4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License version 2 for more details. A copy is
14 * included in the COPYING file that accompanied this code.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
23 * Copyright (c) 2013, 2017, Intel Corporation.
26 * lustre/lfsck/lfsck_namespace.c
28 * Author: Fan, Yong <fan.yong@intel.com>
31 #define DEBUG_SUBSYSTEM S_LFSCK
33 #include <lu_object.h>
34 #include <dt_object.h>
35 #include <md_object.h>
36 #include <lustre_fid.h>
37 #include <lustre_lib.h>
38 #include <lustre_net.h>
40 #include "lfsck_internal.h"
42 #define LFSCK_NAMESPACE_MAGIC_V1 0xA0629D03
43 #define LFSCK_NAMESPACE_MAGIC_V2 0xA0621A0B
44 #define LFSCK_NAMESPACE_MAGIC_V3 0xA06249FF
46 /* For Lustre-2.x (x <= 6), the namespace LFSCK used LFSCK_NAMESPACE_MAGIC_V1
47 * as the trace file magic. When downgrade to such old release, the old LFSCK
48 * will not recognize the new LFSCK_NAMESPACE_MAGIC_V2 in the new trace file,
49 * then it will reset the whole LFSCK, and will not cause start failure. The
50 * similar case will happen when upgrade from such old release. */
51 #define LFSCK_NAMESPACE_MAGIC LFSCK_NAMESPACE_MAGIC_V3
53 enum lfsck_nameentry_check {
54 LFSCK_NAMEENTRY_DEAD = 1, /* The object has been unlinked. */
55 LFSCK_NAMEENTRY_REMOVED = 2, /* The entry has been removed. */
56 LFSCK_NAMEENTRY_RECREATED = 3, /* The entry has been recreated. */
59 static struct lfsck_namespace_req *
60 lfsck_namespace_assistant_req_init(struct lfsck_instance *lfsck,
61 struct lfsck_assistant_object *lso,
62 struct lu_dirent *ent, __u16 type)
64 struct lfsck_namespace_req *lnr;
67 size = sizeof(*lnr) + (ent->lde_namelen & ~3) + 4;
70 return ERR_PTR(-ENOMEM);
72 INIT_LIST_HEAD(&lnr->lnr_lar.lar_list);
73 lnr->lnr_lar.lar_parent = lfsck_assistant_object_get(lso);
74 lnr->lnr_lmv = lfsck_lmv_get(lfsck->li_lmv);
75 lnr->lnr_fid = ent->lde_fid;
76 lnr->lnr_dir_cookie = ent->lde_hash;
77 lnr->lnr_attr = ent->lde_attrs;
80 lnr->lnr_namelen = ent->lde_namelen;
81 memcpy(lnr->lnr_name, ent->lde_name, ent->lde_namelen);
86 static void lfsck_namespace_assistant_req_fini(const struct lu_env *env,
87 struct lfsck_assistant_req *lar)
89 struct lfsck_namespace_req *lnr =
90 container_of0(lar, struct lfsck_namespace_req, lnr_lar);
92 if (lnr->lnr_lmv != NULL)
93 lfsck_lmv_put(env, lnr->lnr_lmv);
95 lfsck_assistant_object_put(env, lar->lar_parent);
96 OBD_FREE(lnr, lnr->lnr_size);
99 static void lfsck_namespace_le_to_cpu(struct lfsck_namespace *dst,
100 struct lfsck_namespace *src)
102 dst->ln_magic = le32_to_cpu(src->ln_magic);
103 dst->ln_status = le32_to_cpu(src->ln_status);
104 dst->ln_flags = le32_to_cpu(src->ln_flags);
105 dst->ln_success_count = le32_to_cpu(src->ln_success_count);
106 dst->ln_run_time_phase1 = le64_to_cpu(src->ln_run_time_phase1);
107 dst->ln_run_time_phase2 = le64_to_cpu(src->ln_run_time_phase2);
108 dst->ln_time_last_complete = le64_to_cpu(src->ln_time_last_complete);
109 dst->ln_time_latest_start = le64_to_cpu(src->ln_time_latest_start);
110 dst->ln_time_last_checkpoint =
111 le64_to_cpu(src->ln_time_last_checkpoint);
112 lfsck_position_le_to_cpu(&dst->ln_pos_latest_start,
113 &src->ln_pos_latest_start);
114 lfsck_position_le_to_cpu(&dst->ln_pos_last_checkpoint,
115 &src->ln_pos_last_checkpoint);
116 lfsck_position_le_to_cpu(&dst->ln_pos_first_inconsistent,
117 &src->ln_pos_first_inconsistent);
118 dst->ln_items_checked = le64_to_cpu(src->ln_items_checked);
119 dst->ln_items_repaired = le64_to_cpu(src->ln_items_repaired);
120 dst->ln_items_failed = le64_to_cpu(src->ln_items_failed);
121 dst->ln_dirs_checked = le64_to_cpu(src->ln_dirs_checked);
122 dst->ln_objs_checked_phase2 = le64_to_cpu(src->ln_objs_checked_phase2);
123 dst->ln_objs_repaired_phase2 =
124 le64_to_cpu(src->ln_objs_repaired_phase2);
125 dst->ln_objs_failed_phase2 = le64_to_cpu(src->ln_objs_failed_phase2);
126 dst->ln_objs_nlink_repaired = le64_to_cpu(src->ln_objs_nlink_repaired);
127 fid_le_to_cpu(&dst->ln_fid_latest_scanned_phase2,
128 &src->ln_fid_latest_scanned_phase2);
129 dst->ln_dirent_repaired = le64_to_cpu(src->ln_dirent_repaired);
130 dst->ln_linkea_repaired = le64_to_cpu(src->ln_linkea_repaired);
131 dst->ln_mul_linked_checked = le64_to_cpu(src->ln_mul_linked_checked);
132 dst->ln_mul_linked_repaired = le64_to_cpu(src->ln_mul_linked_repaired);
133 dst->ln_unknown_inconsistency =
134 le64_to_cpu(src->ln_unknown_inconsistency);
135 dst->ln_unmatched_pairs_repaired =
136 le64_to_cpu(src->ln_unmatched_pairs_repaired);
137 dst->ln_dangling_repaired = le64_to_cpu(src->ln_dangling_repaired);
138 dst->ln_mul_ref_repaired = le64_to_cpu(src->ln_mul_ref_repaired);
139 dst->ln_bad_type_repaired = le64_to_cpu(src->ln_bad_type_repaired);
140 dst->ln_lost_dirent_repaired =
141 le64_to_cpu(src->ln_lost_dirent_repaired);
142 dst->ln_striped_dirs_scanned =
143 le64_to_cpu(src->ln_striped_dirs_scanned);
144 dst->ln_striped_dirs_repaired =
145 le64_to_cpu(src->ln_striped_dirs_repaired);
146 dst->ln_striped_dirs_failed =
147 le64_to_cpu(src->ln_striped_dirs_failed);
148 dst->ln_striped_dirs_disabled =
149 le64_to_cpu(src->ln_striped_dirs_disabled);
150 dst->ln_striped_dirs_skipped =
151 le64_to_cpu(src->ln_striped_dirs_skipped);
152 dst->ln_striped_shards_scanned =
153 le64_to_cpu(src->ln_striped_shards_scanned);
154 dst->ln_striped_shards_repaired =
155 le64_to_cpu(src->ln_striped_shards_repaired);
156 dst->ln_striped_shards_failed =
157 le64_to_cpu(src->ln_striped_shards_failed);
158 dst->ln_striped_shards_skipped =
159 le64_to_cpu(src->ln_striped_shards_skipped);
160 dst->ln_name_hash_repaired = le64_to_cpu(src->ln_name_hash_repaired);
161 dst->ln_local_lpf_scanned = le64_to_cpu(src->ln_local_lpf_scanned);
162 dst->ln_local_lpf_moved = le64_to_cpu(src->ln_local_lpf_moved);
163 dst->ln_local_lpf_skipped = le64_to_cpu(src->ln_local_lpf_skipped);
164 dst->ln_local_lpf_failed = le64_to_cpu(src->ln_local_lpf_failed);
165 dst->ln_bitmap_size = le32_to_cpu(src->ln_bitmap_size);
166 dst->ln_time_latest_reset = le64_to_cpu(src->ln_time_latest_reset);
167 dst->ln_linkea_overflow_cleared =
168 le64_to_cpu(src->ln_linkea_overflow_cleared);
169 dst->ln_agent_entries_repaired =
170 le64_to_cpu(src->ln_agent_entries_repaired);
173 static void lfsck_namespace_cpu_to_le(struct lfsck_namespace *dst,
174 struct lfsck_namespace *src)
176 dst->ln_magic = cpu_to_le32(src->ln_magic);
177 dst->ln_status = cpu_to_le32(src->ln_status);
178 dst->ln_flags = cpu_to_le32(src->ln_flags);
179 dst->ln_success_count = cpu_to_le32(src->ln_success_count);
180 dst->ln_run_time_phase1 = cpu_to_le64(src->ln_run_time_phase1);
181 dst->ln_run_time_phase2 = cpu_to_le64(src->ln_run_time_phase2);
182 dst->ln_time_last_complete = cpu_to_le64(src->ln_time_last_complete);
183 dst->ln_time_latest_start = cpu_to_le64(src->ln_time_latest_start);
184 dst->ln_time_last_checkpoint =
185 cpu_to_le64(src->ln_time_last_checkpoint);
186 lfsck_position_cpu_to_le(&dst->ln_pos_latest_start,
187 &src->ln_pos_latest_start);
188 lfsck_position_cpu_to_le(&dst->ln_pos_last_checkpoint,
189 &src->ln_pos_last_checkpoint);
190 lfsck_position_cpu_to_le(&dst->ln_pos_first_inconsistent,
191 &src->ln_pos_first_inconsistent);
192 dst->ln_items_checked = cpu_to_le64(src->ln_items_checked);
193 dst->ln_items_repaired = cpu_to_le64(src->ln_items_repaired);
194 dst->ln_items_failed = cpu_to_le64(src->ln_items_failed);
195 dst->ln_dirs_checked = cpu_to_le64(src->ln_dirs_checked);
196 dst->ln_objs_checked_phase2 = cpu_to_le64(src->ln_objs_checked_phase2);
197 dst->ln_objs_repaired_phase2 =
198 cpu_to_le64(src->ln_objs_repaired_phase2);
199 dst->ln_objs_failed_phase2 = cpu_to_le64(src->ln_objs_failed_phase2);
200 dst->ln_objs_nlink_repaired = cpu_to_le64(src->ln_objs_nlink_repaired);
201 fid_cpu_to_le(&dst->ln_fid_latest_scanned_phase2,
202 &src->ln_fid_latest_scanned_phase2);
203 dst->ln_dirent_repaired = cpu_to_le64(src->ln_dirent_repaired);
204 dst->ln_linkea_repaired = cpu_to_le64(src->ln_linkea_repaired);
205 dst->ln_mul_linked_checked = cpu_to_le64(src->ln_mul_linked_checked);
206 dst->ln_mul_linked_repaired = cpu_to_le64(src->ln_mul_linked_repaired);
207 dst->ln_unknown_inconsistency =
208 cpu_to_le64(src->ln_unknown_inconsistency);
209 dst->ln_unmatched_pairs_repaired =
210 cpu_to_le64(src->ln_unmatched_pairs_repaired);
211 dst->ln_dangling_repaired = cpu_to_le64(src->ln_dangling_repaired);
212 dst->ln_mul_ref_repaired = cpu_to_le64(src->ln_mul_ref_repaired);
213 dst->ln_bad_type_repaired = cpu_to_le64(src->ln_bad_type_repaired);
214 dst->ln_lost_dirent_repaired =
215 cpu_to_le64(src->ln_lost_dirent_repaired);
216 dst->ln_striped_dirs_scanned =
217 cpu_to_le64(src->ln_striped_dirs_scanned);
218 dst->ln_striped_dirs_repaired =
219 cpu_to_le64(src->ln_striped_dirs_repaired);
220 dst->ln_striped_dirs_failed =
221 cpu_to_le64(src->ln_striped_dirs_failed);
222 dst->ln_striped_dirs_disabled =
223 cpu_to_le64(src->ln_striped_dirs_disabled);
224 dst->ln_striped_dirs_skipped =
225 cpu_to_le64(src->ln_striped_dirs_skipped);
226 dst->ln_striped_shards_scanned =
227 cpu_to_le64(src->ln_striped_shards_scanned);
228 dst->ln_striped_shards_repaired =
229 cpu_to_le64(src->ln_striped_shards_repaired);
230 dst->ln_striped_shards_failed =
231 cpu_to_le64(src->ln_striped_shards_failed);
232 dst->ln_striped_shards_skipped =
233 cpu_to_le64(src->ln_striped_shards_skipped);
234 dst->ln_name_hash_repaired = cpu_to_le64(src->ln_name_hash_repaired);
235 dst->ln_local_lpf_scanned = cpu_to_le64(src->ln_local_lpf_scanned);
236 dst->ln_local_lpf_moved = cpu_to_le64(src->ln_local_lpf_moved);
237 dst->ln_local_lpf_skipped = cpu_to_le64(src->ln_local_lpf_skipped);
238 dst->ln_local_lpf_failed = cpu_to_le64(src->ln_local_lpf_failed);
239 dst->ln_bitmap_size = cpu_to_le32(src->ln_bitmap_size);
240 dst->ln_time_latest_reset = cpu_to_le64(src->ln_time_latest_reset);
241 dst->ln_linkea_overflow_cleared =
242 cpu_to_le64(src->ln_linkea_overflow_cleared);
243 dst->ln_agent_entries_repaired =
244 cpu_to_le64(src->ln_agent_entries_repaired);
247 static void lfsck_namespace_record_failure(const struct lu_env *env,
248 struct lfsck_instance *lfsck,
249 struct lfsck_namespace *ns)
251 struct lfsck_position pos;
253 ns->ln_items_failed++;
254 lfsck_pos_fill(env, lfsck, &pos, false);
255 if (lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent) ||
256 lfsck_pos_is_eq(&pos, &ns->ln_pos_first_inconsistent) < 0) {
257 ns->ln_pos_first_inconsistent = pos;
259 CDEBUG(D_LFSCK, "%s: namespace LFSCK hit first non-repaired "
260 "inconsistency at the pos [%llu, "DFID", %#llx]\n",
261 lfsck_lfsck2name(lfsck),
262 ns->ln_pos_first_inconsistent.lp_oit_cookie,
263 PFID(&ns->ln_pos_first_inconsistent.lp_dir_parent),
264 ns->ln_pos_first_inconsistent.lp_dir_cookie);
269 * Load the MDT bitmap from the lfsck_namespace trace file.
