =
4194 lfsck_dev_idx(lfsck);
4195 ns->ln_flags |= LF_INCONSISTENT;
4196 llmv->ll_lmv_updated = 1;
4199 rc = lfsck_namespace_verify_stripe_slave(env, com,
4200 lfsck->li_obj_dir, llmv);
4203 RETURN(rc > 0 ? 0 : rc);
4206 static int lfsck_namespace_checkpoint(const struct lu_env *env,
4207 struct lfsck_component *com, bool init)
4209 struct lfsck_instance *lfsck = com->lc_lfsck;
4210 struct lfsck_namespace *ns = com->lc_file_ram;
4214 rc = lfsck_checkpoint_generic(env, com);
4219 down_write(&com->lc_sem);
4221 ns->ln_pos_latest_start = lfsck->li_pos_checkpoint;
4223 ns->ln_pos_last_checkpoint = lfsck->li_pos_checkpoint;
4224 ns->ln_run_time_phase1 += ktime_get_seconds() -
4225 lfsck->li_time_last_checkpoint;
4226 ns->ln_time_last_checkpoint = ktime_get_real_seconds();
4227 ns->ln_items_checked += com->lc_new_checked;
4228 com->lc_new_checked = 0;
4231 rc = lfsck_namespace_store(env, com);
4232 up_write(&com->lc_sem);
4235 CDEBUG(D_LFSCK, "%s: namespace LFSCK checkpoint at the pos [%llu"
4236 ", "DFID", %#llx], status = %d: rc = %d\n",
4237 lfsck_lfsck2name(lfsck), lfsck->li_pos_current.lp_oit_cookie,
4238 PFID(&lfsck->li_pos_current.lp_dir_parent),
4239 lfsck->li_pos_current.lp_dir_cookie, ns->ln_status, rc);
4241 return rc > 0 ? 0 : rc;
4244 static int lfsck_namespace_prep(const struct lu_env *env,
4245 struct lfsck_component *com,
4246 struct lfsck_start_param *lsp)
4248 struct lfsck_instance *lfsck = com->lc_lfsck;
4249 struct lfsck_namespace *ns = com->lc_file_ram;
4250 struct lfsck_position *pos = &com->lc_pos_start;
4253 rc = lfsck_namespace_load_bitmap(env, com);
4254 if (rc != 0 || ns->ln_status == LS_COMPLETED) {
4255 rc = lfsck_namespace_reset(env, com, false);
4257 rc = lfsck_set_param(env, lfsck, lsp->lsp_start, true);
4260 CDEBUG(D_LFSCK, "%s: namespace LFSCK prep failed: "
4261 "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
4267 down_write(&com->lc_sem);
4268 ns->ln_time_latest_start = ktime_get_real_seconds();
4269 spin_lock(&lfsck->li_lock);
4271 if (ns->ln_flags & LF_SCANNED_ONCE) {
4272 if (!lfsck->li_drop_dryrun ||
4273 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent)) {
4274 ns->ln_status = LS_SCANNING_PHASE2;
4275 list_move_tail(&com->lc_link,
4276 &lfsck->li_list_double_scan);
4277 if (!list_empty(&com->lc_link_dir))
4278 list_del_init(&com->lc_link_dir);
4279 lfsck_pos_set_zero(pos);
4281 ns->ln_status = LS_SCANNING_PHASE1;
4282 ns->ln_run_time_phase1 = 0;
4283 ns->ln_run_time_phase2 = 0;
4284 ns->ln_items_checked = 0;
4285 ns->ln_items_repaired = 0;
4286 ns->ln_items_failed = 0;
4287 ns->ln_dirs_checked = 0;
4288 ns->ln_objs_checked_phase2 = 0;
4289 ns->ln_objs_repaired_phase2 = 0;
4290 ns->ln_objs_failed_phase2 = 0;
4291 ns->ln_objs_nlink_repaired = 0;
4292 ns->ln_dirent_repaired = 0;
4293 ns->ln_linkea_repaired = 0;
4294 ns->ln_mul_linked_checked = 0;
4295 ns->ln_mul_linked_repaired = 0;
4296 ns->ln_unknown_inconsistency = 0;
4297 ns->ln_unmatched_pairs_repaired = 0;
4298 ns->ln_dangling_repaired = 0;
4299 ns->ln_mul_ref_repaired = 0;
4300 ns->ln_bad_type_repaired = 0;
4301 ns->ln_lost_dirent_repaired = 0;
4302 ns->ln_striped_dirs_scanned = 0;
4303 ns->ln_striped_dirs_repaired = 0;
4304 ns->ln_striped_dirs_failed = 0;
4305 ns->ln_striped_dirs_disabled = 0;
4306 ns->ln_striped_dirs_skipped = 0;
4307 ns->ln_striped_shards_scanned = 0;
4308 ns->ln_striped_shards_repaired = 0;
4309 ns->ln_striped_shards_failed = 0;
4310 ns->ln_striped_shards_skipped = 0;
4311 ns->ln_name_hash_repaired = 0;
4312 fid_zero(&ns->ln_fid_latest_scanned_phase2);
4313 if (list_empty(&com->lc_link_dir))
4314 list_add_tail(&com->lc_link_dir,
4315 &lfsck->li_list_dir);
4316 *pos = ns->ln_pos_first_inconsistent;
4319 ns->ln_status = LS_SCANNING_PHASE1;
4320 if (list_empty(&com->lc_link_dir))
4321 list_add_tail(&com->lc_link_dir,
4322 &lfsck->li_list_dir);
4323 if (!lfsck->li_drop_dryrun ||
4324 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent)) {
4325 *pos = ns->ln_pos_last_checkpoint;
4326 pos->lp_oit_cookie++;
4328 *pos = ns->ln_pos_first_inconsistent;
4332 spin_unlock(&lfsck->li_lock);
4333 up_write(&com->lc_sem);
4335 rc = lfsck_start_assistant(env, com, lsp);
4337 CDEBUG(D_LFSCK, "%s: namespace LFSCK prep done, start pos [%llu, "
4338 DFID", %#llx]: rc = %d\n",
4339 lfsck_lfsck2name(lfsck), pos->lp_oit_cookie,
4340 PFID(&pos->lp_dir_parent), pos->lp_dir_cookie, rc);
4345 static int lfsck_namespace_exec_oit(const struct lu_env *env,
4346 struct lfsck_component *com,
4347 struct dt_object *obj)
4349 struct lfsck_thread_info *info = lfsck_env_info(env);
4350 struct lfsck_namespace *ns = com->lc_file_ram;
4351 struct lfsck_instance *lfsck = com->lc_lfsck;
4352 const struct lu_fid *fid = lfsck_dto2fid(obj);
4353 struct lu_fid *pfid = &info->lti_fid2;
4354 struct lu_name *cname = &info->lti_name;
4355 struct lu_seq_range *range = &info->lti_range;
4356 struct seq_server_site *ss = lfsck_dev_site(lfsck);
4357 struct linkea_data ldata = { NULL };
4358 __u32 idx = lfsck_dev_idx(lfsck);
4360 bool remote = false;
4363 rc = lfsck_links_read(env, obj, &ldata);
4367 /* -EINVAL means crashed linkEA, should be verified. */
4368 if (rc == -EINVAL) {
4369 rc = lfsck_namespace_trace_update(env, com, fid,
4370 LNTF_CHECK_LINKEA, true);
4372 struct lustre_handle lh = { 0 };
4374 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
4375 MDS_INODELOCK_UPDATE |
4376 MDS_INODELOCK_XATTR, LCK_EX);
4378 rc = lfsck_namespace_links_remove(env, com,
4380 lfsck_ibits_unlock(&lh, LCK_EX);
4384 GOTO(out, rc = (rc == -ENOENT ? 0 : rc));
4387 if (rc && rc != -ENODATA)
4390 if (rc == -ENODATA || unlikely(!ldata.ld_leh->leh_reccount)) {
4391 rc = lfsck_namespace_check_for_double_scan(env, com, obj);
4396 linkea_first_entry(&ldata);
4397 while (ldata.ld_lee != NULL) {
4398 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
4400 if (!fid_is_sane(pfid)) {
4401 rc = lfsck_namespace_trace_update(env, com, fid,
4402 LNTF_CHECK_PARENT, true);
4404 fld_range_set_mdt(range);
4405 rc = fld_server_lookup(env, ss->ss_server_fld,
4406 fid_seq(pfid), range);
4407 if ((rc == -ENOENT) ||
4408 (!rc && range->lsr_index != idx)) {
4416 linkea_next_entry(&ldata);
4419 if ((lu_object_has_agent_entry(&obj->do_lu) && !remote) ||
4420 (!lu_object_has_agent_entry(&obj->do_lu) && remote)) {
4421 rc = lfsck_namespace_trace_update(env, com, fid,
4422 LNTF_CHECK_AGENT_ENTRY, true);
4427 /* Record multiple-linked object. */
4428 if (ldata.ld_leh->leh_reccount > 1) {
4429 rc = lfsck_namespace_trace_update(env, com, fid,
4430 LNTF_CHECK_LINKEA, true);
4436 rc = lfsck_namespace_trace_update(env, com, fid,
4437 LNTF_CHECK_LINKEA, true);
4439 rc = lfsck_namespace_check_for_double_scan(env, com, obj);
4444 down_write(&com->lc_sem);
4445 if (S_ISDIR(lfsck_object_type(obj)))
4446 ns->ln_dirs_checked++;
4448 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
4449 up_write(&com->lc_sem);
4454 static int lfsck_namespace_exec_dir(const struct lu_env *env,
4455 struct lfsck_component *com,
4456 struct lfsck_assistant_object *lso,
4457 struct lu_dirent *ent, __u16 type)
4459 struct lfsck_assistant_data *lad = com->lc_data;
4460 struct lfsck_instance *lfsck = com->lc_lfsck;
4461 struct lfsck_namespace_req *lnr;
4462 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4463 struct ptlrpc_thread *mthread = &lfsck->li_thread;
4464 struct ptlrpc_thread *athread = &lad->lad_thread;
4465 struct l_wait_info lwi = { 0 };
4466 bool wakeup = false;
4468 l_wait_event(mthread->t_ctl_waitq,
4469 lad->lad_prefetched < bk->lb_async_windows ||
4470 !thread_is_running(mthread) ||
4471 !thread_is_running(athread),
4474 if (unlikely(!thread_is_running(mthread) ||
4475 !thread_is_running(athread)))
4478 if (unlikely(lfsck_is_dead_obj(lfsck->li_obj_dir)))
4481 lnr = lfsck_namespace_assistant_req_init(com->lc_lfsck, lso, ent, type);
4483 struct lfsck_namespace *ns = com->lc_file_ram;
4485 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
4486 return PTR_ERR(lnr);
4489 spin_lock(&lad->lad_lock);
4490 if (lad->lad_assistant_status < 0 ||