271 * \param[in] env pointer to the thread context
272 * \param[in] com pointer to the lfsck component
274 * \retval 0 for success
275 * \retval negative error number on failure or data corruption
277 static int lfsck_namespace_load_bitmap(const struct lu_env *env,
278 struct lfsck_component *com)
280 struct dt_object *obj = com->lc_obj;
281 struct lfsck_assistant_data *lad = com->lc_data;
282 struct lfsck_namespace *ns = com->lc_file_ram;
283 struct cfs_bitmap *bitmap = lad->lad_bitmap;
289 if (com->lc_lfsck->li_mdt_descs.ltd_tgts_bitmap->size >
291 nbits = com->lc_lfsck->li_mdt_descs.ltd_tgts_bitmap->size;
293 nbits = ns->ln_bitmap_size;
295 if (unlikely(nbits < BITS_PER_LONG))
296 nbits = BITS_PER_LONG;
298 if (nbits > bitmap->size) {
299 __u32 new_bits = bitmap->size;
300 struct cfs_bitmap *new_bitmap;
302 while (new_bits < nbits)
305 new_bitmap = CFS_ALLOCATE_BITMAP(new_bits);
306 if (new_bitmap == NULL)
309 lad->lad_bitmap = new_bitmap;
310 CFS_FREE_BITMAP(bitmap);
314 if (ns->ln_bitmap_size == 0) {
315 clear_bit(LAD_INCOMPLETE, &lad->lad_flags);
316 CFS_RESET_BITMAP(bitmap);
321 size = (ns->ln_bitmap_size + 7) >> 3;
322 rc = dt_xattr_get(env, obj,
323 lfsck_buf_get(env, bitmap->data, size),
324 XATTR_NAME_LFSCK_BITMAP);
326 RETURN(rc >= 0 ? -EINVAL : rc);
328 if (cfs_bitmap_check_empty(bitmap))
329 clear_bit(LAD_INCOMPLETE, &lad->lad_flags);
331 set_bit(LAD_INCOMPLETE, &lad->lad_flags);
337 * Load namespace LFSCK statistics information from the trace file.
339 * \param[in] env pointer to the thread context
340 * \param[in] com pointer to the lfsck component
342 * \retval 0 for success
343 * \retval negative error number on failure
345 static int lfsck_namespace_load(const struct lu_env *env,
346 struct lfsck_component *com)
348 int len = com->lc_file_size;
351 rc = dt_xattr_get(env, com->lc_obj,
352 lfsck_buf_get(env, com->lc_file_disk, len),
353 XATTR_NAME_LFSCK_NAMESPACE);
355 struct lfsck_namespace *ns = com->lc_file_ram;
357 lfsck_namespace_le_to_cpu(ns,
358 (struct lfsck_namespace *)com->lc_file_disk);
359 if (ns->ln_magic != LFSCK_NAMESPACE_MAGIC) {
360 CDEBUG(D_LFSCK, "%s: invalid lfsck_namespace magic "
361 "%#x != %#x\n", lfsck_lfsck2name(com->lc_lfsck),
362 ns->ln_magic, LFSCK_NAMESPACE_MAGIC);
367 } else if (rc != -ENODATA) {
368 CDEBUG(D_LFSCK, "%s: fail to load lfsck_namespace, "
369 "expected = %d: rc = %d\n",
370 lfsck_lfsck2name(com->lc_lfsck), len, rc);
378 static int lfsck_namespace_store(const struct lu_env *env,
379 struct lfsck_component *com)
381 struct dt_object *obj = com->lc_obj;
382 struct lfsck_instance *lfsck = com->lc_lfsck;
383 struct lfsck_namespace *ns = com->lc_file_ram;
384 struct lfsck_assistant_data *lad = com->lc_data;
385 struct dt_device *dev = lfsck_obj2dev(obj);
386 struct cfs_bitmap *bitmap = NULL;
387 struct thandle *handle;
389 int len = com->lc_file_size;
394 bitmap = lad->lad_bitmap;
395 nbits = bitmap->size;
398 LASSERTF((nbits & 7) == 0, "Invalid nbits %u\n", nbits);
401 ns->ln_bitmap_size = nbits;
402 lfsck_namespace_cpu_to_le((struct lfsck_namespace *)com->lc_file_disk,
404 handle = dt_trans_create(env, dev);
406 GOTO(log, rc = PTR_ERR(handle));
408 rc = dt_declare_xattr_set(env, obj,
409 lfsck_buf_get(env, com->lc_file_disk, len),
410 XATTR_NAME_LFSCK_NAMESPACE, 0, handle);
414 if (bitmap != NULL) {
415 rc = dt_declare_xattr_set(env, obj,
416 lfsck_buf_get(env, bitmap->data, nbits >> 3),
417 XATTR_NAME_LFSCK_BITMAP, 0, handle);
422 rc = dt_trans_start_local(env, dev, handle);
426 rc = dt_xattr_set(env, obj,
427 lfsck_buf_get(env, com->lc_file_disk, len),
428 XATTR_NAME_LFSCK_NAMESPACE, 0, handle);
429 if (rc == 0 && bitmap != NULL)
430 rc = dt_xattr_set(env, obj,
431 lfsck_buf_get(env, bitmap->data, nbits >> 3),
432 XATTR_NAME_LFSCK_BITMAP, 0, handle);
437 dt_trans_stop(env, dev, handle);
441 CDEBUG(D_LFSCK, "%s: fail to store lfsck_namespace: rc = %d\n",
442 lfsck_lfsck2name(lfsck), rc);
446 static int lfsck_namespace_init(const struct lu_env *env,
447 struct lfsck_component *com)
449 struct lfsck_namespace *ns = com->lc_file_ram;
452 memset(ns, 0, sizeof(*ns));
453 ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
454 ns->ln_status = LS_INIT;
455 ns->ln_time_latest_reset = ktime_get_real_seconds();
456 down_write(&com->lc_sem);
457 rc = lfsck_namespace_store(env, com);
459 rc = lfsck_load_sub_trace_files(env, com,
460 &dt_lfsck_namespace_features, LFSCK_NAMESPACE, true);
461 up_write(&com->lc_sem);
467 * Update the namespace LFSCK trace file for the given @fid
469 * \param[in] env pointer to the thread context
470 * \param[in] com pointer to the lfsck component
471 * \param[in] fid the fid which flags to be updated in the lfsck
473 * \param[in] add true if add new flags, otherwise remove flags
475 * \retval 0 for success or nothing to be done
476 * \retval negative error number on failure
478 int lfsck_namespace_trace_update(const struct lu_env *env,
479 struct lfsck_component *com,
480 const struct lu_fid *fid,
481 const __u8 flags, bool add)
483 struct lfsck_instance *lfsck = com->lc_lfsck;
484 struct dt_object *obj;
485 struct lu_fid *key = &lfsck_env_info(env)->lti_fid3;
486 struct dt_device *dev;
487 struct thandle *th = NULL;
496 if (unlikely(!fid_is_sane(fid)))
499 idx = lfsck_sub_trace_file_fid2idx(fid);
500 mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
501 obj = com->lc_sub_trace_objs[idx].lsto_obj;
502 if (unlikely(obj == NULL)) {
503 mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
507 lfsck_object_get(obj);
508 dev = lfsck_obj2dev(obj);
509 fid_cpu_to_be(key, fid);
510 rc = dt_lookup(env, obj, (struct dt_rec *)&old,
511 (const struct dt_key *)key);
514 GOTO(unlock, rc = 0);
518 } else if (rc == 0) {
520 if ((old & flags) == flags)
521 GOTO(unlock, rc = 0);
525 if ((old & flags) == 0)
526 GOTO(unlock, rc = 0);
534 th = dt_trans_create(env, dev);
536 GOTO(log, rc = PTR_ERR(th));
539 rc = dt_declare_delete(env, obj,
540 (const struct dt_key *)key, th);
546 rc = dt_declare_insert(env, obj,
547 (const struct dt_rec *)&new,
548 (const struct dt_key *)key, th);
553 rc = dt_trans_start_local(env, dev, th);
558 rc = dt_delete(env, obj, (const struct dt_key *)key, th);
564 rc = dt_insert(env, obj, (const struct dt_rec *)&new,
565 (const struct dt_key *)key, th);
573 if (th != NULL && !IS_ERR(th))
574 dt_trans_stop(env, dev, th);
576 CDEBUG(D_LFSCK, "%s: namespace LFSCK %s flags for "DFID" in the "
577 "trace file, flags %x, old %x, new %x: rc = %d\n",
578 lfsck_lfsck2name(lfsck), add ? "add" : "del", PFID(fid),
579 (__u32)flags, (__u32)old, (__u32)new, rc);
582 mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
583 lfsck_object_put(env, obj);
588 int lfsck_namespace_check_exist(const struct lu_env *env,
589 struct dt_object *dir,
590 struct dt_object *obj, const char *name)
592 struct lu_fid *fid = &lfsck_env_info(env)->lti_fid;
596 if (unlikely(lfsck_is_dead_obj(obj)))
597 RETURN(LFSCK_NAMEENTRY_DEAD);
599 rc = dt_lookup(env, dir, (struct dt_rec *)fid,
600 (const struct dt_key *)name);
602 RETURN(LFSCK_NAMEENTRY_REMOVED);
607 if (!lu_fid_eq(fid, lfsck_dto2fid(obj)))
608 RETURN(LFSCK_NAMEENTRY_RECREATED);
613 static int lfsck_declare_namespace_exec_dir(const struct lu_env *env,
614 struct dt_object *obj,
615 struct thandle *handle)
619 /* For remote updating LINKEA, there may be further LFSCK action
620 * on remote MDT after the updating, so update the LINKEA ASAP. */
621 if (dt_object_remote(obj))
624 /* For destroying all invalid linkEA entries. */
625 rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, handle);
627 /* For insert new linkEA entry. */
628 rc = dt_declare_xattr_set(env, obj,
629 lfsck_buf_get_const(env, NULL, MAX_LINKEA_SIZE),
630 XATTR_NAME_LINK, 0, handle);
634 int __lfsck_links_read(const struct lu_env *env, struct dt_object *obj,
635 struct linkea_data *ldata, bool with_rec)
639 if (ldata->ld_buf->lb_buf == NULL)
642 if (!dt_object_exists(obj))
645 rc = dt_xattr_get(env, obj, ldata->ld_buf, XATTR_NAME_LINK);
647 /* Buf was too small, figure out what we need. */
648 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LINK);
649 if (unlikely(rc == 0))
655 lu_buf_realloc(ldata->ld_buf, rc);
656 if (ldata->ld_buf->lb_buf == NULL)
659 rc = dt_xattr_get(env, obj, ldata->ld_buf, XATTR_NAME_LINK);
662 if (unlikely(rc == 0))
667 rc = linkea_init_with_rec(ldata);
669 rc = linkea_init(ldata);
676 * Remove linkEA for the given object.
678 * The caller should take the ldlm lock before the calling.