4491 unlikely(!thread_is_running(mthread) ||
4492 !thread_is_running(athread))) {
4493 spin_unlock(&lad->lad_lock);
4494 lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar);
4495 return lad->lad_assistant_status;
4498 list_add_tail(&lnr->lnr_lar.lar_list, &lad->lad_req_list);
4499 if (lad->lad_prefetched == 0)
4502 lad->lad_prefetched++;
4503 spin_unlock(&lad->lad_lock);
4505 wake_up_all(&lad->lad_thread.t_ctl_waitq);
4507 down_write(&com->lc_sem);
4508 com->lc_new_checked++;
4509 up_write(&com->lc_sem);
4514 static int lfsck_namespace_post(const struct lu_env *env,
4515 struct lfsck_component *com,
4516 int result, bool init)
4518 struct lfsck_instance *lfsck = com->lc_lfsck;
4519 struct lfsck_namespace *ns = com->lc_file_ram;
4523 lfsck_post_generic(env, com, &result);
4525 down_write(&com->lc_sem);
4526 lfsck_namespace_release_lmv(env, com);
4528 spin_lock(&lfsck->li_lock);
4530 ns->ln_pos_last_checkpoint = lfsck->li_pos_checkpoint;
4532 ns->ln_status = LS_SCANNING_PHASE2;
4533 ns->ln_flags |= LF_SCANNED_ONCE;
4534 ns->ln_flags &= ~LF_UPGRADE;
4535 list_del_init(&com->lc_link_dir);
4536 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
4537 } else if (result == 0) {
4538 if (lfsck->li_status != 0)
4539 ns->ln_status = lfsck->li_status;
4541 ns->ln_status = LS_STOPPED;
4542 if (ns->ln_status != LS_PAUSED) {
4543 list_del_init(&com->lc_link_dir);
4544 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4547 ns->ln_status = LS_FAILED;
4548 list_del_init(&com->lc_link_dir);
4549 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4551 spin_unlock(&lfsck->li_lock);
4554 ns->ln_run_time_phase1 += ktime_get_seconds() -
4555 lfsck->li_time_last_checkpoint;
4556 ns->ln_time_last_checkpoint = ktime_get_real_seconds();
4557 ns->ln_items_checked += com->lc_new_checked;
4558 com->lc_new_checked = 0;
4561 rc = lfsck_namespace_store(env, com);
4562 up_write(&com->lc_sem);
4564 CDEBUG(D_LFSCK, "%s: namespace LFSCK post done: rc = %d\n",
4565 lfsck_lfsck2name(lfsck), rc);
4571 lfsck_namespace_dump(const struct lu_env *env, struct lfsck_component *com,
4574 struct lfsck_instance *lfsck = com->lc_lfsck;
4575 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4576 struct lfsck_namespace *ns = com->lc_file_ram;
4578 down_read(&com->lc_sem);
4579 seq_printf(m, "name: lfsck_namespace\n"
4585 lfsck_status2name(ns->ln_status));
4587 lfsck_bits_dump(m, ns->ln_flags, lfsck_flags_names, "flags");
4589 lfsck_bits_dump(m, bk->lb_param, lfsck_param_names, "param");
4591 lfsck_time_dump(m, ns->ln_time_last_complete, "last_completed");
4593 lfsck_time_dump(m, ns->ln_time_latest_start, "latest_start");
4595 lfsck_time_dump(m, ns->ln_time_last_checkpoint, "last_checkpoint");
4597 lfsck_pos_dump(m, &ns->ln_pos_latest_start, "latest_start_position");
4599 lfsck_pos_dump(m, &ns->ln_pos_last_checkpoint,
4600 "last_checkpoint_position");
4602 lfsck_pos_dump(m, &ns->ln_pos_first_inconsistent,
4603 "first_failure_position");
4605 if (ns->ln_status == LS_SCANNING_PHASE1) {
4606 struct lfsck_position pos;
4607 time64_t duration = ktime_get_seconds() -
4608 lfsck->li_time_last_checkpoint;
4609 u64 checked = ns->ln_items_checked + com->lc_new_checked;
4610 u64 speed = checked;
4611 u64 new_checked = com->lc_new_checked;
4612 time64_t rtime = ns->ln_run_time_phase1 + duration;
4615 new_checked = div64_s64(new_checked, duration);
4618 speed = div64_s64(speed, rtime);
4620 lfsck_namespace_dump_statistics(m, ns, checked, 0, rtime, 0,
4621 bk->lb_param & LPF_DRYRUN);
4622 seq_printf(m, "average_speed_phase1: %llu items/sec\n"
4623 "average_speed_phase2: N/A\n"
4624 "average_speed_total: %llu items/sec\n"
4625 "real_time_speed_phase1: %llu items/sec\n"
4626 "real_time_speed_phase2: N/A\n",
4631 if (likely(lfsck->li_di_oit)) {
4632 const struct dt_it_ops *iops =
4633 &lfsck->li_obj_oit->do_index_ops->dio_it;
4635 /* The low layer otable-based iteration position may NOT
4636 * exactly match the namespace-based directory traversal
4637 * cookie. Generally, it is not a serious issue. But the
4638 * caller should NOT make assumption on that. */
4639 pos.lp_oit_cookie = iops->store(env, lfsck->li_di_oit);
4640 if (!lfsck->li_current_oit_processed)
4641 pos.lp_oit_cookie--;
4643 spin_lock(&lfsck->li_lock);
4644 if (lfsck->li_di_dir) {
4645 pos.lp_dir_cookie = lfsck->li_cookie_dir;
4646 if (pos.lp_dir_cookie >= MDS_DIR_END_OFF) {
4647 fid_zero(&pos.lp_dir_parent);
4648 pos.lp_dir_cookie = 0;
4651 *lfsck_dto2fid(lfsck->li_obj_dir);
4654 fid_zero(&pos.lp_dir_parent);
4655 pos.lp_dir_cookie = 0;
4657 spin_unlock(&lfsck->li_lock);
4659 pos = ns->ln_pos_last_checkpoint;
4662 lfsck_pos_dump(m, &pos, "current_position");
4663 } else if (ns->ln_status == LS_SCANNING_PHASE2) {
4664 time64_t duration = ktime_get_seconds() -
4665 com->lc_time_last_checkpoint;
4666 __u64 checked = ns->ln_objs_checked_phase2 +
4667 com->lc_new_checked;
4668 __u64 speed1 = ns->ln_items_checked;
4669 __u64 speed2 = checked;
4670 __u64 speed0 = speed1 + speed2;
4671 __u64 new_checked = com->lc_new_checked;
4672 time64_t rtime = ns->ln_run_time_phase2 + duration;
4673 time64_t time0 = ns->ln_run_time_phase1 + rtime;
4676 new_checked = div64_s64(new_checked, duration);
4678 if (ns->ln_run_time_phase1 != 0)
4679 speed1 = div64_s64(speed1, ns->ln_run_time_phase1);
4680 else if (ns->ln_items_checked != 0)
4684 speed2 = div64_s64(speed2, rtime);
4685 else if (checked != 0)
4689 speed0 = div64_s64(speed0, time0);
4691 lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked,
4693 ns->ln_run_time_phase1, rtime,
4694 bk->lb_param & LPF_DRYRUN);
4695 seq_printf(m, "average_speed_phase1: %llu items/sec\n"
4696 "average_speed_phase2: %llu objs/sec\n"
4697 "average_speed_total: %llu items/sec\n"
4698 "real_time_speed_phase1: N/A\n"
4699 "real_time_speed_phase2: %llu objs/sec\n"
4700 "current_position: "DFID"\n",
4705 PFID(&ns->ln_fid_latest_scanned_phase2));
4707 __u64 speed1 = ns->ln_items_checked;
4708 __u64 speed2 = ns->ln_objs_checked_phase2;
4709 __u64 speed0 = speed1 + speed2;
4710 time64_t time0 = ns->ln_run_time_phase1 + ns->ln_run_time_phase2;
4712 if (ns->ln_run_time_phase1 != 0)
4713 speed1 = div64_s64(speed1, ns->ln_run_time_phase1);
4714 else if (ns->ln_items_checked != 0)
4717 if (ns->ln_run_time_phase2 != 0)
4718 speed2 = div64_s64(speed2, ns->ln_run_time_phase2);
4719 else if (ns->ln_objs_checked_phase2 != 0)
4723 speed0 = div64_s64(speed0, time0);
4725 lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked,
4726 ns->ln_objs_checked_phase2,
4727 ns->ln_run_time_phase1,
4728 ns->ln_run_time_phase2,
4729 bk->lb_param & LPF_DRYRUN);
4730 seq_printf(m, "average_speed_phase1: %llu items/sec\n"
4731 "average_speed_phase2: %llu objs/sec\n"
4732 "average_speed_total: %llu items/sec\n"
4733 "real_time_speed_phase1: N/A\n"
4734 "real_time_speed_phase2: N/A\n"
4735 "current_position: N/A\n",
4741 up_read(&com->lc_sem);
4744 static int lfsck_namespace_double_scan(const struct lu_env *env,
4745 struct lfsck_component *com)
4747 struct lfsck_namespace *ns = com->lc_file_ram;
4748 struct lfsck_assistant_data *lad = com->lc_data;
4749 struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
4750 struct lfsck_tgt_desc *ltd;
4751 struct lfsck_tgt_desc *next;
4754 rc = lfsck_double_scan_generic(env, com, ns->ln_status);
4755 if (thread_is_stopped(&lad->lad_thread)) {
4756 LASSERT(list_empty(&lad->lad_req_list));
4757 LASSERT(list_empty(&lad->lad_mdt_phase1_list));
4759 spin_lock(<ds->ltd_lock);
4760 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4761 ltd_namespace_phase_list) {
4762 list_del_init(<d->ltd_namespace_phase_list);
4764 spin_unlock(<ds->ltd_lock);
4770 static void lfsck_namespace_data_release(const struct lu_env *env,
4771 struct lfsck_component *com)
4773 struct lfsck_assistant_data *lad = com->lc_data;
4774 struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
4775 struct lfsck_tgt_desc *ltd;
4776 struct lfsck_tgt_desc *next;
4778 LASSERT(lad != NULL);
4779 LASSERT(thread_is_init(&lad->lad_thread) ||
4780 thread_is_stopped(&lad->lad_thread));
4781 LASSERT(list_empty(&lad->lad_req_list));
4783 com->lc_data = NULL;
4784 lfsck_namespace_release_lmv(env, com);
4786 spin_lock(<ds->ltd_lock);
4787 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
4788 ltd_namespace_phase_list) {
4789 list_del_init(<d->ltd_namespace_phase_list);
4791 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4792 ltd_namespace_phase_list) {
4793 list_del_init(<d->ltd_namespace_phase_list);
4795 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_list,
4796 ltd_namespace_list) {
4797 list_del_init(<d->ltd_namespace_list);
4799 spin_unlock(<ds->ltd_lock);
4801 if (likely(lad->lad_bitmap != NULL))
4802 CFS_FREE_BITMAP(lad->lad_bitmap);
4807 static void lfsck_namespace_quit(const struct lu_env *env,
4808 struct lfsck_component *com)
4810 struct lfsck_assistant_data *lad = com->lc_data;
4811 struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
4812 struct lfsck_tgt_desc *ltd;
4813 struct lfsck_tgt_desc *next;
4815 LASSERT(lad != NULL);
4817 lfsck_quit_generic(env, com);
4819 LASSERT(thread_is_init(&lad->lad_thread) ||
4820 thread_is_stopped(&lad->lad_thread));
4821 LASSERT(list_empty(&lad->lad_req_list));
4823 lfsck_namespace_release_lmv(env, com);
4825 spin_lock(<ds->ltd_lock);
4826 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
4827 ltd_namespace_phase_list) {
4828 list_del_init(<d->ltd_namespace_phase_list);
4830 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4831 ltd_namespace_phase_list) {
4832 list_del_init(<d->ltd_namespace_phase_list);
4834 spin_unlock(<ds->ltd_lock);
4837 static int lfsck_namespace_in_notify(const struct lu_env *env,
4838 struct lfsck_component *com,
4839 struct lfsck_request *lr)
4841 struct lfsck_instance *lfsck = com->lc_lfsck;
4842 struct lfsck_namespace *ns = com->lc_file_ram;
4843 struct lfsck_assistant_data *lad = com->lc_data;
4844 struct lfsck_tgt_descs *ltds = &lfsck->li_mdt_descs;
4845 struct lfsck_tgt_desc *ltd;
4850 switch (lr->lr_event) {
4851 case LE_SET_LMV_MASTER: {
4852 struct dt_object *obj;
4854 obj = lfsck_object_find_bottom(env, lfsck, &lr->lr_fid);
4856 RETURN(PTR_ERR(obj));
4858 if (likely(dt_object_exists(obj)))
4859 rc = lfsck_namespace_notify_lmv_master_local(env, com,
4862 lfsck_object_put(env, obj);
4864 RETURN(rc > 0 ? 0 : rc);
4866 case LE_SET_LMV_SLAVE: {
4867 if (!(lr->lr_flags & LEF_RECHECK_NAME_HASH))
4868 ns->ln_striped_shards_repaired++;
4870 rc = lfsck_namespace_trace_update(env, com, &lr->lr_fid,
4871 LNTF_RECHECK_NAME_HASH, true);
4873 RETURN(rc > 0 ? 0 : rc);
4875 case LE_PHASE1_DONE:
4876 case LE_PHASE2_DONE:
4883 CDEBUG(D_LFSCK, "%s: namespace LFSCK handles notify %u from MDT %x, "
4884 "status %d, flags %x\n", lfsck_lfsck2name(lfsck), lr->lr_event,
4885 lr->lr_index, lr->lr_status, lr->lr_flags2);
4887 spin_lock(<ds->ltd_lock);
4888 ltd = lfsck_ltd2tgt(ltds, lr->lr_index);
4890 spin_unlock(<ds->ltd_lock);
4895 list_del_init(<d->ltd_namespace_phase_list);
4896 switch (lr->lr_event) {
4897 case LE_PHASE1_DONE:
4898 if (lr->lr_status <= 0) {
4899 ltd->ltd_namespace_done = 1;
4900 list_del_init(<d->ltd_namespace_list);
4901 CDEBUG(D_LFSCK, "%s: MDT %x failed/stopped at "
4902 "phase1 for namespace LFSCK: rc = %d.\n",
4903 lfsck_lfsck2name(lfsck),
4904 ltd->ltd_index, lr->lr_status);
4905 ns->ln_flags |= LF_INCOMPLETE;
4910 if (lr->lr_flags2 & LF_INCOMPLETE)
4911 ns->ln_flags |= LF_INCOMPLETE;
4913 if (list_empty(<d->ltd_namespace_list))
4914 list_add_tail(<d->ltd_namespace_list,
4915 &lad->lad_mdt_list);
4916 list_add_tail(<d->ltd_namespace_phase_list,
4917 &lad->lad_mdt_phase2_list);
4919 case LE_PHASE2_DONE:
4920 ltd->ltd_namespace_done = 1;
4921 list_del_init(<d->ltd_namespace_list);
4925 ltd->ltd_namespace_done = 1;
4926 list_del_init(<d->ltd_namespace_list);
4927 if (!(lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT)) {
4929 "%s: the peer MDT %x exit namespace LFSCK\n",
4930 lfsck_lfsck2name(lfsck), ltd->ltd_index);
4931 ns->ln_flags |= LF_INCOMPLETE;
4937 spin_unlock(<ds->ltd_lock);
4939 if (fail && lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) {
4940 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
4942 memset(stop, 0, sizeof(*stop));
4943 stop->ls_status = lr->lr_status;
4944 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
4945 lfsck_stop(env, lfsck->li_bottom, stop);
4946 } else if (lfsck_phase2_next_ready(lad)) {
4947 wake_up_all(&lad->lad_thread.t_ctl_waitq);
4953 static void lfsck_namespace_repaired(struct lfsck_namespace *ns, __u64 *count)
4955 *count += ns->ln_objs_nlink_repaired;
4956 *count += ns->ln_dirent_repaired;
4957 *count += ns->ln_linkea_repaired;
4958 *count += ns->ln_mul_linked_repaired;
4959 *count += ns->ln_unmatched_pairs_repaired;
4960 *count += ns->ln_dangling_repaired;
4961 *count += ns->ln_mul_ref_repaired;
4962 *count += ns->ln_bad_type_repaired;
4963 *count += ns->ln_lost_dirent_repaired;
4964 *count += ns->ln_striped_dirs_disabled;
4965 *count += ns->ln_striped_dirs_repaired;
4966 *count += ns->ln_striped_shards_repaired;
4967 *count += ns->ln_name_hash_repaired;
4968 *count += ns->ln_local_lpf_moved;
4971 static int lfsck_namespace_query_all(const struct lu_env *env,
4972 struct lfsck_component *com,
4973 __u32 *mdts_count, __u64 *repaired)
4975 struct lfsck_namespace *ns = com->lc_file_ram;
4976 struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
4977 struct lfsck_tgt_desc *ltd;
4982 rc = lfsck_query_all(env, com);
4986 down_read(<ds->ltd_rw_sem);
4987 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
4988 ltd = lfsck_ltd2tgt(ltds, idx);
4989 LASSERT(ltd != NULL);
4991 mdts_count[ltd->ltd_namespace_status]++;
4992 *repaired += ltd->ltd_namespace_repaired;
4994 up_read(<ds->ltd_rw_sem);
4996 down_read(&com->lc_sem);
4997 mdts_count[ns->ln_status]++;
4998 lfsck_namespace_repaired(ns, repaired);
4999 up_read(&com->lc_sem);
5004 static int lfsck_namespace_query(const struct lu_env *env,
5005 struct lfsck_component *com,
5006 struct lfsck_request *req,
5007 struct lfsck_reply *rep,
5008 struct lfsck_query *que, int idx)
5010 struct lfsck_namespace *ns = com->lc_file_ram;
5014 LASSERT(com->lc_lfsck->li_master);
5016 rc = lfsck_namespace_query_all(env, com,
5017 que->lu_mdts_count[idx],
5018 &que->lu_repaired[idx]);
5020 down_read(&com->lc_sem);
5021 rep->lr_status = ns->ln_status;
5022 if (req->lr_flags & LEF_QUERY_ALL)
5023 lfsck_namespace_repaired(ns, &rep->lr_repaired);
5024 up_read(&com->lc_sem);
5030 static struct lfsck_operations lfsck_namespace_ops = {
5031 .lfsck_reset = lfsck_namespace_reset,
5032 .lfsck_fail = lfsck_namespace_fail,
5033 .lfsck_close_dir = lfsck_namespace_close_dir,
5034 .lfsck_open_dir = lfsck_namespace_open_dir,
5035 .lfsck_checkpoint = lfsck_namespace_checkpoint,
5036 .lfsck_prep = lfsck_namespace_prep,
5037 .lfsck_exec_oit = lfsck_namespace_exec_oit,
5038 .lfsck_exec_dir = lfsck_namespace_exec_dir,
5039 .lfsck_post = lfsck_namespace_post,
5040 .lfsck_dump = lfsck_namespace_dump,
5041 .lfsck_double_scan = lfsck_namespace_double_scan,
5042 .lfsck_data_release = lfsck_namespace_data_release,
5043 .lfsck_quit = lfsck_namespace_quit,
5044 .lfsck_in_notify = lfsck_namespace_in_notify,
5045 .lfsck_query = lfsck_namespace_query,
5049 * Repair dangling name entry.
5051 * For the name entry with dangling reference, we need to repare the
5052 * inconsistency according to the LFSCK sponsor's requirement:
5054 * 1) Keep the inconsistency there and report the inconsistency case,
5055 * then give the chance to the application to find related issues,
5056 * and the users can make the decision about how to handle it with
5057 * more human knownledge. (by default)
5059 * 2) Re-create the missing MDT-object with the FID information.
5061 * \param[in] env pointer to the thread context
5062 * \param[in] com pointer to the lfsck component
5063 * \param[in] parent pointer to the dir object that contains the dangling
5065 * \param[in] child pointer to the object corresponding to the dangling
5067 * \param[in] lnr pointer to the namespace request that contains the
5068 * name's name, parent object, parent's LMV, and ect.