680 * \param[in] env pointer to the thread context
681 * \param[in] com pointer to the lfsck component
682 * \param[in] obj pointer to the dt_object to be handled
684 * \retval 0 for repaired cases
685 * \retval negative error number on failure
687 static int lfsck_namespace_links_remove(const struct lu_env *env,
688 struct lfsck_component *com,
689 struct dt_object *obj)
691 struct lfsck_instance *lfsck = com->lc_lfsck;
692 struct dt_device *dev = lfsck_obj2dev(obj);
693 struct thandle *th = NULL;
697 LASSERT(dt_object_remote(obj) == 0);
699 th = dt_trans_create(env, dev);
701 GOTO(log, rc = PTR_ERR(th));
703 rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, th);
707 rc = dt_trans_start_local(env, dev, th);
711 dt_write_lock(env, obj, 0);
712 if (unlikely(lfsck_is_dead_obj(obj)))
713 GOTO(unlock, rc = -ENOENT);
715 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
716 GOTO(unlock, rc = 0);
718 rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, th);
723 dt_write_unlock(env, obj);
726 dt_trans_stop(env, dev, th);
729 CDEBUG(D_LFSCK, "%s: namespace LFSCK remove invalid linkEA "
730 "for the object "DFID": rc = %d\n",
731 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
734 struct lfsck_namespace *ns = com->lc_file_ram;
736 ns->ln_flags |= LF_INCONSISTENT;
742 static int lfsck_links_write(const struct lu_env *env, struct dt_object *obj,
743 struct linkea_data *ldata, struct thandle *handle)
748 lfsck_buf_init(&buf, ldata->ld_buf->lb_buf, ldata->ld_leh->leh_len);
751 rc = dt_xattr_set(env, obj, &buf, XATTR_NAME_LINK, 0, handle);
752 if (unlikely(rc == -ENOSPC)) {
753 rc = linkea_overflow_shrink(ldata);
754 if (likely(rc > 0)) {
763 static inline bool linkea_reclen_is_valid(const struct linkea_data *ldata)
765 if (ldata->ld_reclen <= 0)
768 if ((char *)ldata->ld_lee + ldata->ld_reclen >
769 (char *)ldata->ld_leh + ldata->ld_leh->leh_len)
775 static inline bool linkea_entry_is_valid(const struct linkea_data *ldata,
776 const struct lu_name *cname,
777 const struct lu_fid *pfid)
779 if (!linkea_reclen_is_valid(ldata))
782 if (cname->ln_namelen <= 0 || cname->ln_namelen > NAME_MAX)
785 if (!fid_is_sane(pfid))
791 static int lfsck_namespace_unpack_linkea_entry(struct linkea_data *ldata,
792 struct lu_name *cname,
794 char *buf, const int buflen)
796 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, cname, pfid);
797 if (unlikely(!linkea_entry_is_valid(ldata, cname, pfid)))
800 /* To guarantee the 'name' is terminated with '0'. */
801 memcpy(buf, cname->ln_name, cname->ln_namelen);
802 buf[cname->ln_namelen] = 0;
803 cname->ln_name = buf;
808 static void lfsck_linkea_del_buf(struct linkea_data *ldata,
809 const struct lu_name *lname)
811 LASSERT(ldata->ld_leh != NULL && ldata->ld_lee != NULL);
813 /* If current record is corrupted, all the subsequent
814 * records will be dropped. */
815 if (unlikely(!linkea_reclen_is_valid(ldata))) {
816 void *ptr = ldata->ld_lee;
818 ldata->ld_leh->leh_len = sizeof(struct link_ea_header);
819 ldata->ld_leh->leh_reccount = 0;
820 linkea_first_entry(ldata);
821 while (ldata->ld_lee != NULL &&
822 (char *)ldata->ld_lee < (char *)ptr) {
823 int reclen = (ldata->ld_lee->lee_reclen[0] << 8) |
824 ldata->ld_lee->lee_reclen[1];
826 ldata->ld_leh->leh_len += reclen;
827 ldata->ld_leh->leh_reccount++;
828 ldata->ld_lee = (struct link_ea_entry *)
829 ((char *)ldata->ld_lee + reclen);
832 ldata->ld_lee = NULL;
834 linkea_del_buf(ldata, lname);
838 static int lfsck_namespace_filter_linkea_entry(struct linkea_data *ldata,
839 struct lu_name *cname,
843 struct link_ea_entry *oldlee;
847 oldlee = ldata->ld_lee;
848 oldlen = ldata->ld_reclen;
849 linkea_next_entry(ldata);
850 while (ldata->ld_lee != NULL) {
851 ldata->ld_reclen = (ldata->ld_lee->lee_reclen[0] << 8) |
852 ldata->ld_lee->lee_reclen[1];
853 if (unlikely(!linkea_reclen_is_valid(ldata))) {
854 lfsck_linkea_del_buf(ldata, NULL);
855 LASSERT(ldata->ld_lee == NULL);
856 } else if (unlikely(ldata->ld_reclen == oldlen &&
857 memcmp(ldata->ld_lee, oldlee, oldlen) == 0)) {
862 lfsck_linkea_del_buf(ldata, cname);
864 linkea_next_entry(ldata);
867 ldata->ld_lee = oldlee;
868 ldata->ld_reclen = oldlen;
874 * Insert orphan into .lustre/lost+found/MDTxxxx/ locally.
876 * Add the specified orphan MDT-object to the .lustre/lost+found/MDTxxxx/
877 * with the given type to generate the name, the detailed rules for name
878 * have been described as following.
880 * The function also generates the linkEA corresponding to the name entry
881 * under the .lustre/lost+found/MDTxxxx/ for the orphan MDT-object.
883 * \param[in] env pointer to the thread context
884 * \param[in] com pointer to the lfsck component
885 * \param[in] orphan pointer to the orphan MDT-object
886 * \param[in] infix additional information for the orphan name, such as
887 * the FID for original
888 * \param[in] type the type for describing why the orphan MDT-object is
889 * created. The rules are as following:
891 * type "D": The MDT-object is a directory, it may knows its parent
892 * but because there is no valid linkEA, the LFSCK cannot
893 * know where to put it back to the namespace.
894 * type "O": The MDT-object has no linkEA, and there is no name
895 * entry that references the MDT-object.
897 * type "S": The orphan MDT-object is a shard of a striped directory
899 * \see lfsck_layout_recreate_parent() for more types.
901 * The orphan name will be like:
902 * ${FID}-${infix}-${type}-${conflict_version}
904 * \param[out] count if some others inserted some linkEA entries by race,
905 * then return the linkEA entries count.
907 * \retval positive number for repaired cases
908 * \retval 0 if needs to repair nothing
909 * \retval negative error number on failure
911 static int lfsck_namespace_insert_orphan(const struct lu_env *env,
912 struct lfsck_component *com,
913 struct dt_object *orphan,
914 const char *infix, const char *type,
917 struct lfsck_thread_info *info = lfsck_env_info(env);
918 struct lu_name *cname = &info->lti_name;
919 struct dt_insert_rec *rec = &info->lti_dt_rec;
920 struct lu_attr *la = &info->lti_la2;
921 const struct lu_fid *cfid = lfsck_dto2fid(orphan);
922 const struct lu_fid *pfid;
924 struct lfsck_instance *lfsck = com->lc_lfsck;
925 struct dt_device *dev = lfsck_obj2dev(orphan);
926 struct dt_object *parent;
927 struct thandle *th = NULL;
928 struct lfsck_lock_handle *pllh = &info->lti_llh;
929 struct lustre_handle clh = { 0 };
930 struct linkea_data ldata2 = { NULL };
931 struct lu_buf linkea_buf;
938 cname->ln_name = NULL;
939 if (unlikely(lfsck->li_lpf_obj == NULL))
940 GOTO(log, rc = -ENXIO);
942 parent = lfsck->li_lpf_obj;
943 pfid = lfsck_dto2fid(parent);
947 namelen = snprintf(info->lti_key, NAME_MAX, DFID"%s-%s-%d",
948 PFID(cfid), infix, type, idx++);
949 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
950 (const struct dt_key *)info->lti_key);
951 if (rc != 0 && rc != -ENOENT)
954 if (unlikely(rc == 0 && lu_fid_eq(cfid, &tfid)))
956 } while (rc == 0 && !exist);
958 rc = lfsck_lock(env, lfsck, parent, info->lti_key, pllh,
959 MDS_INODELOCK_UPDATE, LCK_PW);
963 /* Re-check whether the name conflict with othrs after taken
965 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
966 (const struct dt_key *)info->lti_key);
968 if (!lu_fid_eq(cfid, &tfid)) {
975 } else if (rc != -ENOENT) {
981 cname->ln_name = info->lti_key;
982 cname->ln_namelen = namelen;
983 rc = linkea_links_new(&ldata2, &info->lti_linkea_buf2,
988 rc = lfsck_ibits_lock(env, lfsck, orphan, &clh,
989 MDS_INODELOCK_UPDATE | MDS_INODELOCK_LOOKUP |
990 MDS_INODELOCK_XATTR, LCK_EX);
994 lfsck_buf_init(&linkea_buf, ldata2.ld_buf->lb_buf,
995 ldata2.ld_leh->leh_len);
996 th = dt_trans_create(env, dev);
998 GOTO(log, rc = PTR_ERR(th));
1000 if (S_ISDIR(lfsck_object_type(orphan))) {
1001 rc = dt_declare_delete(env, orphan,
1002 (const struct dt_key *)dotdot, th);
1006 rec->rec_type = S_IFDIR;
1007 rec->rec_fid = pfid;
1008 rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
1009 (const struct dt_key *)dotdot, th);
1014 rc = dt_declare_xattr_set(env, orphan, &linkea_buf,
1015 XATTR_NAME_LINK, 0, th);
1020 rec->rec_type = lfsck_object_type(orphan) & S_IFMT;
1021 rec->rec_fid = cfid;
1022 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1023 (const struct dt_key *)cname->ln_name,
1028 if (S_ISDIR(rec->rec_type)) {
1029 rc = dt_declare_ref_add(env, parent, th);
1035 memset(la, 0, sizeof(*la));
1036 la->la_ctime = ktime_get_real_seconds();
1037 la->la_valid = LA_CTIME;
1038 rc = dt_declare_attr_set(env, orphan, la, th);
1042 rc = dt_trans_start_local(env, dev, th);
1046 dt_write_lock(env, orphan, 0);
1047 rc = lfsck_links_read2_with_rec(env, orphan, &ldata2);
1048 if (likely(rc == -ENODATA || rc == -EINVAL)) {
1049 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1050 GOTO(unlock, rc = 1);
1052 if (S_ISDIR(lfsck_object_type(orphan))) {
1053 rc = dt_delete(env, orphan,
1054 (const struct dt_key *)dotdot, th);
1058 rec->rec_type = S_IFDIR;
1059 rec->rec_fid = pfid;
1060 rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1061 (const struct dt_key *)dotdot, th);
1066 rc = dt_xattr_set(env, orphan, &linkea_buf, XATTR_NAME_LINK, 0,
1069 if (rc == 0 && count != NULL)
1070 *count = ldata2.ld_leh->leh_reccount;
1074 dt_write_unlock(env, orphan);
1076 if (rc == 0 && !exist) {
1077 rec->rec_type = lfsck_object_type(orphan) & S_IFMT;
1078 rec->rec_fid = cfid;
1079 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1080 (const struct dt_key *)cname->ln_name, th);
1081 if (rc == 0 && S_ISDIR(rec->rec_type)) {
1082 dt_write_lock(env, parent, 0);
1083 rc = dt_ref_add(env, parent, th);
1084 dt_write_unlock(env, parent);
1089 rc = dt_attr_set(env, orphan, la, th);
1091 GOTO(stop, rc = (rc == 0 ? 1 : rc));
1094 dt_write_unlock(env, orphan);
1097 dt_trans_stop(env, dev, th);
1100 lfsck_ibits_unlock(&clh, LCK_EX);
1102 CDEBUG(D_LFSCK, "%s: namespace LFSCK insert orphan for the "
1103 "object "DFID", name = %s: rc = %d\n",
1104 lfsck_lfsck2name(lfsck), PFID(cfid),
1105 cname->ln_name != NULL ? cname->ln_name : "<NULL>", rc);
1108 struct lfsck_namespace *ns = com->lc_file_ram;
1110 ns->ln_flags |= LF_INCONSISTENT;
1116 static int lfsck_lmv_set(const struct lu_env *env,
1117 struct lfsck_instance *lfsck,
1118 struct dt_object *obj,
1119 struct lmv_mds_md_v1 *lmv)
1121 struct dt_device *dev = lfsck->li_next;
1122 struct thandle *th = NULL;
1123 struct lu_buf buf = { lmv, sizeof(*lmv) };
1128 th = dt_trans_create(env, dev);
1130 RETURN(PTR_ERR(th));
1132 rc = dt_declare_xattr_set(env, obj, &buf, XATTR_NAME_LMV".set", 0, th);
1136 rc = dt_trans_start_local(env, dev, th);
1140 rc = dt_xattr_set(env, obj, &buf, XATTR_NAME_LMV".set", 0, th);
1146 dt_trans_stop(env, dev, th);
1151 static int lfsck_lmv_delete(const struct lu_env *env,
1152 struct lfsck_instance *lfsck,
1153 struct dt_object *obj)
1155 struct dt_device *dev = lfsck->li_next;
1156 struct thandle *th = NULL;
1161 th = dt_trans_create(env, dev);
1163 RETURN(PTR_ERR(th));
1165 rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LMV, th);
1169 rc = dt_trans_start_local(env, dev, th);
1173 rc = dt_xattr_del(env, obj, XATTR_NAME_LMV, th);
1179 dt_trans_stop(env, dev, th);
1184 static inline int lfsck_object_is_shard(const struct lu_env *env,
1185 struct lfsck_instance *lfsck,
1186 struct dt_object *obj,
1187 const struct lu_name *lname)
1189 struct lfsck_thread_info *info = lfsck_env_info(env);
1190 struct lmv_mds_md_v1 *lmv = &info->lti_lmv;
1193 rc = lfsck_shard_name_to_index(env, lname->ln_name, lname->ln_namelen,
1194 lfsck_object_type(obj),
1195 lfsck_dto2fid(obj));
1199 rc = lfsck_read_stripe_lmv(env, lfsck, obj, lmv);
1203 if (!rc && lmv->lmv_magic == LMV_MAGIC_STRIPE)
1210 * Add the specified name entry back to namespace.
1212 * If there is a linkEA entry that back references a name entry under
1213 * some parent directory, but such parent directory does not have the
1214 * claimed name entry. On the other hand, the linkEA entries count is
1215 * not larger than the MDT-object's hard link count. Under such case,
1216 * it is quite possible that the name entry is lost. Then the LFSCK
1217 * should add the name entry back to the namespace.
1219 * If \a child is shard, which means \a parent is a striped directory,
1220 * if \a parent has LMV, we need to delete it before insertion because
1221 * now parent's striping is broken and can't be parsed correctly.
1223 * \param[in] env pointer to the thread context
1224 * \param[in] com pointer to the lfsck component
1225 * \param[in] parent pointer to the directory under which the name entry
1226 * will be inserted into
1227 * \param[in] child pointer to the object referenced by the name entry
1228 * that to be inserted into the parent
1229 * \param[in] lname the name for the child in the parent directory
1231 * \retval positive number for repaired cases
1232 * \retval 0 if nothing to be repaired
1233 * \retval negative error number on failure
1235 static int lfsck_namespace_insert_normal(const struct lu_env *env,
1236 struct lfsck_component *com,
1237 struct dt_object *parent,
1238 struct dt_object *child,
1239 const struct lu_name *lname)
1241 struct lfsck_thread_info *info = lfsck_env_info(env);
1242 struct lu_attr *la = &info->lti_la;
1243 struct dt_insert_rec *rec = &info->lti_dt_rec;
1244 struct lfsck_instance *lfsck = com->lc_lfsck;
1245 /* The child and its name may be on different MDTs. */
1246 const struct lu_fid *pfid = lfsck_dto2fid(parent);
1247 const struct lu_fid *cfid = lfsck_dto2fid(child);
1248 struct dt_device *dev = lfsck->li_next;
1249 struct thandle *th = NULL;
1250 struct lfsck_lock_handle *llh = &info->lti_llh;
1251 struct lmv_mds_md_v1 *lmv = &info->lti_lmv;
1252 struct lu_buf buf = { lmv, sizeof(*lmv) };
1253 /* whether parent's LMV is deleted before insertion */
1254 bool parent_lmv_deleted = false;
1255 /* whether parent's LMV is missing */
1256 bool parent_lmv_lost = false;
1261 /* @parent/@child may be based on lfsck->li_bottom,
1262 * but here we need the object based on the lfsck->li_next. */
1264 parent = lfsck_object_locate(dev, parent);
1266 GOTO(log, rc = PTR_ERR(parent));
1268 child = lfsck_object_locate(dev, child);
1270 GOTO(log, rc = PTR_ERR(child));
1272 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1275 rc = lfsck_lock(env, lfsck, parent, lname->ln_name, llh,
1276 MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
1277 MDS_INODELOCK_XATTR, LCK_EX);
1281 rc = lfsck_object_is_shard(env, lfsck, child, lname);
1286 rc = lfsck_read_stripe_lmv(env, lfsck, parent, lmv);
1289 * To add a shard, we need to convert parent to a
1290 * plain directory by deleting its LMV, and after
1291 * insertion set it back.
1293 rc = lfsck_lmv_delete(env, lfsck, parent);
1296 parent_lmv_deleted = true;
1297 lmv->lmv_layout_version++;
1298 lfsck_lmv_header_cpu_to_le(lmv, lmv);
1299 } else if (rc == -ENODATA) {
1300 struct lu_seq_range *range = &info->lti_range;
1301 struct seq_server_site *ss = lfsck_dev_site(lfsck);
1303 rc = lfsck_read_stripe_lmv(env, lfsck, child, lmv);
1307 fld_range_set_mdt(range);
1308 rc = fld_server_lookup(env, ss->ss_server_fld,
1309 fid_seq(lfsck_dto2fid(parent)), range);
1313 parent_lmv_lost = true;
1314 lmv->lmv_magic = LMV_MAGIC;
1315 lmv->lmv_master_mdt_index = range->lsr_index;
1316 lmv->lmv_layout_version++;
1317 lfsck_lmv_header_cpu_to_le(lmv, lmv);
1323 if (unlikely(!dt_try_as_dir(env, parent)))
1324 GOTO(unlock, rc = -ENOTDIR);
1326 th = dt_trans_create(env, dev);
1328 GOTO(unlock, rc = PTR_ERR(th));
1330 rec->rec_type = lfsck_object_type(child) & S_IFMT;
1331 rec->rec_fid = cfid;
1332 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1333 (const struct dt_key *)lname->ln_name, th);
1337 if (S_ISDIR(rec->rec_type)) {
1338 rc = dt_declare_ref_add(env, parent, th);
1343 if (parent_lmv_lost) {
1344 rc = dt_declare_xattr_set(env, parent, &buf,
1345 XATTR_NAME_LMV".set", 0, th);
1350 la->la_ctime = ktime_get_real_seconds();
1351 la->la_valid = LA_CTIME;
1352 rc = dt_declare_attr_set(env, parent, la, th);
1356 rc = dt_declare_attr_set(env, child, la, th);
1360 rc = dt_trans_start_local(env, dev, th);
1364 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1365 (const struct dt_key *)lname->ln_name, th);
1369 if (S_ISDIR(rec->rec_type)) {
1370 dt_write_lock(env, parent, 0);
1371 rc = dt_ref_add(env, parent, th);
1372 dt_write_unlock(env, parent);
1377 if (parent_lmv_lost) {
1378 rc = dt_xattr_set(env, parent, &buf, XATTR_NAME_LMV".set", 0,
1384 rc = dt_attr_set(env, parent, la, th);
1388 rc = dt_attr_set(env, child, la, th);
1390 GOTO(stop, rc = (rc == 0 ? 1 : rc));
1393 dt_trans_stop(env, dev, th);
1396 if (parent_lmv_deleted)
1397 lfsck_lmv_set(env, lfsck, parent, lmv);
1402 CDEBUG(D_LFSCK, "%s: namespace LFSCK insert object "DFID" with "
1403 "the name %s and type %o to the parent "DFID": rc = %d\n",
1404 lfsck_lfsck2name(lfsck), PFID(cfid), lname->ln_name,
1405 lfsck_object_type(child) & S_IFMT, PFID(pfid), rc);
1408 struct lfsck_namespace *ns = com->lc_file_ram;
1410 ns->ln_flags |= LF_INCONSISTENT;
1412 ns->ln_lost_dirent_repaired++;
1419 * Create the specified orphan directory.
1421 * For the case that the parent MDT-object stored in some MDT-object's
1422 * linkEA entry is lost, the LFSCK will re-create the parent object as
1423 * an orphan and insert it into .lustre/lost+found/MDTxxxx/ directory
1424 * with the name ${FID}-P-${conflict_version}.
1426 * \param[in] env pointer to the thread context
1427 * \param[in] com pointer to the lfsck component
1428 * \param[in] orphan pointer to the orphan MDT-object to be created
1429 * \param[in] lmv pointer to master LMV EA that will be set to the orphan
1431 * \retval positive number for repaired cases
1432 * \retval negative error number on failure
1434 static int lfsck_namespace_create_orphan_dir(const struct lu_env *env,
1435 struct lfsck_component *com,
1436 struct dt_object *orphan,
1437 struct lmv_mds_md_v1 *lmv)
1439 struct lfsck_thread_info *info = lfsck_env_info(env);
1440 struct lu_attr *la = &info->lti_la;
1441 struct dt_allocation_hint *hint = &info->lti_hint;
1442 struct dt_object_format *dof = &info->lti_dof;
1443 struct lu_name *cname = &info->lti_name2;
1444 struct dt_insert_rec *rec = &info->lti_dt_rec;
1445 struct lmv_mds_md_v1 *lmv2 = &info->lti_lmv2;
1446 const struct lu_fid *cfid = lfsck_dto2fid(orphan);
1448 struct lfsck_instance *lfsck = com->lc_lfsck;
1449 struct lfsck_namespace *ns = com->lc_file_ram;
1450 struct dt_device *dev = lfsck_obj2dev(orphan);
1451 struct dt_object *parent = NULL;
1452 struct thandle *th = NULL;
1453 struct lfsck_lock_handle *llh = &info->lti_llh;
1454 struct linkea_data ldata = { NULL };
1455 struct lu_buf linkea_buf;
1456 struct lu_buf lmv_buf;
1464 LASSERT(!dt_object_exists(orphan));
1466 cname->ln_name = NULL;
1467 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1470 if (dt_object_remote(orphan)) {
1471 LASSERT(lfsck->li_lpf_root_obj != NULL);
1473 idx = lfsck_find_mdt_idx_by_fid(env, lfsck, cfid);
1475 GOTO(log, rc = idx);
1477 snprintf(name, 8, "MDT%04x", idx);
1478 rc = dt_lookup(env, lfsck->li_lpf_root_obj,
1479 (struct dt_rec *)&tfid,
1480 (const struct dt_key *)name);
1482 GOTO(log, rc = (rc == -ENOENT ? -ENXIO : rc));
1484 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
1486 GOTO(log, rc = PTR_ERR(parent));
1488 if (unlikely(!dt_try_as_dir(env, parent)))
1489 GOTO(log, rc = -ENOTDIR);
1491 if (unlikely(lfsck->li_lpf_obj == NULL))
1492 GOTO(log, rc = -ENXIO);
1494 parent = lfsck->li_lpf_obj;
1497 dev = lfsck_find_dev_by_fid(env, lfsck, cfid);
1499 GOTO(log, rc = PTR_ERR(dev));
1505 namelen = snprintf(name, 31, DFID"-P-%d",
1507 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
1508 (const struct dt_key *)name);
1509 if (rc != 0 && rc != -ENOENT)
1513 rc = lfsck_lock(env, lfsck, parent, name, llh,
1514 MDS_INODELOCK_UPDATE, LCK_PW);
1518 /* Re-check whether the name conflict with othrs after taken
1520 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
1521 (const struct dt_key *)name);
1522 if (unlikely(rc == 0)) {
1530 cname->ln_name = name;
1531 cname->ln_namelen = namelen;
1533 memset(la, 0, sizeof(*la));
1534 la->la_mode = S_IFDIR | 0700;
1535 la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
1536 LA_ATIME | LA_MTIME | LA_CTIME;
1538 orphan->do_ops->do_ah_init(env, hint, parent, orphan,
1539 la->la_mode & S_IFMT);
1541 memset(dof, 0, sizeof(*dof));
1542 dof->dof_type = dt_mode_to_dft(S_IFDIR);
1544 rc = linkea_links_new(&ldata, &info->lti_linkea_buf2,
1545 cname, lfsck_dto2fid(parent));
1549 th = dt_trans_create(env, dev);
1551 GOTO(unlock1, rc = PTR_ERR(th));
1553 /* Sync the remote transaction to guarantee that the subsequent
1554 * lock against the @orphan can find the @orphan in time. */
1555 if (dt_object_remote(orphan))
1558 rc = dt_declare_create(env, orphan, la, hint, dof, th);
1562 if (unlikely(!dt_try_as_dir(env, orphan)))
1563 GOTO(stop, rc = -ENOTDIR);
1565 rc = dt_declare_ref_add(env, orphan, th);
1569 rec->rec_type = S_IFDIR;
1570 rec->rec_fid = cfid;
1571 rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
1572 (const struct dt_key *)dot, th);
1576 rec->rec_fid = lfsck_dto2fid(parent);
1577 rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
1578 (const struct dt_key *)dotdot, th);
1583 lmv->lmv_magic = LMV_MAGIC;
1584 lmv->lmv_master_mdt_index = lfsck_dev_idx(lfsck);
1585 lfsck_lmv_header_cpu_to_le(lmv2, lmv);
1586 lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
1587 rc = dt_declare_xattr_set(env, orphan, &lmv_buf,
1588 XATTR_NAME_LMV, 0, th);
1593 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
1594 ldata.ld_leh->leh_len);
1595 rc = dt_declare_xattr_set(env, orphan, &linkea_buf,
1596 XATTR_NAME_LINK, 0, th);
1600 rec->rec_fid = cfid;
1601 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1602 (const struct dt_key *)name, th);
1604 rc = dt_declare_ref_add(env, parent, th);
1609 rc = dt_trans_start_local(env, dev, th);
1613 dt_write_lock(env, orphan, 0);
1614 rc = dt_create(env, orphan, la, hint, dof, th);
1618 rc = dt_ref_add(env, orphan, th);
1622 rec->rec_fid = cfid;
1623 rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1624 (const struct dt_key *)dot, th);
1628 rec->rec_fid = lfsck_dto2fid(parent);
1629 rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1630 (const struct dt_key *)dotdot, th);
1635 rc = dt_xattr_set(env, orphan, &lmv_buf, XATTR_NAME_LMV, 0, th);
1640 rc = dt_xattr_set(env, orphan, &linkea_buf,
1641 XATTR_NAME_LINK, 0, th);
1642 dt_write_unlock(env, orphan);
1646 rec->rec_fid = cfid;
1647 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1648 (const struct dt_key *)name, th);
1650 dt_write_lock(env, parent, 0);
1651 rc = dt_ref_add(env, parent, th);
1652 dt_write_unlock(env, parent);
1655 GOTO(stop, rc = (rc == 0 ? 1 : rc));
1658 dt_write_unlock(env, orphan);
1661 rc1 = dt_trans_stop(env, dev, th);
1662 if (rc1 != 0 && rc > 0)
1669 CDEBUG(D_LFSCK, "%s: namespace LFSCK create orphan dir for "
1670 "the object "DFID", name = %s: rc = %d\n",
1671 lfsck_lfsck2name(lfsck), PFID(cfid),
1672 cname->ln_name != NULL ? cname->ln_name : "<NULL>", rc);
1674 if (parent != NULL && !IS_ERR(parent) && parent != lfsck->li_lpf_obj)
1675 lfsck_object_put(env, parent);
1678 ns->ln_flags |= LF_INCONSISTENT;
1684 * Remove the specified entry from the linkEA.