5070 * \retval positive number if no need to repair
5071 * \retval zero for repaired successfully
5072 * \retval negative error number on failure
5074 int lfsck_namespace_repair_dangling(const struct lu_env *env,
5075 struct lfsck_component *com,
5076 struct dt_object *parent,
5077 struct dt_object *child,
5078 struct lfsck_namespace_req *lnr)
5080 struct lfsck_thread_info *info = lfsck_env_info(env);
5081 struct lu_attr *la = &info->lti_la;
5082 struct dt_allocation_hint *hint = &info->lti_hint;
5083 struct dt_object_format *dof = &info->lti_dof;
5084 struct dt_insert_rec *rec = &info->lti_dt_rec;
5085 struct lmv_mds_md_v1 *lmv2 = &info->lti_lmv2;
5086 const struct lu_name *cname;
5087 const struct lu_fid *pfid = lfsck_dto2fid(parent);
5088 const struct lu_fid *cfid = lfsck_dto2fid(child);
5089 struct linkea_data ldata = { NULL };
5090 struct lfsck_lock_handle *llh = &info->lti_llh;
5091 struct lu_buf linkea_buf;
5092 struct lu_buf lmv_buf;
5093 struct lfsck_instance *lfsck = com->lc_lfsck;
5094 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5095 struct dt_device *dev = lfsck->li_next;
5096 struct thandle *th = NULL;
5098 __u16 type = lnr->lnr_type;
5102 cname = lfsck_name_get_const(env, lnr->lnr_name, lnr->lnr_namelen);
5103 if (bk->lb_param & LPF_CREATE_MDTOBJ)
5108 if (!create || bk->lb_param & LPF_DRYRUN)
5111 /* We may need to create the sub-objects of the @child via LOD,
5112 * so make the modification based on lfsck->li_next. */
5114 parent = lfsck_object_locate(dev, parent);
5116 GOTO(log, rc = PTR_ERR(parent));
5118 if (unlikely(!dt_try_as_dir(env, parent)))
5119 GOTO(log, rc = -ENOTDIR);
5121 child = lfsck_object_locate(dev, child);
5123 GOTO(log, rc = PTR_ERR(child));
5125 rc = linkea_links_new(&ldata, &info->lti_linkea_buf2,
5130 rc = lfsck_lock(env, lfsck, parent, lnr->lnr_name, llh,
5131 MDS_INODELOCK_UPDATE, LCK_PR);
5135 rc = lfsck_namespace_check_exist(env, parent, child, lnr->lnr_name);
5139 /* Set the ctime as zero, then others can know it is created for
5140 * repairing dangling name entry by LFSCK. And if the LFSCK made
5141 * wrong decision and the real MDT-object has been found later,
5142 * then the LFSCK has chance to fix the incosistency properly. */
5143 memset(la, 0, sizeof(*la));
5144 la->la_mode = (type & S_IFMT) | 0600;
5145 la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
5146 LA_ATIME | LA_MTIME | LA_CTIME;
5148 child->do_ops->do_ah_init(env, hint, parent, child,
5149 la->la_mode & S_IFMT);
5151 memset(dof, 0, sizeof(*dof));
5152 dof->dof_type = dt_mode_to_dft(type);
5153 /* If the target is a regular file, then the LFSCK will only create
5154 * the MDT-object without stripes (dof->dof_reg.striped = 0). related
5155 * OST-objects will be created when write open. */
5157 th = dt_trans_create(env, dev);
5159 GOTO(log, rc = PTR_ERR(th));
5161 /* 1a. create child. */
5162 rc = dt_declare_create(env, child, la, hint, dof, th);
5166 if (S_ISDIR(type)) {
5167 if (unlikely(!dt_try_as_dir(env, child)))
5168 GOTO(stop, rc = -ENOTDIR);
5170 /* 2a. increase child nlink */
5171 rc = dt_declare_ref_add(env, child, th);
5175 /* 3a. insert dot into child dir */
5176 rec->rec_type = S_IFDIR;
5177 rec->rec_fid = cfid;
5178 rc = dt_declare_insert(env, child,
5179 (const struct dt_rec *)rec,
5180 (const struct dt_key *)dot, th);
5184 /* 4a. insert dotdot into child dir */
5185 rec->rec_fid = pfid;
5186 rc = dt_declare_insert(env, child,
5187 (const struct dt_rec *)rec,
5188 (const struct dt_key *)dotdot, th);
5192 /* 5a. generate slave LMV EA. */
5193 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
5196 idx = lfsck_shard_name_to_index(env,
5197 lnr->lnr_name, lnr->lnr_namelen,
5199 if (unlikely(idx < 0))
5200 GOTO(stop, rc = idx);
5202 *lmv2 = lnr->lnr_lmv->ll_lmv;
5203 lmv2->lmv_magic = LMV_MAGIC_STRIPE;
5204 lmv2->lmv_master_mdt_index = idx;
5206 lfsck_lmv_header_cpu_to_le(lmv2, lmv2);
5207 lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
5208 rc = dt_declare_xattr_set(env, child, &lmv_buf,
5209 XATTR_NAME_LMV, 0, th);
5215 /* 6a. insert linkEA for child */
5216 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
5217 ldata.ld_leh->leh_len);
5218 rc = dt_declare_xattr_set(env, child, &linkea_buf,
5219 XATTR_NAME_LINK, 0, th);
5223 rc = dt_trans_start_local(env, dev, th);
5225 GOTO(stop, rc = (rc == -EEXIST ? 1 : rc));
5227 dt_write_lock(env, child, 0);
5228 /* 1b. create child */
5229 rc = dt_create(env, child, la, hint, dof, th);
5231 GOTO(unlock, rc = (rc == -EEXIST ? 1 : rc));
5233 if (S_ISDIR(type)) {
5234 /* 2b. increase child nlink */
5235 rc = dt_ref_add(env, child, th);
5239 /* 3b. insert dot into child dir */
5240 rec->rec_type = S_IFDIR;
5241 rec->rec_fid = cfid;
5242 rc = dt_insert(env, child, (const struct dt_rec *)rec,
5243 (const struct dt_key *)dot, th, 1);
5247 /* 4b. insert dotdot into child dir */
5248 rec->rec_fid = pfid;
5249 rc = dt_insert(env, child, (const struct dt_rec *)rec,
5250 (const struct dt_key *)dotdot, th, 1);
5254 /* 5b. generate slave LMV EA. */
5255 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
5256 rc = dt_xattr_set(env, child, &lmv_buf, XATTR_NAME_LMV,
5263 /* 6b. insert linkEA for child. */
5264 rc = dt_xattr_set(env, child, &linkea_buf,
5265 XATTR_NAME_LINK, 0, th);
5270 dt_write_unlock(env, child);
5273 dt_trans_stop(env, dev, th);
5277 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found dangling "
5278 "reference for: parent "DFID", child "DFID", type %u, "
5279 "name %s. %s: rc = %d\n", lfsck_lfsck2name(lfsck),
5280 PFID(pfid), PFID(cfid), type, cname->ln_name,
5281 create ? "Create the lost MDT-object as required" :
5282 "Keep the MDT-object there by default", rc);
5285 struct lfsck_namespace *ns = com->lc_file_ram;
5287 ns->ln_flags |= LF_INCONSISTENT;
5293 static int lfsck_namespace_assistant_handler_p1(const struct lu_env *env,
5294 struct lfsck_component *com,
5295 struct lfsck_assistant_req *lar)
5297 struct lfsck_thread_info *info = lfsck_env_info(env);
5298 struct lu_attr *la = &info->lti_la;
5299 struct lfsck_instance *lfsck = com->lc_lfsck;
5300 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5301 struct lfsck_namespace *ns = com->lc_file_ram;
5302 struct lfsck_assistant_data *lad = com->lc_data;
5303 struct linkea_data ldata = { NULL };
5304 const struct lu_name *cname;
5305 struct thandle *handle = NULL;
5306 struct lfsck_namespace_req *lnr =
5307 container_of0(lar, struct lfsck_namespace_req, lnr_lar);
5308 struct dt_object *dir = NULL;
5309 struct dt_object *obj = NULL;
5310 struct lfsck_assistant_object *lso = lar->lar_parent;
5311 const struct lu_fid *pfid = &lso->lso_fid;
5312 struct dt_device *dev = NULL;
5313 struct lustre_handle lh = { 0 };
5314 bool repaired = false;
5315 bool dtlocked = false;
5319 bool bad_hash = false;
5320 bool bad_linkea = false;
5324 enum lfsck_namespace_inconsistency_type type = LNIT_NONE;
5331 if (lnr->lnr_attr & LUDA_UPGRADE) {
5332 ns->ln_flags |= LF_UPGRADE;
5333 ns->ln_dirent_repaired++;
5335 } else if (lnr->lnr_attr & LUDA_REPAIR) {
5336 ns->ln_flags |= LF_INCONSISTENT;
5337 ns->ln_dirent_repaired++;
5341 if (unlikely(fid_is_zero(&lnr->lnr_fid))) {
5342 if (strcmp(lnr->lnr_name, dotdot) != 0)
5345 rc = lfsck_namespace_trace_update(env, com, pfid,
5346 LNTF_CHECK_PARENT, true);
5351 if (unlikely(!fid_is_sane(&lnr->lnr_fid))) {
5352 CDEBUG(D_LFSCK, "%s: dir scan find invalid FID "DFID
5353 " for the name entry %.*s under "DFID"\n",
5354 lfsck_lfsck2name(lfsck), PFID(&lnr->lnr_fid),
5355 lnr->lnr_namelen, lnr->lnr_name, PFID(pfid));
5357 if (strcmp(lnr->lnr_name, dotdot) != 0)
5358 /* invalid FID means bad name entry, remove it. */
5359 type = LNIT_BAD_DIRENT;
5361 /* If the parent FID is invalid, we cannot remove
5362 * the ".." entry directly. */
5363 rc = lfsck_namespace_trace_update(env, com, pfid,
5364 LNTF_CHECK_PARENT, true);
5369 if (unlikely(lnr->lnr_dir_cookie == MDS_DIR_END_OFF)) {
5370 rc = lfsck_namespace_striped_dir_rescan(env, com, lnr);
5375 if (fid_seq_is_dot(fid_seq(&lnr->lnr_fid)))
5378 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
5379 rc = lfsck_namespace_handle_striped_master(env, com, lnr);
5384 idx = lfsck_find_mdt_idx_by_fid(env, lfsck, &lnr->lnr_fid);
5386 GOTO(out, rc = idx);
5388 if (idx == lfsck_dev_idx(lfsck)) {
5389 if (unlikely(strcmp(lnr->lnr_name, dotdot) == 0))
5392 dev = lfsck->li_bottom;
5394 struct lfsck_tgt_desc *ltd;
5396 /* Usually, some local filesystem consistency verification
5397 * tools can guarantee the local namespace tree consistenct.