1686 * Locate the linkEA entry with the given @cname and @pfid, then
1687 * remove this entry or the other entries those are repeated with
1690 * \param[in] env pointer to the thread context
1691 * \param[in] com pointer to the lfsck component
1692 * \param[in] obj pointer to the dt_object to be handled
1693 * \param[in,out]ldata pointer to the buffer that holds the linkEA
1694 * \param[in] cname the name for the child in the parent directory
1695 * \param[in] pfid the parent directory's FID for the linkEA
1696 * \param[in] next if true, then remove the first found linkEA
1697 * entry, and move the ldata->ld_lee to next entry
1699 * \retval positive number for repaired cases
1700 * \retval 0 if nothing to be repaired
1701 * \retval negative error number on failure
1703 static int lfsck_namespace_shrink_linkea(const struct lu_env *env,
1704 struct lfsck_component *com,
1705 struct dt_object *obj,
1706 struct linkea_data *ldata,
1707 struct lu_name *cname,
1708 struct lu_fid *pfid,
1711 struct lfsck_instance *lfsck = com->lc_lfsck;
1712 struct dt_device *dev = lfsck_obj2dev(obj);
1713 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
1714 struct thandle *th = NULL;
1715 struct lustre_handle lh = { 0 };
1716 struct linkea_data ldata_new = { NULL };
1717 struct lu_buf linkea_buf;
1722 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
1723 MDS_INODELOCK_UPDATE | MDS_INODELOCK_XATTR,
1729 lfsck_linkea_del_buf(ldata, cname);
1731 lfsck_namespace_filter_linkea_entry(ldata, cname, pfid,
1733 if (ldata->ld_leh->leh_reccount > 0 ||
1734 unlikely(ldata->ld_leh->leh_overflow_time)) {
1735 lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
1736 ldata->ld_leh->leh_len);
1737 buflen = linkea_buf.lb_len;
1741 th = dt_trans_create(env, dev);
1743 GOTO(unlock1, rc = PTR_ERR(th));
1746 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
1747 XATTR_NAME_LINK, 0, th);
1749 rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, th);
1753 rc = dt_trans_start_local(env, dev, th);
1757 dt_write_lock(env, obj, 0);
1758 if (unlikely(lfsck_is_dead_obj(obj)))
1759 GOTO(unlock2, rc = -ENOENT);
1761 rc = lfsck_links_read2_with_rec(env, obj, &ldata_new);
1763 GOTO(unlock2, rc = (rc == -ENODATA ? 0 : rc));
1765 /* The specified linkEA entry has been removed by race. */
1766 rc = linkea_links_find(&ldata_new, cname, pfid);
1768 GOTO(unlock2, rc = 0);
1770 if (bk->lb_param & LPF_DRYRUN)
1771 GOTO(unlock2, rc = 1);
1774 lfsck_linkea_del_buf(&ldata_new, cname);
1776 lfsck_namespace_filter_linkea_entry(&ldata_new, cname, pfid,
1780 * linkea may change because it doesn't take lock in the first read, if
1781 * it becomes larger, restart from beginning.
1783 if ((ldata_new.ld_leh->leh_reccount > 0 ||
1784 unlikely(ldata_new.ld_leh->leh_overflow_time)) &&
1785 buflen < ldata_new.ld_leh->leh_len) {
1786 dt_write_unlock(env, obj);
1787 dt_trans_stop(env, dev, th);
1788 lfsck_buf_init(&linkea_buf, ldata_new.ld_buf->lb_buf,
1789 ldata_new.ld_leh->leh_len);
1790 buflen = linkea_buf.lb_len;
1795 rc = lfsck_links_write(env, obj, &ldata_new, th);
1797 rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, th);
1799 GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
1802 dt_write_unlock(env, obj);
1805 dt_trans_stop(env, dev, th);
1808 lfsck_ibits_unlock(&lh, LCK_EX);
1811 CDEBUG(D_LFSCK, "%s: namespace LFSCK remove %s linkEA entry "
1812 "for the object: "DFID", parent "DFID", name %.*s\n",
1813 lfsck_lfsck2name(lfsck), next ? "invalid" : "redundant",
1814 PFID(lfsck_dto2fid(obj)), PFID(pfid), cname->ln_namelen,
1818 struct lfsck_namespace *ns = com->lc_file_ram;
1820 ns->ln_flags |= LF_INCONSISTENT;
1827 * Conditionally remove the specified entry from the linkEA.
1829 * Take the parent lock firstly, then check whether the specified
1830 * name entry exists or not: if yes, do nothing; otherwise, call
1831 * lfsck_namespace_shrink_linkea() to remove the linkea entry.
1833 * \param[in] env pointer to the thread context
1834 * \param[in] com pointer to the lfsck component
1835 * \param[in] parent pointer to the parent directory
1836 * \param[in] child pointer to the child object that holds the linkEA
1837 * \param[in,out]ldata pointer to the buffer that holds the linkEA
1838 * \param[in] cname the name for the child in the parent directory
1839 * \param[in] pfid the parent directory's FID for the linkEA
1841 * \retval positive number for repaired cases
1842 * \retval 0 if nothing to be repaired
1843 * \retval negative error number on failure
1845 static int lfsck_namespace_shrink_linkea_cond(const struct lu_env *env,
1846 struct lfsck_component *com,
1847 struct dt_object *parent,
1848 struct dt_object *child,
1849 struct linkea_data *ldata,
1850 struct lu_name *cname,
1851 struct lu_fid *pfid)
1853 struct lfsck_thread_info *info = lfsck_env_info(env);
1854 struct lu_fid *cfid = &info->lti_fid3;
1855 struct lfsck_lock_handle *llh = &info->lti_llh;
1859 rc = lfsck_lock(env, com->lc_lfsck, parent, cname->ln_name, llh,
1860 MDS_INODELOCK_UPDATE, LCK_PR);
1864 dt_read_lock(env, parent, 0);
1865 if (unlikely(lfsck_is_dead_obj(parent))) {
1866 dt_read_unlock(env, parent);
1868 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
1874 rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
1875 (const struct dt_key *)cname->ln_name);
1876 dt_read_unlock(env, parent);
1878 /* It is safe to release the ldlm lock, because when the logic come
1879 * here, we have got all the needed information above whether the
1880 * linkEA entry is valid or not. It is not important that others
1881 * may add new linkEA entry after the ldlm lock released. If other
1882 * has removed the specified linkEA entry by race, then it is OK,
1883 * because the subsequent lfsck_namespace_shrink_linkea() can handle
1886 if (rc == -ENOENT) {
1887 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
1896 /* The LFSCK just found some internal status of cross-MDTs
1897 * create operation. That is normal. */
1898 if (lu_fid_eq(cfid, lfsck_dto2fid(child))) {
1899 linkea_next_entry(ldata);
1904 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata, cname,
1911 * Conditionally replace name entry in the parent.
1913 * As required, the LFSCK may re-create the lost MDT-object for dangling
1914 * name entry, but such repairing may be wrong because of bad FID in the
1915 * name entry. As the LFSCK processing, the real MDT-object may be found,
1916 * then the LFSCK should check whether the former re-created MDT-object
1917 * has been modified or not, if not, then destroy it and update the name
1918 * entry in the parent to reference the real MDT-object.
1920 * \param[in] env pointer to the thread context
1921 * \param[in] com pointer to the lfsck component
1922 * \param[in] parent pointer to the parent directory
1923 * \param[in] child pointer to the MDT-object that may be the real
1924 * MDT-object corresponding to the name entry in parent
1925 * \param[in] cfid the current FID in the name entry
1926 * \param[in] cname contains the name of the child in the parent directory
1928 * \retval positive number for repaired cases
1929 * \retval 0 if nothing to be repaired
1930 * \retval negative error number on failure
1932 static int lfsck_namespace_replace_cond(const struct lu_env *env,
1933 struct lfsck_component *com,
1934 struct dt_object *parent,
1935 struct dt_object *child,
1936 const struct lu_fid *cfid,
1937 const struct lu_name *cname)
1939 struct lfsck_thread_info *info = lfsck_env_info(env);
1940 struct lu_attr *la = &info->lti_la;
1941 struct dt_insert_rec *rec = &info->lti_dt_rec;
1943 struct lfsck_instance *lfsck = com->lc_lfsck;
1944 /* The child and its name may be on different MDTs. */
1945 struct dt_device *dev = lfsck->li_next;
1946 const char *name = cname->ln_name;
1947 const struct lu_fid *pfid = lfsck_dto2fid(parent);
1948 struct dt_object *cobj = NULL;
1949 struct lfsck_lock_handle *pllh = &info->lti_llh;
1950 struct lustre_handle clh = { 0 };
1951 struct linkea_data ldata = { NULL };
1952 struct thandle *th = NULL;
1957 /* @parent/@child may be based on lfsck->li_bottom,
1958 * but here we need the object based on the lfsck->li_next. */
1960 parent = lfsck_object_locate(dev, parent);
1962 GOTO(log, rc = PTR_ERR(parent));
1964 if (unlikely(!dt_try_as_dir(env, parent)))
1965 GOTO(log, rc = -ENOTDIR);
1967 rc = lfsck_lock(env, lfsck, parent, name, pllh,
1968 MDS_INODELOCK_UPDATE, LCK_PW);
1972 if (!fid_is_sane(cfid)) {
1977 cobj = lfsck_object_find_by_dev(env, dev, cfid);
1980 if (rc == -ENOENT) {
1988 if (!dt_object_exists(cobj)) {
1993 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
1994 (const struct dt_key *)name);
1995 if (rc == -ENOENT) {
2003 /* Someone changed the name entry, cannot replace it. */
2004 if (!lu_fid_eq(cfid, &tfid))
2007 /* lock the object to be destroyed. */
2008 rc = lfsck_ibits_lock(env, lfsck, cobj, &clh,
2009 MDS_INODELOCK_UPDATE |
2010 MDS_INODELOCK_UPDATE | MDS_INODELOCK_XATTR,
2015 if (unlikely(lfsck_is_dead_obj(cobj))) {
2020 rc = dt_attr_get(env, cobj, la);
2024 /* The object has been modified by other(s), or it is not created by
2025 * LFSCK, the two cases are indistinguishable. So cannot replace it. */
2026 if (la->la_ctime != 0)
2029 if (S_ISREG(la->la_mode)) {
2030 rc = dt_xattr_get(env, cobj, &LU_BUF_NULL, XATTR_NAME_LOV);
2031 /* If someone has created related OST-object(s),
2033 if ((rc > 0) || (rc < 0 && rc != -ENODATA))
2034 GOTO(log, rc = (rc > 0 ? 0 : rc));
2038 dt_read_lock(env, child, 0);
2039 rc = lfsck_links_read2_with_rec(env, child, &ldata);
2040 dt_read_unlock(env, child);
2042 /* Someone changed the child, no need to replace. */
2049 rc = linkea_links_find(&ldata, cname, pfid);
2050 /* Someone moved the child, no need to replace. */
2054 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2057 th = dt_trans_create(env, dev);
2059 GOTO(log, rc = PTR_ERR(th));
2062 rc = dt_declare_destroy(env, cobj, th);
2067 rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
2071 rec->rec_type = S_IFDIR;
2072 rec->rec_fid = lfsck_dto2fid(child);
2073 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
2074 (const struct dt_key *)name, th);
2078 rc = dt_trans_start_local(env, dev, th);
2083 rc = dt_destroy(env, cobj, th);
2088 /* The old name entry maybe not exist. */
2089 rc = dt_delete(env, parent, (const struct dt_key *)name, th);
2090 if (rc != 0 && rc != -ENOENT)
2093 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
2094 (const struct dt_key *)name, th);
2096 GOTO(stop, rc = (rc == 0 ? 1 : rc));
2099 dt_trans_stop(env, dev, th);
2102 lfsck_ibits_unlock(&clh, LCK_EX);
2105 if (cobj != NULL && !IS_ERR(cobj))
2106 lfsck_object_put(env, cobj);
2108 CDEBUG(D_LFSCK, "%s: namespace LFSCK conditionally destroy the "
2109 "object "DFID" because of conflict with the object "DFID
2110 " under the parent "DFID" with name %s: rc = %d\n",
2111 lfsck_lfsck2name(lfsck), PFID(cfid),
2112 PFID(lfsck_dto2fid(child)), PFID(pfid), name, rc);
2118 * Overwrite the linkEA for the object with the given ldata.
2120 * The caller should take the ldlm lock before the calling.
2122 * \param[in] env pointer to the thread context
2123 * \param[in] com pointer to the lfsck component
2124 * \param[in] obj pointer to the dt_object to be handled
2125 * \param[in] ldata pointer to the new linkEA data
2127 * \retval positive number for repaired cases
2128 * \retval 0 if nothing to be repaired
2129 * \retval negative error number on failure
2131 int lfsck_namespace_rebuild_linkea(const struct lu_env *env,
2132 struct lfsck_component *com,
2133 struct dt_object *obj,
2134 struct linkea_data *ldata)
2136 struct lfsck_instance *lfsck = com->lc_lfsck;
2137 struct dt_device *dev = lfsck_obj2dev(obj);
2138 struct thandle *th = NULL;
2139 struct lu_buf linkea_buf;
2143 th = dt_trans_create(env, dev);
2145 GOTO(log, rc = PTR_ERR(th));
2147 lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
2148 ldata->ld_leh->leh_len);
2149 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
2150 XATTR_NAME_LINK, 0, th);
2154 rc = dt_trans_start_local(env, dev, th);
2158 dt_write_lock(env, obj, 0);
2159 if (unlikely(lfsck_is_dead_obj(obj)))
2160 GOTO(unlock, rc = 0);
2162 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2163 GOTO(unlock, rc = 1);
2165 rc = dt_xattr_set(env, obj, &linkea_buf,
2166 XATTR_NAME_LINK, 0, th);
2168 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2171 dt_write_unlock(env, obj);
2174 dt_trans_stop(env, dev, th);
2177 CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild linkEA for the "
2178 "object "DFID": rc = %d\n",
2179 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
2182 struct lfsck_namespace *ns = com->lc_file_ram;
2184 ns->ln_flags |= LF_INCONSISTENT;
2191 * Repair invalid name entry.
2193 * If the name entry contains invalid information, such as bad file type
2194 * or (and) corrupted object FID, then either remove the name entry or
2195 * udpate the name entry with the given (right) information.
2197 * \param[in] env pointer to the thread context
2198 * \param[in] com pointer to the lfsck component
2199 * \param[in] parent pointer to the parent directory
2200 * \param[in] child pointer to the object referenced by the name entry
2201 * \param[in] name the old name of the child under the parent directory
2202 * \param[in] name2 the new name of the child under the parent directory
2203 * \param[in] type the type claimed by the name entry
2204 * \param[in] update update the name entry if true; otherwise, remove it
2205 * \param[in] dec decrease the parent nlink count if true
2207 * \retval positive number for repaired successfully
2208 * \retval 0 if nothing to be repaired
2209 * \retval negative error number on failure
2211 int lfsck_namespace_repair_dirent(const struct lu_env *env,
2212 struct lfsck_component *com,
2213 struct dt_object *parent,
2214 struct dt_object *child,
2215 const char *name, const char *name2,
2216 __u16 type, bool update, bool dec)
2218 struct lfsck_thread_info *info = lfsck_env_info(env);
2219 struct dt_insert_rec *rec = &info->lti_dt_rec;
2220 const struct lu_fid *pfid = lfsck_dto2fid(parent);
2221 struct lu_fid cfid = {0};
2223 struct lfsck_instance *lfsck = com->lc_lfsck;
2224 struct dt_device *dev = lfsck->li_next;
2225 struct thandle *th = NULL;
2226 struct lfsck_lock_handle *llh = &info->lti_llh;
2227 struct lustre_handle lh = { 0 };
2232 cfid = *lfsck_dto2fid(child);
2233 parent = lfsck_object_locate(dev, parent);
2235 GOTO(log, rc = PTR_ERR(parent));
2237 if (unlikely(!dt_try_as_dir(env, parent)))
2238 GOTO(log, rc = -ENOTDIR);
2240 if (!update || strcmp(name, name2) == 0)
2241 rc = lfsck_lock(env, lfsck, parent, name, llh,
2242 MDS_INODELOCK_UPDATE, LCK_PW);
2244 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
2245 MDS_INODELOCK_UPDATE, LCK_PW);
2249 th = dt_trans_create(env, dev);
2251 GOTO(unlock1, rc = PTR_ERR(th));
2253 rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
2258 rec->rec_type = lfsck_object_type(child) & S_IFMT;
2259 LASSERT(!fid_is_zero(&cfid));
2260 rec->rec_fid = &cfid;
2261 rc = dt_declare_insert(env, parent,
2262 (const struct dt_rec *)rec,
2263 (const struct dt_key *)name2, th);
2268 if (dec && S_ISDIR(type)) {
2269 rc = dt_declare_ref_del(env, parent, th);
2274 rc = dt_trans_start_local(env, dev, th);
2279 dt_write_lock(env, parent, 0);
2280 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
2281 (const struct dt_key *)name);
2282 /* Someone has removed the bad name entry by race. */
2284 GOTO(unlock2, rc = 0);
2289 /* Someone has removed the bad name entry and reused it for other
2290 * object by race. */
2291 if (!lu_fid_eq(&tfid, &cfid))
2292 GOTO(unlock2, rc = 0);
2294 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2295 GOTO(unlock2, rc = 1);
2297 rc = dt_delete(env, parent, (const struct dt_key *)name, th);
2302 rc = dt_insert(env, parent,
2303 (const struct dt_rec *)rec,
2304 (const struct dt_key *)name2, th);
2309 if (dec && S_ISDIR(type)) {
2310 rc = dt_ref_del(env, parent, th);
2315 GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
2318 dt_write_unlock(env, parent);
2321 dt_trans_stop(env, dev, th);
2323 /* We are not sure whether the child will become orphan or not.
2324 * Record it in the LFSCK trace file for further checking in
2325 * the second-stage scanning. */
2326 if (!update && !dec && child && rc == 0)
2327 lfsck_namespace_trace_update(env, com, &cfid,
2328 LNTF_CHECK_LINKEA, true);
2331 /* It is harmless even if unlock the unused lock_handle */
2332 lfsck_ibits_unlock(&lh, LCK_PW);
2336 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found bad name "
2337 "entry for: parent "DFID", child "DFID", name %s, type "
2338 "in name entry %o, type claimed by child %o. repair it "
2339 "by %s with new name2 %s: rc = %d\n",
2340 lfsck_lfsck2name(lfsck), PFID(pfid), PFID(&cfid),
2341 name, type, update ? lfsck_object_type(child) : 0,
2342 update ? "updating" : "removing", name2, rc);
2345 struct lfsck_namespace *ns = com->lc_file_ram;
2347 ns->ln_flags |= LF_INCONSISTENT;
2354 * Update the ".." name entry for the given object.
2356 * The object's ".." is corrupted, this function will update the ".." name
2357 * entry with the given pfid, and the linkEA with the given ldata.
2359 * The caller should take the ldlm lock before the calling.
2361 * \param[in] env pointer to the thread context
2362 * \param[in] com pointer to the lfsck component
2363 * \param[in] obj pointer to the dt_object to be handled
2364 * \param[in] pfid the new fid for the object's ".." name entry
2365 * \param[in] cname the name for the @obj in the parent directory
2367 * \retval positive number for repaired cases
2368 * \retval 0 if nothing to be repaired
2369 * \retval negative error number on failure
2371 static int lfsck_namespace_repair_unmatched_pairs(const struct lu_env *env,
2372 struct lfsck_component *com,
2373 struct dt_object *obj,
2374 const struct lu_fid *pfid,
2375 struct lu_name *cname)
2377 struct lfsck_thread_info *info = lfsck_env_info(env);
2378 struct dt_insert_rec *rec = &info->lti_dt_rec;
2379 struct lfsck_instance *lfsck = com->lc_lfsck;
2380 struct dt_device *dev = lfsck_obj2dev(obj);
2381 struct thandle *th = NULL;
2382 struct linkea_data ldata = { NULL };
2383 struct lu_buf linkea_buf;
2387 LASSERT(!dt_object_remote(obj));
2388 LASSERT(S_ISDIR(lfsck_object_type(obj)));
2390 rc = linkea_links_new(&ldata, &info->lti_big_buf, cname, pfid);
2394 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
2395 ldata.ld_leh->leh_len);
2397 th = dt_trans_create(env, dev);
2399 GOTO(log, rc = PTR_ERR(th));
2401 rc = dt_declare_delete(env, obj, (const struct dt_key *)dotdot, th);
2405 rec->rec_type = S_IFDIR;
2406 rec->rec_fid = pfid;
2407 rc = dt_declare_insert(env, obj, (const struct dt_rec *)rec,
2408 (const struct dt_key *)dotdot, th);
2412 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
2413 XATTR_NAME_LINK, 0, th);
2417 rc = dt_trans_start_local(env, dev, th);
2421 dt_write_lock(env, obj, 0);
2422 if (unlikely(lfsck_is_dead_obj(obj)))
2423 GOTO(unlock, rc = 0);
2425 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2426 GOTO(unlock, rc = 1);
2428 /* The old ".." name entry maybe not exist. */
2429 dt_delete(env, obj, (const struct dt_key *)dotdot, th);
2431 rc = dt_insert(env, obj, (const struct dt_rec *)rec,
2432 (const struct dt_key *)dotdot, th);
2436 rc = lfsck_links_write(env, obj, &ldata, th);
2438 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2441 dt_write_unlock(env, obj);
2444 dt_trans_stop(env, dev, th);
2447 CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild dotdot name entry for "
2448 "the object "DFID", new parent "DFID": rc = %d\n",
2449 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)),
2453 struct lfsck_namespace *ns = com->lc_file_ram;
2455 ns->ln_flags |= LF_INCONSISTENT;
2462 * Handle orphan @obj during Double Scan Directory.
2464 * Remove the @obj's current (invalid) linkEA entries, and insert
2465 * it in the directory .lustre/lost+found/MDTxxxx/ with the name:
2466 * ${FID}-${PFID}-D-${conflict_version}
2468 * The caller should take the ldlm lock before the calling.
2470 * \param[in] env pointer to the thread context
2471 * \param[in] com pointer to the lfsck component
2472 * \param[in] obj pointer to the orphan object to be handled
2473 * \param[in] pfid the new fid for the object's ".." name entry
2474 * \param[in,out] lh ldlm lock handler for the given @obj
2475 * \param[out] type to tell the caller what the inconsistency is
2477 * \retval positive number for repaired cases
2478 * \retval 0 if nothing to be repaired
2479 * \retval negative error number on failure
2482 lfsck_namespace_dsd_orphan(const struct lu_env *env,
2483 struct lfsck_component *com,
2484 struct dt_object *obj,
2485 const struct lu_fid *pfid,
2486 struct lustre_handle *lh,
2487 enum lfsck_namespace_inconsistency_type *type)
2489 struct lfsck_thread_info *info = lfsck_env_info(env);
2490 struct lfsck_namespace *ns = com->lc_file_ram;
2494 /* Remove the unrecognized linkEA. */
2495 rc = lfsck_namespace_links_remove(env, com, obj);
2496 lfsck_ibits_unlock(lh, LCK_EX);
2497 if (rc < 0 && rc != -ENODATA)
2500 *type = LNIT_MUL_REF;
2502 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
2503 * ever tried to verify some remote MDT-object that resides on this
2504 * MDT, but this MDT failed to respond such request. So means there
2505 * may be some remote name entry on other MDT that references this
2506 * object with another name, so we cannot know whether this linkEA
2507 * is valid or not. So keep it there and maybe resolved when next
2509 if (ns->ln_flags & LF_INCOMPLETE)
2512 /* The unique linkEA is invalid, even if the ".." name entry may be
2513 * valid, we still cannot know via which name entry this directory
2514 * will be referenced. Then handle it as pure orphan. */
2515 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2516 "-"DFID, PFID(pfid));
2517 rc = lfsck_namespace_insert_orphan(env, com, obj,
2518 info->lti_tmpbuf, "D", NULL);
2524 * Double Scan Directory object for single linkEA entry case.
2526 * The given @child has unique linkEA entry. If the linkEA entry is valid,
2527 * then check whether the name is in the namespace or not, if not, add the
2528 * missing name entry back to namespace. If the linkEA entry is invalid,
2529 * then remove it and insert the @child in the .lustre/lost+found/MDTxxxx/
2532 * \param[in] env pointer to the thread context
2533 * \param[in] com pointer to the lfsck component
2534 * \param[in] child pointer to the directory to be double scanned
2535 * \param[in] pfid the FID corresponding to the ".." entry
2536 * \param[in] ldata pointer to the linkEA data for the given @child
2537 * \param[in,out] lh ldlm lock handler for the given @child
2538 * \param[out] type to tell the caller what the inconsistency is
2539 * \param[in] retry if found inconsistency, but the caller does not hold
2540 * ldlm lock on the @child, then set @retry as true
2541 * \param[in] unknown set if does not know how to repair the inconsistency
2543 * \retval positive number for repaired cases
2544 * \retval 0 if nothing to be repaired
2545 * \retval negative error number on failure
2548 lfsck_namespace_dsd_single(const struct lu_env *env,
2549 struct lfsck_component *com,
2550 struct dt_object *child,
2551 const struct lu_fid *pfid,
2552 struct linkea_data *ldata,
2553 struct lustre_handle *lh,
2554 enum lfsck_namespace_inconsistency_type *type,
2555 bool *retry, bool *unknown)
2557 struct lfsck_thread_info *info = lfsck_env_info(env);
2558 struct lu_name *cname = &info->lti_name;
2559 const struct lu_fid *cfid = lfsck_dto2fid(child);
2561 struct lfsck_namespace *ns = com->lc_file_ram;
2562 struct lfsck_instance *lfsck = com->lc_lfsck;
2563 struct dt_object *parent = NULL;
2564 struct lmv_mds_md_v1 *lmv;
2568 rc = lfsck_namespace_unpack_linkea_entry(ldata, cname, &tfid,
2570 sizeof(info->lti_key));
2571 /* The unique linkEA entry with bad parent will be handled as orphan. */
2573 if (!lustre_handle_is_used(lh) && retry != NULL)
2576 rc = lfsck_namespace_dsd_orphan(env, com, child,
2582 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
2584 GOTO(out, rc = PTR_ERR(parent));
2586 /* We trust the unique linkEA entry in spite of whether it matches the
2587 * ".." name entry or not. Because even if the linkEA entry is wrong
2588 * and the ".." name entry is right, we still cannot know via which
2589 * name entry the child will be referenced, since all known entries
2590 * have been verified during the first-stage scanning. */
2591 if (!dt_object_exists(parent)) {
2592 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2593 * has ever tried to verify some remote MDT-object that resides
2594 * on this MDT, but this MDT failed to respond such request. So
2595 * means there may be some remote name entry on other MDT that
2596 * references this object with another name, so we cannot know
2597 * whether this linkEA is valid or not. So keep it there and
2598 * maybe resolved when next LFSCK run. */
2599 if (ns->ln_flags & LF_INCOMPLETE)
2602 if (!lustre_handle_is_used(lh) && retry != NULL) {
2608 lfsck_ibits_unlock(lh, LCK_EX);
2611 lmv = &info->lti_lmv;
2612 rc = lfsck_read_stripe_lmv(env, lfsck, child, lmv);
2613 if (rc != 0 && rc != -ENODATA)
2616 if (rc == -ENODATA || lmv->lmv_magic != LMV_MAGIC_STRIPE) {
2618 } else if (lfsck_shard_name_to_index(env,
2619 cname->ln_name, cname->ln_namelen,
2620 S_IFDIR, cfid) < 0) {
2621 /* It is an invalid name entry, we
2622 * cannot trust the parent also. */
2623 rc = lfsck_namespace_shrink_linkea(env, com, child,
2624 ldata, cname, &tfid, true);
2628 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2629 "-"DFID, PFID(pfid));
2630 rc = lfsck_namespace_insert_orphan(env, com, child,
2631 info->lti_tmpbuf, "S", NULL);
2636 /* Create the lost parent as an orphan. */
2637 rc = lfsck_namespace_create_orphan_dir(env, com, parent, lmv);
2639 /* Add the missing name entry to the parent. */
2640 rc = lfsck_namespace_insert_normal(env, com, parent,
2642 if (unlikely(rc == -EEXIST)) {
2643 /* Unfortunately, someone reused the name
2644 * under the parent by race. So we have
2645 * to remove the linkEA entry from
2646 * current child object. It means that the
2647 * LFSCK cannot recover the system
2648 * totally back to its original status,
2649 * but it is necessary to make the
2650 * current system to be consistent. */
2651 rc = lfsck_namespace_shrink_linkea(env,
2653 cname, &tfid, true);
2655 snprintf(info->lti_tmpbuf,
2656 sizeof(info->lti_tmpbuf),
2657 "-"DFID, PFID(pfid));
2658 rc = lfsck_namespace_insert_orphan(env,
2659 com, child, info->lti_tmpbuf,
2666 } /* !dt_object_exists(parent) */
2668 /* The unique linkEA entry with bad parent will be handled as orphan. */
2669 if (unlikely(!dt_try_as_dir(env, parent))) {
2670 if (!lustre_handle_is_used(lh) && retry != NULL)
2673 rc = lfsck_namespace_dsd_orphan(env, com, child,
2679 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
2680 (const struct dt_key *)cname->ln_name);
2681 if (rc == -ENOENT) {
2682 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2683 * has ever tried to verify some remote MDT-object that resides
2684 * on this MDT, but this MDT failed to respond such request. So
2685 * means there may be some remote name entry on other MDT that
2686 * references this object with another name, so we cannot know
2687 * whether this linkEA is valid or not. So keep it there and
2688 * maybe resolved when next LFSCK run. */
2689 if (ns->ln_flags & LF_INCOMPLETE)
2692 if (!lustre_handle_is_used(lh) && retry != NULL) {
2698 lfsck_ibits_unlock(lh, LCK_EX);
2699 rc = lfsck_namespace_check_name(env, lfsck, parent, child,
2707 /* It is an invalid name entry, drop it. */
2708 if (unlikely(rc > 0)) {
2709 rc = lfsck_namespace_shrink_linkea(env, com, child,
2710 ldata, cname, &tfid, true);
2712 snprintf(info->lti_tmpbuf,
2713 sizeof(info->lti_tmpbuf),
2714 "-"DFID, PFID(pfid));
2715 rc = lfsck_namespace_insert_orphan(env, com,
2716 child, info->lti_tmpbuf, "D", NULL);
2722 /* Add the missing name entry back to the namespace. */
2723 rc = lfsck_namespace_insert_normal(env, com, parent, child,
2725 if (unlikely(rc == -ESTALE))
2726 /* It may happen when the remote object has been
2727 * removed, but the local MDT is not aware of that. */
2730 if (unlikely(rc == -EEXIST)) {
2731 /* Unfortunately, someone reused the name under the
2732 * parent by race. So we have to remove the linkEA
2733 * entry from current child object. It means that the
2734 * LFSCK cannot recover the system totally back to
2735 * its original status, but it is necessary to make
2736 * the current system to be consistent.