5398 * So the LFSCK will only verify the remote directory. */
5399 if (unlikely(strcmp(lnr->lnr_name, dotdot) == 0)) {
5400 rc = lfsck_namespace_trace_update(env, com, pfid,
5401 LNTF_CHECK_PARENT, true);
5406 ltd = lfsck_ltd2tgt(&lfsck->li_mdt_descs, idx);
5407 if (unlikely(ltd == NULL)) {
5408 CDEBUG(D_LFSCK, "%s: cannot talk with MDT %x which "
5409 "did not join the namespace LFSCK\n",
5410 lfsck_lfsck2name(lfsck), idx);
5411 lfsck_lad_set_bitmap(env, com, idx);
5413 GOTO(out, rc = -ENODEV);
5419 obj = lfsck_object_find_by_dev(env, dev, &lnr->lnr_fid);
5421 GOTO(out, rc = PTR_ERR(obj));
5423 cname = lfsck_name_get_const(env, lnr->lnr_name, lnr->lnr_namelen);
5424 if (dt_object_exists(obj) == 0) {
5428 dir = lfsck_assistant_object_load(env, lfsck, lso);
5432 GOTO(trace, rc == -ENOENT ? 0 : rc);
5436 rc = lfsck_namespace_check_exist(env, dir, obj, lnr->lnr_name);
5438 if (!lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5439 lnr->lnr_name, lnr->lnr_namelen)) {
5440 type = LNIT_BAD_DIRENT;
5445 type = LNIT_DANGLING;
5446 rc = lfsck_namespace_repair_dangling(env, com, dir,
5455 if (!(bk->lb_param & LPF_DRYRUN) && lad->lad_advance_lock) {
5458 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
5459 MDS_INODELOCK_UPDATE |
5460 MDS_INODELOCK_XATTR, LCK_EX);
5464 handle = dt_trans_create(env, dev);
5466 GOTO(out, rc = PTR_ERR(handle));
5468 rc = lfsck_declare_namespace_exec_dir(env, obj, handle);
5472 rc = dt_trans_start_local(env, dev, handle);
5476 dt_write_lock(env, obj, 0);
5480 rc = lfsck_links_read(env, obj, &ldata);
5481 if (unlikely(rc == -ENOENT)) {
5482 if (handle != NULL) {
5483 dt_write_unlock(env, obj);
5486 dt_trans_stop(env, dev, handle);
5489 lfsck_ibits_unlock(&lh, LCK_EX);
5492 /* It may happen when the remote object has been removed,
5493 * but the local MDT is not aware of that. */
5495 } else if (rc == 0) {
5496 count = ldata.ld_leh->leh_reccount;
5497 rc = linkea_links_find(&ldata, cname, pfid);
5499 (count == 1 || !S_ISDIR(lfsck_object_type(obj)))) {
5500 if ((lfsck_object_type(obj) & S_IFMT) !=
5502 ns->ln_flags |= LF_INCONSISTENT;
5503 type = LNIT_BAD_TYPE;
5509 /* If the name entry hash does not match the slave striped
5510 * directory, and the name entry does not match also, then
5511 * it is quite possible that name entry is corrupted. */
5512 if (!lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5513 lnr->lnr_name, lnr->lnr_namelen)) {
5514 ns->ln_flags |= LF_INCONSISTENT;
5515 type = LNIT_BAD_DIRENT;
5520 /* If the file type stored in the name entry does not match
5521 * the file type claimed by the object, and the object does
5522 * not recognize the name entry, then it is quite possible
5523 * that the name entry is corrupted. */
5524 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type) {
5525 ns->ln_flags |= LF_INCONSISTENT;
5526 type = LNIT_BAD_DIRENT;
5531 /* For sub-dir object, we cannot make sure whether the sub-dir
5532 * back references the parent via ".." name entry correctly or
5533 * not in the LFSCK first-stage scanning. It may be that the
5534 * (remote) sub-dir ".." name entry has no parent FID after
5535 * file-level backup/restore and its linkEA may be wrong.
5536 * So under such case, we should replace the linkEA according
5537 * to current name entry. But this needs to be done during the
5538 * LFSCK second-stage scanning. The LFSCK will record the name
5539 * entry for further possible using. */
5543 } else if (unlikely(rc == -EINVAL)) {
5544 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type)
5545 type = LNIT_BAD_TYPE;
5548 /* The magic crashed, we are not sure whether there are more
5549 * corrupt data in the linkea, so remove all linkea entries. */
5553 } else if (rc == -ENODATA) {
5554 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type)
5555 type = LNIT_BAD_TYPE;
5562 if (bk->lb_param & LPF_DRYRUN) {
5564 ns->ln_flags |= LF_UPGRADE;
5566 ns->ln_flags |= LF_INCONSISTENT;
5567 ns->ln_linkea_repaired++;
5573 if (!lustre_handle_is_used(&lh)) {
5581 LASSERT(handle != NULL);
5584 dir = lfsck_assistant_object_load(env, lfsck, lso);
5588 GOTO(stop, rc == -ENOENT ? 0 : rc);
5592 rc = lfsck_namespace_check_exist(env, dir, obj, lnr->lnr_name);
5597 if (!remove && newdata)
5598 ns->ln_flags |= LF_UPGRADE;
5599 else if (remove || !(ns->ln_flags & LF_UPGRADE))
5600 ns->ln_flags |= LF_INCONSISTENT;
5605 rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, handle);
5606 if (rc != 0 && rc != -ENOENT && rc != -ENODATA)
5611 rc = linkea_data_new(&ldata,
5612 &lfsck_env_info(env)->lti_linkea_buf);
5617 rc = linkea_add_buf(&ldata, cname, pfid);
5619 rc = lfsck_links_write(env, obj, &ldata, handle);
5623 count = ldata.ld_leh->leh_reccount;
5624 if (!S_ISDIR(lfsck_object_type(obj)) ||
5625 !dt_object_remote(obj)) {
5626 ns->ln_linkea_repaired++;
5636 dt_write_unlock(env, obj);
5638 if (handle != NULL && !IS_ERR(handle))
5639 dt_trans_stop(env, dev, handle);
5642 lfsck_ibits_unlock(&lh, LCK_EX);
5644 if (!name_is_dot_or_dotdot(lnr->lnr_name, lnr->lnr_namelen) &&
5645 !lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5646 lnr->lnr_name, lnr->lnr_namelen) &&
5647 type != LNIT_BAD_DIRENT) {
5648 ns->ln_flags |= LF_INCONSISTENT;
5652 dir = lfsck_assistant_object_load(env, lfsck, lso);
5656 GOTO(trace, rc == -ENOENT ? 0 : rc);
5660 rc = lfsck_namespace_repair_bad_name_hash(env, com, dir,
5661 lnr->lnr_lmv, lnr->lnr_name);
5667 if (type != LNIT_NONE && dir == NULL) {
5668 dir = lfsck_assistant_object_load(env, lfsck, lso);
5672 GOTO(trace, rc == -ENOENT ? 0 : rc);
5679 rc = lfsck_namespace_repair_dirent(env, com, dir,
5680 obj, lnr->lnr_name, lnr->lnr_name,
5681 lnr->lnr_type, true, false);
5685 case LNIT_BAD_DIRENT:
5687 /* XXX: This is a bad dirent, we do not know whether
5688 * the original name entry reference a regular
5689 * file or a directory, then keep the parent's
5690 * nlink count unchanged here. */
5691 rc = lfsck_namespace_repair_dirent(env, com, dir,
5692 obj, lnr->lnr_name, lnr->lnr_name,
5693 lnr->lnr_type, false, false);
5701 if (obj != NULL && count == 1 &&
5702 S_ISREG(lfsck_object_type(obj)))
5703 dt_attr_get(env, obj, la);
5707 down_write(&com->lc_sem);
5709 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail to handle "
5710 "the entry: "DFID", parent "DFID", name %.*s: rc = %d\n",
5711 lfsck_lfsck2name(lfsck), PFID(&lnr->lnr_fid), PFID(pfid),
5712 lnr->lnr_namelen, lnr->lnr_name, rc);
5714 lfsck_namespace_record_failure(env, lfsck, ns);
5715 if ((rc == -ENOTCONN || rc == -ESHUTDOWN || rc == -EREMCHG ||
5716 rc == -ETIMEDOUT || rc == -EHOSTDOWN ||
5717 rc == -EHOSTUNREACH || rc == -EINPROGRESS) &&
5718 dev != NULL && dev != lfsck->li_bottom)
5719 lfsck_lad_set_bitmap(env, com, idx);
5721 if (!(bk->lb_param & LPF_FAILOUT))
5725 ns->ln_items_repaired++;
5727 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant "
5728 "repaired the entry: "DFID", parent "DFID
5729 ", name %.*s, type %d\n",
5730 lfsck_lfsck2name(lfsck),
5731 PFID(&lnr->lnr_fid), PFID(pfid),
5732 lnr->lnr_namelen, lnr->lnr_name, type);
5736 ns->ln_dangling_repaired++;
5739 ns->ln_bad_type_repaired++;
5741 case LNIT_BAD_DIRENT:
5742 ns->ln_dirent_repaired++;
5748 if (bk->lb_param & LPF_DRYRUN &&
5749 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent))
5750 lfsck_pos_fill(env, lfsck,
5751 &ns->ln_pos_first_inconsistent,
5756 ns->ln_name_hash_repaired++;
5758 /* Not count repeatedly. */
5760 ns->ln_items_repaired++;
5762 CDEBUG(D_LFSCK, "%s: namespace LFSCK "
5763 "assistant repaired the entry: "
5766 lfsck_lfsck2name(lfsck),
5767 PFID(&lnr->lnr_fid), PFID(pfid),
5768 lnr->lnr_namelen, lnr->lnr_name);
5771 if (bk->lb_param & LPF_DRYRUN &&
5772 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent))
5773 lfsck_pos_fill(env, lfsck,
5774 &ns->ln_pos_first_inconsistent,
5781 if (count > 1 || la->la_nlink > 1)
5782 ns->ln_mul_linked_checked++;
5784 up_write(&com->lc_sem);
5786 if (obj != NULL && !IS_ERR(obj))
5787 lfsck_object_put(env, obj);
5789 if (dir != NULL && !IS_ERR(dir))
5790 lfsck_object_put(env, dir);
5792 lad->lad_advance_lock = bad_linkea;
5798 * Handle one orphan under the backend /lost+found directory
5800 * Insert the orphan FID into the namespace LFSCK trace file for further
5801 * processing (via the subsequent namespace LFSCK second-stage scanning).
5802 * At the same time, remove the orphan name entry from backend /lost+found
5803 * directory. There is an interval between the orphan name entry removed
5804 * from the backend /lost+found directory and the orphan FID in the LFSCK
5805 * trace file handled. In such interval, the LFSCK can be reset, then
5806 * all the FIDs recorded in the namespace LFSCK trace file will be dropped.
5807 * To guarantee that the orphans can be found when LFSCK run next time
5808 * without e2fsck again, when remove the orphan name entry, the LFSCK
5809 * will set the orphan's ctime attribute as 1. Since normal applications
5810 * cannot change the object's ctime attribute as 1. Then when LFSCK run
5811 * next time, it can record the object (that ctime is 1) in the namespace
5812 * LFSCK trace file during the first-stage scanning.
5814 * \param[in] env pointer to the thread context
5815 * \param[in] com pointer to the lfsck component
5816 * \param[in] parent pointer to the object for the backend /lost+found
5817 * \param[in] ent pointer to the name entry for the target under the
5818 * backend /lost+found
5820 * \retval positive for repaired
5821 * \retval 0 if needs to repair nothing
5822 * \retval negative error number on failure
5824 static int lfsck_namespace_scan_local_lpf_one(const struct lu_env *env,
5825 struct lfsck_component *com,
5826 struct dt_object *parent,
5827 struct lu_dirent *ent)
5829 struct lfsck_thread_info *info = lfsck_env_info(env);
5830 struct lu_fid *key = &info->lti_fid;
5831 struct lu_attr *la = &info->lti_la;
5832 struct lfsck_instance *lfsck = com->lc_lfsck;
5833 struct dt_object *obj;
5834 struct dt_device *dev = lfsck->li_bottom;
5835 struct dt_object *child = NULL;
5836 struct thandle *th = NULL;
5843 child = lfsck_object_find_by_dev(env, dev, &ent->lde_fid);
5845 RETURN(PTR_ERR(child));
5847 LASSERT(dt_object_exists(child));
5848 LASSERT(!dt_object_remote(child));
5850 idx = lfsck_sub_trace_file_fid2idx(&ent->lde_fid);
5851 obj = com->lc_sub_trace_objs[idx].lsto_obj;
5852 fid_cpu_to_be(key, &ent->lde_fid);
5853 rc = dt_lookup(env, obj, (struct dt_rec *)&flags,
5854 (const struct dt_key *)key);
5857 flags |= LNTF_CHECK_ORPHAN;
5858 } else if (rc == -ENOENT) {
5859 flags = LNTF_CHECK_ORPHAN;
5864 th = dt_trans_create(env, dev);
5866 GOTO(out, rc = PTR_ERR(th));
5868 /* a1. remove name entry from backend /lost+found */
5869 rc = dt_declare_delete(env, parent,
5870 (const struct dt_key *)ent->lde_name, th);
5874 if (S_ISDIR(lfsck_object_type(child))) {
5875 /* a2. decrease parent's nlink */
5876 rc = dt_declare_ref_del(env, parent, th);
5882 /* a3. remove child's FID from the LFSCK trace file. */
5883 rc = dt_declare_delete(env, obj,
5884 (const struct dt_key *)key, th);
5888 /* a4. set child's ctime as 1 */
5889 memset(la, 0, sizeof(*la));
5891 la->la_valid = LA_CTIME;
5892 rc = dt_declare_attr_set(env, child, la, th);
5897 /* a5. insert child's FID into the LFSCK trace file. */
5898 rc = dt_declare_insert(env, obj, (const struct dt_rec *)&flags,
5899 (const struct dt_key *)key, th);
5903 rc = dt_trans_start_local(env, dev, th);
5907 /* b1. remove name entry from backend /lost+found */
5908 rc = dt_delete(env, parent, (const struct dt_key *)ent->lde_name, th);
5912 if (S_ISDIR(lfsck_object_type(child))) {
5913 /* b2. decrease parent's nlink */
5914 dt_write_lock(env, parent, 0);
5915 rc = dt_ref_del(env, parent, th);
5916 dt_write_unlock(env, parent);
5922 /* a3. remove child's FID from the LFSCK trace file. */
5923 rc = dt_delete(env, obj, (const struct dt_key *)key, th);
5927 /* b4. set child's ctime as 1 */
5928 rc = dt_attr_set(env, child, la, th);
5933 /* b5. insert child's FID into the LFSCK trace file. */
5934 rc = dt_insert(env, obj, (const struct dt_rec *)&flags,
5935 (const struct dt_key *)key, th, 1);
5937 GOTO(stop, rc = (rc == 0 ? 1 : rc));
5940 dt_trans_stop(env, dev, th);
5943 lfsck_object_put(env, child);
5949 * Handle orphans under the backend /lost+found directory
5951 * Some backend checker, such as e2fsck for ldiskfs may find some orphans
5952 * and put them under the backend /lost+found directory that is invisible
5953 * to client. The LFSCK will scan such directory, for the original client
5954 * visible orphans, add their fids into the namespace LFSCK trace file,
5955 * then the subsenquent namespace LFSCK second-stage scanning can handle
5956 * them as other objects to be double scanned: either move back to normal
5957 * namespace, or to the global visible orphan directory:
5958 * /ROOT/.lustre/lost+found/MDTxxxx/
5960 * \param[in] env pointer to the thread context
5961 * \param[in] com pointer to the lfsck component
5963 static void lfsck_namespace_scan_local_lpf(const struct lu_env *env,
5964 struct lfsck_component *com)
5966 struct lfsck_thread_info *info = lfsck_env_info(env);
5967 struct lu_dirent *ent =
5968 (struct lu_dirent *)info->lti_key;
5969 struct lu_seq_range *range = &info->lti_range;
5970 struct lfsck_instance *lfsck = com->lc_lfsck;
5971 struct ptlrpc_thread *thread = &lfsck->li_thread;
5972 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5973 struct lfsck_namespace *ns = com->lc_file_ram;
5974 struct dt_object *parent;
5975 const struct dt_it_ops *iops;
5977 struct seq_server_site *ss = lfsck_dev_site(lfsck);
5979 __u32 idx = lfsck_dev_idx(lfsck);
5984 parent = lfsck_object_find_by_dev(env, lfsck->li_bottom,
5985 &LU_BACKEND_LPF_FID);
5986 if (IS_ERR(parent)) {
5987 CERROR("%s: fail to find backend /lost+found: rc = %ld\n",
5988 lfsck_lfsck2name(lfsck), PTR_ERR(parent));
5992 /* It is normal that the /lost+found does not exist for ZFS backend. */
5993 if (!dt_object_exists(parent))
5996 if (unlikely(!dt_try_as_dir(env, parent)))
5997 GOTO(out, rc = -ENOTDIR);
5999 CDEBUG(D_LFSCK, "%s: start to scan backend /lost+found\n",
6000 lfsck_lfsck2name(lfsck));
6002 com->lc_new_scanned = 0;
6003 iops = &parent->do_index_ops->dio_it;
6004 di = iops->init(env, parent, LUDA_64BITHASH | LUDA_TYPE);
6006 GOTO(out, rc = PTR_ERR(di));
6008 rc = iops->load(env, di, 0);
6010 rc = iops->next(env, di);
6015 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
6016 unlikely(!thread_is_running(thread)))
6019 rc = iops->rec(env, di, (struct dt_rec *)ent,
6020 LUDA_64BITHASH | LUDA_TYPE);
6022 rc = lfsck_unpack_ent(ent, &cookie, &type);
6024 if (unlikely(rc != 0)) {
6025 CDEBUG(D_LFSCK, "%s: fail to iterate backend "
6026 "/lost+found: rc = %d\n",
6027 lfsck_lfsck2name(lfsck), rc);
6032 /* skip dot and dotdot entries */
6033 if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen))
6036 if (!fid_seq_in_fldb(fid_seq(&ent->lde_fid)))
6039 if (fid_is_norm(&ent->lde_fid)) {
6040 fld_range_set_mdt(range);
6041 rc = fld_local_lookup(env, ss->ss_server_fld,
6042 fid_seq(&ent->lde_fid), range);
6045 } else if (idx != 0) {
6046 /* If the returned FID is IGIF, then there are three
6049 * 1) The object is upgraded from old Lustre-1.8 with
6050 * IGIF assigned to such object.
6051 * 2) The object is a backend local object and is
6052 * invisible to client.
6053 * 3) The object lost its LMV EA, and since there is
6054 * no FID-in-dirent for the orphan in the backend
6055 * /lost+found directory, then the low layer will
6056 * return IGIF for such object.
6058 * For MDTx (x != 0), it is either case 2) or case 3),
6059 * but from the LFSCK view, they are indistinguishable.
6060 * To be safe, the LFSCK will keep it there and report
6061 * some message, then the adminstrator can handle that
6064 * For MDT0, it is more possible the case 1). The LFSCK
6065 * will handle the orphan as an upgraded object. */
6066 CDEBUG(D_LFSCK, "%s: the orphan %.*s with IGIF "DFID
6067 "in the backend /lost+found on the MDT %04x, "
6068 "to be safe, skip it.\n",
6069 lfsck_lfsck2name(lfsck), ent->lde_namelen,
6070 ent->lde_name, PFID(&ent->lde_fid), idx);
6074 rc = lfsck_namespace_scan_local_lpf_one(env, com, parent, ent);
6077 down_write(&com->lc_sem);
6078 com->lc_new_scanned++;
6079 ns->ln_local_lpf_scanned++;
6081 ns->ln_local_lpf_moved++;
6083 ns->ln_local_lpf_skipped++;
6085 ns->ln_local_lpf_failed++;
6086 up_write(&com->lc_sem);
6088 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
6092 lfsck_control_speed_by_self(com);
6093 if (unlikely(!thread_is_running(thread))) {
6098 rc = iops->next(env, di);
6102 iops->fini(env, di);
6107 CDEBUG(D_LFSCK, "%s: stop to scan backend /lost+found: rc = %d\n",
6108 lfsck_lfsck2name(lfsck), rc);
6110 lfsck_object_put(env, parent);
6114 * Rescan the striped directory after the master LMV EA reset.
6116 * Sometimes, the master LMV EA of the striped directory maybe lost, so when
6117 * the namespace LFSCK engine scan the striped directory for the first time,
6118 * it will be regarded as a normal directory. As the LFSCK processing, some
6119 * other LFSCK instance on other MDT will find the shard of this striped dir,
6120 * and find that the master MDT-object of the striped directory lost its LMV
6121 * EA, then such remote LFSCK instance will regenerate the master LMV EA and
6122 * notify the LFSCK instance on this MDT to rescan the striped directory.