2738 * It also may be because of the LFSCK found some
2739 * internal status of create operation. Under such
2740 * case, nothing to be done. */
2741 rc = lfsck_namespace_shrink_linkea_cond(env, com,
2742 parent, child, ldata, cname, &tfid);
2744 snprintf(info->lti_tmpbuf,
2745 sizeof(info->lti_tmpbuf),
2746 "-"DFID, PFID(pfid));
2747 rc = lfsck_namespace_insert_orphan(env, com,
2748 child, info->lti_tmpbuf, "D", NULL);
2753 } /* rc == -ENOENT */
2758 if (!lu_fid_eq(&tfid, cfid)) {
2759 if (!lustre_handle_is_used(lh) && retry != NULL) {
2765 lfsck_ibits_unlock(lh, LCK_EX);
2766 /* The name entry references another MDT-object that
2767 * may be created by the LFSCK for repairing dangling
2768 * name entry. Try to replace it. */
2769 rc = lfsck_namespace_replace_cond(env, com, parent, child,
2772 rc = lfsck_namespace_dsd_orphan(env, com, child,
2778 /* Zero FID may because the remote directroy object has invalid linkEA,
2779 * or lost linkEA. Under such case, the LFSCK on this MDT does not know
2780 * how to repair the inconsistency, but the namespace LFSCK on the MDT
2781 * where its name entry resides may has more information (name, FID) to
2782 * repair such inconsistency. So here, keep the inconsistency to avoid
2783 * some imporper repairing. */
2784 if (fid_is_zero(pfid)) {
2791 /* The ".." name entry is wrong, update it. */
2792 if (!lu_fid_eq(pfid, lfsck_dto2fid(parent))) {
2793 if (!lustre_handle_is_used(lh) && retry != NULL) {
2799 *type = LNIT_UNMATCHED_PAIRS;
2800 rc = lfsck_namespace_repair_unmatched_pairs(env, com, child,
2801 lfsck_dto2fid(parent), cname);
2807 if (parent != NULL && !IS_ERR(parent))
2808 lfsck_object_put(env, parent);
2814 * Double Scan Directory object for multiple linkEA entries case.
2816 * The given @child has multiple linkEA entries. There is at most one linkEA
2817 * entry will be valid, all the others will be removed. Firstly, the function
2818 * will try to find out the linkEA entry for which the name entry exists under
2819 * the given parent (@pfid). If there is no linkEA entry that matches the given
2820 * ".." name entry, then tries to find out the first linkEA entry that both the
2821 * parent and the name entry exist to rebuild a new ".." name entry.
2823 * \param[in] env pointer to the thread context
2824 * \param[in] com pointer to the lfsck component
2825 * \param[in] child pointer to the directory to be double scanned
2826 * \param[in] pfid the FID corresponding to the ".." entry
2827 * \param[in] ldata pointer to the linkEA data for the given @child
2828 * \param[in,out] lh ldlm lock handler for the given @child
2829 * \param[out] type to tell the caller what the inconsistency is
2830 * \param[in] lpf true if the ".." entry is under lost+found/MDTxxxx/
2831 * \param[in] unknown set if does not know how to repair the inconsistency
2833 * \retval positive number for repaired cases
2834 * \retval 0 if nothing to be repaired
2835 * \retval negative error number on failure
2838 lfsck_namespace_dsd_multiple(const struct lu_env *env,
2839 struct lfsck_component *com,
2840 struct dt_object *child,
2841 const struct lu_fid *pfid,
2842 struct linkea_data *ldata,
2843 struct lustre_handle *lh,
2844 enum lfsck_namespace_inconsistency_type *type,
2845 bool lpf, bool *unknown)
2847 struct lfsck_thread_info *info = lfsck_env_info(env);
2848 struct lu_name *cname = &info->lti_name;
2849 const struct lu_fid *cfid = lfsck_dto2fid(child);
2850 struct lu_fid *pfid2 = &info->lti_fid3;
2852 struct lfsck_namespace *ns = com->lc_file_ram;
2853 struct lfsck_instance *lfsck = com->lc_lfsck;
2854 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
2855 struct dt_object *parent = NULL;
2856 struct linkea_data ldata_new = { NULL };
2857 int dirent_count = 0;
2863 while (ldata->ld_lee != NULL) {
2864 rc = lfsck_namespace_unpack_linkea_entry(ldata, cname, &tfid,
2866 sizeof(info->lti_key));
2867 /* Drop invalid linkEA entry. */
2869 lfsck_linkea_del_buf(ldata, cname);
2873 /* Drop repeated linkEA entries. */
2874 lfsck_namespace_filter_linkea_entry(ldata, cname, &tfid, true);
2876 /* If current dotdot is the .lustre/lost+found/MDTxxxx/,
2877 * then it is possible that: the directry object has ever
2878 * been lost, but its name entry was there. In the former
2879 * LFSCK run, during the first-stage scanning, the LFSCK
2880 * found the dangling name entry, but it did not recreate
2881 * the lost object, and when moved to the second-stage
2882 * scanning, some children objects of the lost directory
2883 * object were found, then the LFSCK recreated such lost
2884 * directory object as an orphan.
2886 * When the LFSCK runs again, if the dangling name is still
2887 * there, the LFSCK should move the orphan directory object
2888 * back to the normal namespace. */
2889 if (!lpf && !fid_is_zero(pfid) &&
2890 !lu_fid_eq(pfid, &tfid) && once) {
2891 linkea_next_entry(ldata);
2895 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
2897 RETURN(PTR_ERR(parent));
2899 if (!dt_object_exists(parent)) {
2900 lfsck_object_put(env, parent);
2901 if (ldata->ld_leh->leh_reccount > 1) {
2902 /* If it is NOT the last linkEA entry, then
2903 * there is still other chance to make the
2904 * child to be visible via other parent, then
2905 * remove this linkEA entry. */
2906 lfsck_linkea_del_buf(ldata, cname);
2913 /* The linkEA entry with bad parent will be removed. */
2914 if (unlikely(!dt_try_as_dir(env, parent))) {
2915 lfsck_object_put(env, parent);
2916 lfsck_linkea_del_buf(ldata, cname);
2920 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
2921 (const struct dt_key *)cname->ln_name);
2922 *pfid2 = *lfsck_dto2fid(parent);
2923 if (rc == -ENOENT) {
2924 lfsck_object_put(env, parent);
2925 linkea_next_entry(ldata);
2930 lfsck_object_put(env, parent);
2935 if (lu_fid_eq(&tfid, cfid)) {
2936 lfsck_object_put(env, parent);
2937 /* If the parent (that is declared via linkEA entry)
2938 * directory contains the specified child, but such
2939 * parent does not match the dotdot name entry, then
2940 * trust the linkEA. */
2941 if (!fid_is_zero(pfid) && !lu_fid_eq(pfid, pfid2)) {
2942 *type = LNIT_UNMATCHED_PAIRS;
2943 rc = lfsck_namespace_repair_unmatched_pairs(env,
2944 com, child, pfid2, cname);
2950 /* It is the most common case that we find the
2951 * name entry corresponding to the linkEA entry
2952 * that matches the ".." name entry. */
2953 rc = linkea_links_new(&ldata_new, &info->lti_big_buf,
2958 rc = lfsck_namespace_rebuild_linkea(env, com, child,
2963 lfsck_linkea_del_buf(ldata, cname);
2964 linkea_first_entry(ldata);
2965 /* There may be some invalid dangling name entries under
2966 * other parent directories, remove all of them. */
2967 while (ldata->ld_lee != NULL) {
2968 rc = lfsck_namespace_unpack_linkea_entry(ldata,
2969 cname, &tfid, info->lti_key,
2970 sizeof(info->lti_key));
2974 parent = lfsck_object_find_bottom(env, lfsck,
2976 if (IS_ERR(parent)) {
2977 rc = PTR_ERR(parent);
2978 if (rc != -ENOENT &&
2979 bk->lb_param & LPF_FAILOUT)
2985 if (!dt_object_exists(parent)) {
2986 lfsck_object_put(env, parent);
2990 rc = lfsck_namespace_repair_dirent(env, com,
2991 parent, child, cname->ln_name,
2992 cname->ln_name, S_IFDIR, false, true);
2993 lfsck_object_put(env, parent);
2995 if (bk->lb_param & LPF_FAILOUT)
3004 lfsck_linkea_del_buf(ldata, cname);
3007 ns->ln_dirent_repaired += dirent_count;
3010 } /* lu_fid_eq(&tfid, lfsck_dto2fid(child)) */
3012 lfsck_ibits_unlock(lh, LCK_EX);
3013 /* The name entry references another MDT-object that may be
3014 * created by the LFSCK for repairing dangling name entry.
3015 * Try to replace it. */
3016 rc = lfsck_namespace_replace_cond(env, com, parent, child,
3018 lfsck_object_put(env, parent);
3025 lfsck_linkea_del_buf(ldata, cname);
3026 } /* while (ldata->ld_lee != NULL) */
3028 /* If there is still linkEA overflow, return. */
3029 if (unlikely(ldata->ld_leh->leh_overflow_time))
3032 linkea_first_entry(ldata);
3033 if (ldata->ld_leh->leh_reccount == 1) {
3034 rc = lfsck_namespace_dsd_single(env, com, child, pfid, ldata,
3035 lh, type, NULL, unknown);
3040 /* All linkEA entries are invalid and removed, then handle the @child
3042 if (ldata->ld_leh->leh_reccount == 0) {
3043 rc = lfsck_namespace_dsd_orphan(env, com, child, pfid, lh,
3049 /* If the dangling name entry for the orphan directory object has
3050 * been remvoed, then just check whether the directory object is
3051 * still under the .lustre/lost+found/MDTxxxx/ or not. */
3057 /* There is no linkEA entry that matches the ".." name entry. Find
3058 * the first linkEA entry that both parent and name entry exist to
3059 * rebuild a new ".." name entry. */
3069 * Repair the object's nlink attribute.
3071 * If all the known name entries have been verified, then the object's hard
3072 * link attribute should match the object's linkEA entries count unless the
3073 * object's has too many hard link to be recorded in the linkEA. Such cases
3074 * should have been marked in the LFSCK trace file. Otherwise, trust the
3075 * linkEA to update the object's nlink attribute.