6124 * \param[in] env pointer to the thread context
6125 * \param[in] com pointer to the lfsck component
6126 * \param[in] llu the lfsck_lmv_unit that contains the striped directory
6129 * \retval positive number for success
6130 * \retval 0 for LFSCK stopped/paused
6131 * \retval negative error number on failure
6133 static int lfsck_namespace_rescan_striped_dir(const struct lu_env *env,
6134 struct lfsck_component *com,
6135 struct lfsck_lmv_unit *llu)
6137 struct lfsck_thread_info *info = lfsck_env_info(env);
6138 struct lfsck_instance *lfsck = com->lc_lfsck;
6139 struct lfsck_assistant_data *lad = com->lc_data;
6140 struct dt_object *dir;
6141 const struct dt_it_ops *iops;
6143 struct lu_dirent *ent =
6144 (struct lu_dirent *)info->lti_key;
6145 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
6146 struct ptlrpc_thread *thread = &lfsck->li_thread;
6147 struct lfsck_assistant_object *lso = NULL;
6148 struct lfsck_namespace_req *lnr;
6149 struct lfsck_assistant_req *lar;
6154 LASSERT(list_empty(&lad->lad_req_list));
6156 lfsck->li_lmv = &llu->llu_lmv;
6157 lfsck->li_obj_dir = lfsck_object_get(llu->llu_obj);
6158 rc = lfsck_open_dir(env, lfsck, 0);
6162 dir = lfsck->li_obj_dir;
6163 di = lfsck->li_di_dir;
6164 iops = &dir->do_index_ops->dio_it;
6166 rc = iops->rec(env, di, (struct dt_rec *)ent,
6167 lfsck->li_args_dir);
6169 rc = lfsck_unpack_ent(ent, &lfsck->li_cookie_dir,
6173 if (bk->lb_param & LPF_FAILOUT)
6179 if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen))
6183 lso = lfsck_assistant_object_init(env,
6184 lfsck_dto2fid(dir), NULL,
6185 lfsck->li_pos_current.lp_oit_cookie, true);
6187 if (bk->lb_param & LPF_FAILOUT)
6188 GOTO(out, rc = PTR_ERR(lso));
6195 lnr = lfsck_namespace_assistant_req_init(lfsck, lso, ent, type);
6197 if (bk->lb_param & LPF_FAILOUT)
6198 GOTO(out, rc = PTR_ERR(lnr));
6203 lar = &lnr->lnr_lar;
6204 rc = lfsck_namespace_assistant_handler_p1(env, com, lar);
6205 lfsck_namespace_assistant_req_fini(env, lar);
6206 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
6209 if (unlikely(!thread_is_running(thread)))
6213 rc = iops->next(env, di);
6217 if (lso != NULL && !IS_ERR(lso))
6218 lfsck_assistant_object_put(env, lso);
6220 lfsck_close_dir(env, lfsck, rc);
6224 /* The close_dir() may insert a dummy lnr in the lad->lad_req_list. */
6225 if (list_empty(&lad->lad_req_list))
6228 spin_lock(&lad->lad_lock);
6229 lar = list_entry(lad->lad_req_list.next, struct lfsck_assistant_req,
6231 list_del_init(&lar->lar_list);
6232 spin_unlock(&lad->lad_lock);
6234 rc = lfsck_namespace_assistant_handler_p1(env, com, lar);
6235 lfsck_namespace_assistant_req_fini(env, lar);
6237 RETURN(rc == 0 ? 1 : rc);
6241 lfsck_namespace_double_scan_one_trace_file(const struct lu_env *env,
6242 struct lfsck_component *com,
6243 struct dt_object *obj, bool first)
6245 struct lfsck_instance *lfsck = com->lc_lfsck;
6246 struct ptlrpc_thread *thread = &lfsck->li_thread;
6247 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
6248 struct lfsck_namespace *ns = com->lc_file_ram;
6249 const struct dt_it_ops *iops = &obj->do_index_ops->dio_it;
6250 struct dt_object *target;
6258 di = iops->init(env, obj, 0);
6260 RETURN(PTR_ERR(di));
6263 fid_cpu_to_be(&fid, &ns->ln_fid_latest_scanned_phase2);
6266 rc = iops->get(env, di, (const struct dt_key *)&fid);
6271 /* The start one either has been processed or does not exist,
6273 rc = iops->next(env, di);
6279 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
6280 unlikely(!thread_is_running(thread)))
6283 key = iops->key(env, di);
6292 fid_be_to_cpu(&fid, (const struct lu_fid *)key);
6293 if (!fid_is_sane(&fid)) {
6298 target = lfsck_object_find_bottom(env, lfsck, &fid);
6299 if (IS_ERR(target)) {
6300 rc = PTR_ERR(target);
6304 if (dt_object_exists(target)) {
6305 rc = iops->rec(env, di, (struct dt_rec *)&flags, 0);
6307 rc = lfsck_namespace_double_scan_one(env, com,
6314 lfsck_object_put(env, target);
6317 down_write(&com->lc_sem);
6318 com->lc_new_checked++;
6319 com->lc_new_scanned++;
6321 ns->ln_fid_latest_scanned_phase2 = fid;
6324 ns->ln_objs_repaired_phase2++;
6326 ns->ln_objs_failed_phase2++;
6327 up_write(&com->lc_sem);
6329 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
6332 if (unlikely(com->lc_time_next_checkpoint <=
6333 ktime_get_seconds()) &&
6334 com->lc_new_checked != 0) {
6335 down_write(&com->lc_sem);
6336 ns->ln_run_time_phase2 += ktime_get_seconds() -
6337 com->lc_time_last_checkpoint;
6338 ns->ln_time_last_checkpoint = ktime_get_real_seconds();
6339 ns->ln_objs_checked_phase2 += com->lc_new_checked;
6340 com->lc_new_checked = 0;
6341 lfsck_namespace_store(env, com);
6342 up_write(&com->lc_sem);
6344 com->lc_time_last_checkpoint = ktime_get_seconds();
6345 com->lc_time_next_checkpoint =
6346 com->lc_time_last_checkpoint +
6347 LFSCK_CHECKPOINT_INTERVAL;
6350 lfsck_control_speed_by_self(com);
6351 if (unlikely(!thread_is_running(thread)))
6354 rc = iops->next(env, di);
6363 iops->fini(env, di);
6368 static int lfsck_namespace_assistant_handler_p2(const struct lu_env *env,
6369 struct lfsck_component *com)
6371 struct lfsck_instance *lfsck = com->lc_lfsck;
6372 struct lfsck_namespace *ns = com->lc_file_ram;
6377 while (!list_empty(&lfsck->li_list_lmv)) {
6378 struct lfsck_lmv_unit *llu;
6380 spin_lock(&lfsck->li_lock);
6381 llu = list_entry(lfsck->li_list_lmv.next,
6382 struct lfsck_lmv_unit, llu_link);
6383 list_del_init(&llu->llu_link);
6384 spin_unlock(&lfsck->li_lock);
6386 rc = lfsck_namespace_rescan_striped_dir(env, com, llu);
6391 CDEBUG(D_LFSCK, "%s: namespace LFSCK phase2 scan start\n",
6392 lfsck_lfsck2name(lfsck));
6394 lfsck_namespace_scan_local_lpf(env, com);
6396 com->lc_new_checked = 0;
6397 com->lc_new_scanned = 0;
6398 com->lc_time_last_checkpoint = ktime_get_seconds();
6399 com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
6400 LFSCK_CHECKPOINT_INTERVAL;
6402 i = lfsck_sub_trace_file_fid2idx(&ns->ln_fid_latest_scanned_phase2);
6403 rc = lfsck_namespace_double_scan_one_trace_file(env, com,
6404 com->lc_sub_trace_objs[i].lsto_obj, true);
6405 while (rc > 0 && ++i < LFSCK_STF_COUNT)
6406 rc = lfsck_namespace_double_scan_one_trace_file(env, com,
6407 com->lc_sub_trace_objs[i].lsto_obj, false);
6409 CDEBUG(D_LFSCK, "%s: namespace LFSCK phase2 scan stop at the No. %d "
6410 "trace file: rc = %d\n", lfsck_lfsck2name(lfsck), i, rc);
6415 static void lfsck_namespace_assistant_fill_pos(const struct lu_env *env,
6416 struct lfsck_component *com,
6417 struct lfsck_position *pos)
6419 struct lfsck_assistant_data *lad = com->lc_data;
6420 struct lfsck_namespace_req *lnr;
6422 if (((struct lfsck_namespace *)(com->lc_file_ram))->ln_status !=
6426 if (list_empty(&lad->lad_req_list))
6429 lnr = list_entry(lad->lad_req_list.next,
6430 struct lfsck_namespace_req,
6432 pos->lp_oit_cookie = lnr->lnr_lar.lar_parent->lso_oit_cookie;
6433 pos->lp_dir_cookie = lnr->lnr_dir_cookie - 1;
6434 pos->lp_dir_parent = lnr->lnr_lar.lar_parent->lso_fid;
6437 static int lfsck_namespace_double_scan_result(const struct lu_env *env,
6438 struct lfsck_component *com,
6441 struct lfsck_instance *lfsck = com->lc_lfsck;
6442 struct lfsck_namespace *ns = com->lc_file_ram;
6444 down_write(&com->lc_sem);
6445 ns->ln_run_time_phase2 += ktime_get_seconds() -
6446 com->lc_time_last_checkpoint;
6447 ns->ln_time_last_checkpoint = ktime_get_real_seconds();
6448 ns->ln_objs_checked_phase2 += com->lc_new_checked;
6449 com->lc_new_checked = 0;
6452 if (ns->ln_flags & LF_INCOMPLETE)
6453 ns->ln_status = LS_PARTIAL;
6455 ns->ln_status = LS_COMPLETED;
6456 ns->ln_flags &= ~LF_SCANNED_ONCE;
6457 if (!(lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN))
6458 ns->ln_flags &= ~LF_INCONSISTENT;
6459 ns->ln_time_last_complete = ns->ln_time_last_checkpoint;
6460 ns->ln_success_count++;
6461 } else if (rc == 0) {
6462 if (lfsck->li_status != 0)
6463 ns->ln_status = lfsck->li_status;
6465 ns->ln_status = LS_STOPPED;
6467 ns->ln_status = LS_FAILED;
6470 rc = lfsck_namespace_store(env, com);
6471 up_write(&com->lc_sem);
6477 lfsck_namespace_assistant_sync_failures_interpret(const struct lu_env *env,
6478 struct ptlrpc_request *req,
6482 struct lfsck_async_interpret_args *laia = args;
6483 struct lfsck_tgt_desc *ltd = laia->laia_ltd;
6485 ltd->ltd_synced_failures = 1;
6492 * Notify remote LFSCK instances about former failures.
6494 * The local LFSCK instance has recorded which MDTs have ever failed to respond
6495 * some LFSCK verification requests (maybe because of network issues or the MDT
6496 * itself trouble). During the respond gap the MDT may missed some name entries
6497 * verification, then the MDT cannot know whether related MDT-objects have been
6498 * referenced by related name entries or not, then in the second-stage scanning,
6499 * these MDT-objects will be regarded as orphan, if the MDT-object contains bad
6500 * linkEA for back reference, then it will misguide the LFSCK to generate wrong
6501 * name entry for repairing the orphan.
6503 * To avoid above trouble, when layout LFSCK finishes the first-stage scanning,
6504 * it will scan the bitmap for the ever failed MDTs, and notify them that they
6505 * have ever missed some name entries verification and should skip the handling
6506 * for orphan MDT-objects.
6508 * \param[in] env pointer to the thread context
6509 * \param[in] com pointer to the lfsck component
6510 * \param[in] lr pointer to the lfsck request
6512 static void lfsck_namespace_assistant_sync_failures(const struct lu_env *env,
6513 struct lfsck_component *com,
6514 struct lfsck_request *lr)
6516 struct lfsck_async_interpret_args *laia =
6517 &lfsck_env_info(env)->lti_laia2;
6518 struct lfsck_assistant_data *lad = com->lc_data;
6519 struct lfsck_namespace *ns = com->lc_file_ram;
6520 struct lfsck_instance *lfsck = com->lc_lfsck;
6521 struct lfsck_tgt_descs *ltds = &lfsck->li_mdt_descs;
6522 struct lfsck_tgt_desc *ltd;
6523 struct ptlrpc_request_set *set;
6528 if (!lad->lad_incomplete)
6531 set = ptlrpc_prep_set();
6533 GOTO(out, rc = -ENOMEM);
6535 lr->lr_flags2 = ns->ln_flags | LF_INCOMPLETE;
6536 memset(laia, 0, sizeof(*laia));
6537 lad->lad_touch_gen++;
6539 down_read(<ds->ltd_rw_sem);
6540 cfs_foreach_bit(lad->lad_bitmap, idx) {
6541 ltd = lfsck_ltd2tgt(ltds, idx);
6545 laia->laia_ltd = ltd;
6546 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
6547 lfsck_namespace_assistant_sync_failures_interpret,
6548 laia, LFSCK_NOTIFY);
6550 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail "
6551 "to sync failure with MDT %x: rc = %d\n",
6552 lfsck_lfsck2name(lfsck), ltd->ltd_index, rc);
6554 up_read(<ds->ltd_rw_sem);
6556 rc = ptlrpc_set_wait(set);
6557 ptlrpc_set_destroy(set);
6563 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail "
6564 "to sync failure with MDTs, and related MDTs "
6565 "may handle orphan improperly: rc = %d\n",
6566 lfsck_lfsck2name(lfsck), rc);
6571 struct lfsck_assistant_operations lfsck_namespace_assistant_ops = {
6572 .la_handler_p1 = lfsck_namespace_assistant_handler_p1,
6573 .la_handler_p2 = lfsck_namespace_assistant_handler_p2,
6574 .la_fill_pos = lfsck_namespace_assistant_fill_pos,
6575 .la_double_scan_result = lfsck_namespace_double_scan_result,
6576 .la_req_fini = lfsck_namespace_assistant_req_fini,
6577 .la_sync_failures = lfsck_namespace_assistant_sync_failures,
6581 * Verify the specified linkEA entry for the given directory object.
6582 * If the object has no such linkEA entry or it has more other linkEA
6583 * entries, then re-generate the linkEA with the given information.
6585 * \param[in] env pointer to the thread context
6586 * \param[in] obj pointer to the dt_object to be handled
6587 * \param[in] cname the name for the child in the parent directory
6588 * \param[in] pfid the parent directory's FID for the linkEA
6590 * \retval 0 for success
6591 * \retval negative error number on failure
6593 int lfsck_verify_linkea(const struct lu_env *env, struct dt_object *obj,
6594 const struct lu_name *cname, const struct lu_fid *pfid)
6596 struct dt_device *dev = lfsck_obj2dev(obj);
6597 struct linkea_data ldata = { NULL };
6598 struct lu_buf linkea_buf;
6601 int fl = LU_XATTR_CREATE;
6605 LASSERT(S_ISDIR(lfsck_object_type(obj)));
6607 rc = lfsck_links_read_with_rec(env, obj, &ldata);
6608 if (rc == -ENODATA) {
6610 } else if (rc == 0) {
6611 fl = LU_XATTR_REPLACE;
6612 if (ldata.ld_leh->leh_reccount != 1) {
6615 rc = linkea_links_find(&ldata, cname, pfid);
6624 rc = linkea_links_new(&ldata, &lfsck_env_info(env)->lti_linkea_buf,
6629 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
6630 ldata.ld_leh->leh_len);
6631 th = dt_trans_create(env, dev);
6633 RETURN(PTR_ERR(th));
6635 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
6636 XATTR_NAME_LINK, fl, th);
6640 rc = dt_trans_start_local(env, dev, th);
6644 dt_write_lock(env, obj, 0);
6645 rc = dt_xattr_set(env, obj, &linkea_buf,
6646 XATTR_NAME_LINK, fl, th);
6647 dt_write_unlock(env, obj);
6652 dt_trans_stop(env, dev, th);
6657 * Get the name and parent directory's FID from the first linkEA entry.
6659 * \param[in] env pointer to the thread context
6660 * \param[in] obj pointer to the object which get linkEA from
6661 * \param[out] name pointer to the buffer to hold the name
6662 * in the first linkEA entry
6663 * \param[out] pfid pointer to the buffer to hold the parent
6664 * directory's FID in the first linkEA entry
6666 * \retval 0 for success
6667 * \retval negative error number on failure
6669 int lfsck_links_get_first(const struct lu_env *env, struct dt_object *obj,
6670 char *name, struct lu_fid *pfid)
6672 struct lu_name *cname = &lfsck_env_info(env)->lti_name;
6673 struct linkea_data ldata = { NULL };
6676 rc = lfsck_links_read_with_rec(env, obj, &ldata);
6680 linkea_first_entry(&ldata);
6681 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, cname, pfid);
6682 /* To guarantee the 'name' is terminated with '0'. */
6683 memcpy(name, cname->ln_name, cname->ln_namelen);
6684 name[cname->ln_namelen] = 0;
6690 * Update the object's name entry with the given FID.
6692 * \param[in] env pointer to the thread context
6693 * \param[in] lfsck pointer to the lfsck instance
6694 * \param[in] dir pointer to the directory that holds
6696 * \param[in] name the name for the entry to be updated
6697 * \param[in] fid the new FID for the name entry referenced
6698 * \param[in] type the type for the name entry to be updated
6700 * \retval 0 for success
6701 * \retval negative error number on failure
6703 int lfsck_update_name_entry(const struct lu_env *env,
6704 struct lfsck_instance *lfsck,
6705 struct dt_object *dir, const char *name,
6706 const struct lu_fid *fid, __u32 type)
6708 struct lfsck_thread_info *info = lfsck_env_info(env);
6709 struct dt_insert_rec *rec = &info->lti_dt_rec;
6710 struct lfsck_lock_handle *llh = &info->lti_llh;
6711 struct dt_device *dev = lfsck_obj2dev(dir);
6717 rc = lfsck_lock(env, lfsck, dir, name, llh,
6718 MDS_INODELOCK_UPDATE, LCK_PW);
6722 th = dt_trans_create(env, dev);
6724 GOTO(unlock, rc = PTR_ERR(th));
6726 rc = dt_declare_delete(env, dir, (const struct dt_key *)name, th);
6730 rec->rec_type = type;
6732 rc = dt_declare_insert(env, dir, (const struct dt_rec *)rec,
6733 (const struct dt_key *)name, th);
6737 rc = dt_declare_ref_add(env, dir, th);
6741 rc = dt_trans_start_local(env, dev, th);
6745 rc = dt_delete(env, dir, (const struct dt_key *)name, th);
6746 if (rc == -ENOENT) {
6754 rc = dt_insert(env, dir, (const struct dt_rec *)rec,
6755 (const struct dt_key *)name, th, 1);
6756 if (rc == 0 && S_ISDIR(type) && !exists) {
6757 dt_write_lock(env, dir, 0);
6758 rc = dt_ref_add(env, dir, th);
6759 dt_write_unlock(env, dir);
6765 dt_trans_stop(env, dev, th);
6769 CDEBUG(D_LFSCK, "%s: update name entry "DFID"/%s with the FID "DFID
6770 " and the type %o: rc = %d\n", lfsck_lfsck2name(lfsck),
6771 PFID(lfsck_dto2fid(dir)), name, PFID(fid), type, rc);
6776 int lfsck_namespace_setup(const struct lu_env *env,
6777 struct lfsck_instance *lfsck)
6779 struct lfsck_component *com;
6780 struct lfsck_namespace *ns;
6781 struct dt_object *root = NULL;
6782 struct dt_object *obj;
6787 LASSERT(lfsck->li_master);
6793 INIT_LIST_HEAD(&com->lc_link);
6794 INIT_LIST_HEAD(&com->lc_link_dir);
6795 init_rwsem(&com->lc_sem);
6796 atomic_set(&com->lc_ref, 1);
6797 com->lc_lfsck = lfsck;
6798 com->lc_type = LFSCK_TYPE_NAMESPACE;
6799 com->lc_ops = &lfsck_namespace_ops;
6800 com->lc_data = lfsck_assistant_data_init(
6801 &lfsck_namespace_assistant_ops,
6803 if (com->lc_data == NULL)
6804 GOTO(out, rc = -ENOMEM);
6806 com->lc_file_size = sizeof(struct lfsck_namespace);
6807 OBD_ALLOC(com->lc_file_ram, com->lc_file_size);
6808 if (com->lc_file_ram == NULL)
6809 GOTO(out, rc = -ENOMEM);
6811 OBD_ALLOC(com->lc_file_disk, com->lc_file_size);
6812 if (com->lc_file_disk == NULL)
6813 GOTO(out, rc = -ENOMEM);
6815 for (i = 0; i < LFSCK_STF_COUNT; i++)
6816 mutex_init(&com->lc_sub_trace_objs[i].lsto_mutex);
6818 root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
6820 GOTO(out, rc = PTR_ERR(root));
6822 if (unlikely(!dt_try_as_dir(env, root)))
6823 GOTO(out, rc = -ENOTDIR);
6825 obj = local_index_find_or_create(env, lfsck->li_los, root,
6827 S_IFREG | S_IRUGO | S_IWUSR,
6828 &dt_lfsck_namespace_features);
6830 GOTO(out, rc = PTR_ERR(obj));
6833 rc = lfsck_namespace_load(env, com);
6834 if (rc == -ENODATA) {
6835 rc = lfsck_namespace_init(env, com);
6836 } else if (rc < 0) {
6837 rc = lfsck_namespace_reset(env, com, true);
6839 rc = lfsck_load_sub_trace_files(env, com,
6840 &dt_lfsck_namespace_features, LFSCK_NAMESPACE, false);
6842 rc = lfsck_namespace_reset(env, com, true);
6847 ns = com->lc_file_ram;
6848 switch (ns->ln_status) {
6853 spin_lock(&lfsck->li_lock);
6854 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
6855 spin_unlock(&lfsck->li_lock);
6858 CERROR("%s: unknown lfsck_namespace status %d\n",
6859 lfsck_lfsck2name(lfsck), ns->ln_status);
6861 case LS_SCANNING_PHASE1:
6862 case LS_SCANNING_PHASE2:
6863 /* No need to store the status to disk right now.
6864 * If the system crashed before the status stored,
6865 * it will be loaded back when next time. */
6866 ns->ln_status = LS_CRASHED;
6870 spin_lock(&lfsck->li_lock);
6871 list_add_tail(&com->lc_link, &lfsck->li_list_scan);
6872 list_add_tail(&com->lc_link_dir, &lfsck->li_list_dir);
6873 spin_unlock(&lfsck->li_lock);
6880 if (root != NULL && !IS_ERR(root))
6881 lfsck_object_put(env, root);
6883 lfsck_component_cleanup(env, com);
6884 CERROR("%s: fail to init namespace LFSCK component: rc = %d\n",
6885 lfsck_lfsck2name(lfsck), rc);