3077 * \param[in] env pointer to the thread context
3078 * \param[in] com pointer to the lfsck component
3079 * \param[in] obj pointer to the dt_object to be handled
3080 * \param[in,out] la pointer to buffer to object's attribute before
3081 * and after the repairing
3083 * \retval positive number for repaired cases
3084 * \retval 0 if nothing to be repaired
3085 * \retval negative error number on failure
3087 static int lfsck_namespace_repair_nlink(const struct lu_env *env,
3088 struct lfsck_component *com,
3089 struct dt_object *obj,
3092 struct lfsck_namespace *ns = com->lc_file_ram;
3093 struct lfsck_instance *lfsck = com->lc_lfsck;
3094 struct dt_device *dev = lfsck_obj2dev(obj);
3095 const struct lu_fid *cfid = lfsck_dto2fid(obj);
3096 struct thandle *th = NULL;
3097 struct linkea_data ldata = { NULL };
3098 struct lustre_handle lh = { 0 };
3099 __u32 old = la->la_nlink;
3103 LASSERT(!dt_object_remote(obj));
3105 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
3106 MDS_INODELOCK_UPDATE, LCK_PW);
3110 th = dt_trans_create(env, dev);
3112 GOTO(log, rc = PTR_ERR(th));
3114 la->la_valid = LA_NLINK;
3115 rc = dt_declare_attr_set(env, obj, la, th);
3119 rc = dt_trans_start_local(env, dev, th);
3123 dt_write_lock(env, obj, 0);
3124 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
3125 * ever tried to verify some remote MDT-object that resides on this
3126 * MDT, but this MDT failed to respond such request. So means there
3127 * may be some remote name entry on other MDT that references this
3128 * object with another name, so we cannot know whether this linkEA
3129 * is valid or not. So keep it there and maybe resolved when next
3131 if (ns->ln_flags & LF_INCOMPLETE)
3132 GOTO(unlock, rc = 0);
3134 rc = dt_attr_get(env, obj, la);
3136 GOTO(unlock, rc = (rc == -ENOENT ? 0 : rc));
3138 rc = lfsck_links_read2_with_rec(env, obj, &ldata);
3140 GOTO(unlock, rc = (rc == -ENODATA ? 0 : rc));
3142 /* XXX: Currently, we only update the nlink attribute if the known
3143 * linkEA entries is larger than the nlink attribute. That is
3145 if (la->la_nlink >= ldata.ld_leh->leh_reccount ||
3146 unlikely(la->la_nlink == 0 ||
3147 ldata.ld_leh->leh_overflow_time))
3148 GOTO(unlock, rc = 0);
3150 la->la_nlink = ldata.ld_leh->leh_reccount;
3151 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
3152 GOTO(unlock, rc = 1);
3154 rc = dt_attr_set(env, obj, la, th);
3156 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
3159 dt_write_unlock(env, obj);
3162 dt_trans_stop(env, dev, th);
3165 lfsck_ibits_unlock(&lh, LCK_PW);
3166 CDEBUG(D_LFSCK, "%s: namespace LFSCK repaired the object "DFID"'s "
3167 "nlink count from %u to %u: rc = %d\n",
3168 lfsck_lfsck2name(lfsck), PFID(cfid), old, la->la_nlink, rc);
3171 ns->ln_flags |= LF_INCONSISTENT;
3177 * Double scan the directory object for namespace LFSCK.
3179 * This function will verify the <parent, child> pairs in the namespace tree:
3180 * the parent references the child via some name entry that should be in the
3181 * child's linkEA entry, the child should back references the parent via its
3184 * The LFSCK will scan every linkEA entry in turn until find out the first
3185 * matched pairs. If found, then all other linkEA entries will be dropped.
3186 * If all the linkEA entries cannot match the ".." name entry, then there
3187 * are serveral possible cases:
3189 * 1) If there is only one linkEA entry, then trust it as long as the PFID
3190 * in the linkEA entry is valid.
3192 * 2) If there are multiple linkEA entries, then try to find the linkEA
3193 * that matches the ".." name entry. If found, then all other entries
3194 * are invalid; otherwise, it is quite possible that the ".." name entry
3195 * is corrupted. Under such case, the LFSCK will rebuild the ".." name
3196 * entry according to the first valid linkEA entry (both the parent and
3197 * the name entry should exist).
3199 * 3) If the directory object has no (valid) linkEA entry, then the
3200 * directory object will be handled as pure orphan and inserted
3201 * in the .lustre/lost+found/MDTxxxx/ with the name:
3202 * ${self_FID}-${PFID}-D-${conflict_version}
3204 * \param[in] env pointer to the thread context
3205 * \param[in] com pointer to the lfsck component
3206 * \param[in] child pointer to the directory object to be handled
3207 * \param[in] flags to indicate the specical checking on the @child
3209 * \retval positive number for repaired cases
3210 * \retval 0 if nothing to be repaired
3211 * \retval negative error number on failure
3213 static int lfsck_namespace_double_scan_dir(const struct lu_env *env,
3214 struct lfsck_component *com,
3215 struct dt_object *child, __u8 flags)
3217 struct lfsck_thread_info *info = lfsck_env_info(env);
3218 const struct lu_fid *cfid = lfsck_dto2fid(child);
3219 struct lu_fid *pfid = &info->lti_fid2;
3220 struct lfsck_namespace *ns = com->lc_file_ram;
3221 struct lfsck_instance *lfsck = com->lc_lfsck;
3222 struct lustre_handle lh = { 0 };
3223 struct linkea_data ldata = { NULL };
3224 bool unknown = false;
3227 enum lfsck_namespace_inconsistency_type type = LNIT_BAD_LINKEA;
3231 LASSERT(!dt_object_remote(child));
3233 if (flags & LNTF_UNCERTAIN_LMV) {
3234 if (flags & LNTF_RECHECK_NAME_HASH) {
3235 rc = lfsck_namespace_scan_shard(env, com, child);
3239 ns->ln_striped_shards_scanned++;
3241 ns->ln_striped_shards_skipped++;
3245 flags &= ~(LNTF_RECHECK_NAME_HASH | LNTF_UNCERTAIN_LMV);
3249 if (flags & (LNTF_CHECK_LINKEA | LNTF_CHECK_PARENT) &&
3250 !(lfsck->li_bookmark_ram.lb_param & LPF_ALL_TGT)) {
3252 "%s: some MDT(s) maybe NOT take part in the the namespace LFSCK, then the LFSCK cannot guarantee all the name entries have been verified in first-stage scanning. So have to skip orphan related handling for the directory object "DFID" with remote name entry\n",
3253 lfsck_lfsck2name(lfsck), PFID(cfid));
3258 if (unlikely(!dt_try_as_dir(env, child)))
3259 GOTO(out, rc = -ENOTDIR);
3261 /* We only take ldlm lock on the @child when required. When the
3262 * logic comes here for the first time, it is always false. */
3266 rc = lfsck_ibits_lock(env, lfsck, child, &lh,
3267 MDS_INODELOCK_UPDATE |
3268 MDS_INODELOCK_XATTR, LCK_EX);
3273 dt_read_lock(env, child, 0);
3274 if (unlikely(lfsck_is_dead_obj(child))) {
3275 dt_read_unlock(env, child);
3280 rc = dt_lookup(env, child, (struct dt_rec *)pfid,
3281 (const struct dt_key *)dotdot);
3283 if (rc != -ENOENT && rc != -ENODATA && rc != -EINVAL) {
3284 dt_read_unlock(env, child);
3289 if (!lustre_handle_is_used(&lh)) {
3290 dt_read_unlock(env, child);
3295 } else if (lfsck->li_lpf_obj != NULL &&
3296 lu_fid_eq(pfid, lfsck_dto2fid(lfsck->li_lpf_obj))) {
3298 } else if (unlikely(!fid_is_sane(pfid))) {
3302 rc = lfsck_links_read(env, child, &ldata);
3303 dt_read_unlock(env, child);
3305 if (rc != -ENODATA && rc != -EINVAL)
3308 if (!lustre_handle_is_used(&lh))
3311 if (rc == -EINVAL && !fid_is_zero(pfid)) {
3312 /* Remove the corrupted linkEA. */
3313 rc = lfsck_namespace_links_remove(env, com, child);
3315 /* Here, because of the crashed linkEA, we
3316 * cannot know whether there is some parent
3317 * that references the child directory via
3318 * some name entry or not. So keep it there,
3319 * when the LFSCK run next time, if there is
3320 * some parent that references this object,
3321 * then the LFSCK can rebuild the linkEA;
3322 * otherwise, this object will be handled
3323 * as orphan as above. */
3326 /* 1. If we have neither ".." nor linkEA,
3327 * then it is an orphan.
3329 * 2. If we only have the ".." name entry,
3330 * but no parent references this child
3331 * directory, then handle it as orphan. */
3332 lfsck_ibits_unlock(&lh, LCK_EX);
3333 type = LNIT_MUL_REF;
3335 /* If the LFSCK is marked as LF_INCOMPLETE,
3336 * then means some MDT has ever tried to
3337 * verify some remote MDT-object that resides
3338 * on this MDT, but this MDT failed to respond
3339 * such request. So means there may be some
3340 * remote name entry on other MDT that
3341 * references this object with another name,
3342 * so we cannot know whether this linkEA is
3343 * valid or not. So keep it there and maybe
3344 * resolved when next LFSCK run. */
3345 if (ns->ln_flags & LF_INCOMPLETE)
3348 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
3349 "-"DFID, PFID(pfid));
3350 rc = lfsck_namespace_insert_orphan(env, com, child,
3351 info->lti_tmpbuf, "D", NULL);
3357 linkea_first_entry(&ldata);
3358 /* This is the most common case: the object has unique linkEA entry. */
3359 if (ldata.ld_leh->leh_reccount == 1) {
3360 rc = lfsck_namespace_dsd_single(env, com, child, pfid, &ldata,
3361 &lh, &type, &retry, &unknown);
3363 LASSERT(!lustre_handle_is_used(&lh));
3372 if (!lustre_handle_is_used(&lh))
3375 if (unlikely(ldata.ld_leh->leh_reccount == 0)) {
3376 rc = lfsck_namespace_dsd_orphan(env, com, child, pfid, &lh,
3382 /* When we come here, the cases usually like that:
3383 * 1) The directory object has a corrupted linkEA entry. During the
3384 * first-stage scanning, the LFSCK cannot know such corruption,
3385 * then it appends the right linkEA entry according to the found
3386 * name entry after the bad one.
3388 * 2) The directory object has a right linkEA entry. During the
3389 * first-stage scanning, the LFSCK finds some bad name entry,
3390 * but the LFSCK cannot aware that at that time, then it adds
3391 * the bad linkEA entry for further processing. */
3392 rc = lfsck_namespace_dsd_multiple(env, com, child, pfid, &ldata,
3393 &lh, &type, lpf, &unknown);
3398 lfsck_ibits_unlock(&lh, LCK_EX);
3401 case LNIT_BAD_LINKEA:
3402 ns->ln_linkea_repaired++;
3404 case LNIT_UNMATCHED_PAIRS:
3405 ns->ln_unmatched_pairs_repaired++;
3408 ns->ln_mul_ref_repaired++;
3416 ns->ln_unknown_inconsistency++;
3422 lfsck_namespace_linkea_stale_overflow(struct linkea_data *ldata,
3423 struct lfsck_namespace *ns)
3425 /* Both the leh_overflow_time and ln_time_latest_reset are
3426 * local time based, so need NOT to care about clock drift
3427 * among the servers. */
3428 return ldata->ld_leh->leh_overflow_time &&
3429 ldata->ld_leh->leh_overflow_time < ns->ln_time_latest_reset;
3433 * Clear the object's linkEA overflow timestamp.
3435 * If the MDT-object has too many hard links as to the linkEA cannot hold
3436 * all of them, then overflow timestamp will be set in the linkEA header.
3437 * If some hard links are removed after that, then it is possible to hold
3438 * other missed linkEA entries. If the namespace LFSCK have added all the
3439 * related linkEA entries, then it will remove the overflow timestamp.
3441 * \param[in] env pointer to the thread context
3442 * \param[in] com pointer to the lfsck component
3443 * \param[in] ldata pointer to the linkEA data for the given @obj
3444 * \param[in] obj pointer to the dt_object to be handled
3446 * \retval positive number for repaired cases
3447 * \retval 0 if nothing to be repaired
3448 * \retval negative error number on failure
3450 static int lfsck_namespace_linkea_clear_overflow(const struct lu_env *env,
3451 struct lfsck_component *com,
3452 struct linkea_data *ldata,
3453 struct dt_object *obj)
3455 struct lfsck_namespace *ns = com->lc_file_ram;
3456 struct lfsck_instance *lfsck = com->lc_lfsck;
3457 struct dt_device *dev = lfsck_obj2dev(obj);
3458 struct thandle *th = NULL;
3459 struct lustre_handle lh = { 0 };
3460 struct lu_buf linkea_buf;
3464 LASSERT(!dt_object_remote(obj));
3466 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
3467 MDS_INODELOCK_UPDATE, LCK_PW);
3471 th = dt_trans_create(env, dev);
3473 GOTO(log, rc = PTR_ERR(th));
3475 rc = dt_declare_xattr_set(env, obj,
3476 lfsck_buf_get_const(env, NULL, MAX_LINKEA_SIZE),
3477 XATTR_NAME_LINK, 0, th);
3481 rc = dt_trans_start_local(env, dev, th);
3485 dt_write_lock(env, obj, 0);
3486 rc = lfsck_links_read(env, obj, ldata);
3490 if (unlikely(!lfsck_namespace_linkea_stale_overflow(ldata, ns)))
3491 GOTO(unlock, rc = 0);
3493 ldata->ld_leh->leh_overflow_time = 0;
3494 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
3495 GOTO(unlock, rc = 1);
3497 /* If all known entries are in the linkEA, then the 'leh_reccount'
3498 * should NOT be zero. */
3499 LASSERT(ldata->ld_leh->leh_reccount > 0);
3501 lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
3502 ldata->ld_leh->leh_len);
3503 rc = dt_xattr_set(env, obj, &linkea_buf, XATTR_NAME_LINK, 0, th);
3504 if (unlikely(rc == -ENOSPC))
3512 dt_write_unlock(env, obj);
3515 dt_trans_stop(env, dev, th);
3518 lfsck_ibits_unlock(&lh, LCK_PW);
3519 CDEBUG(D_LFSCK, "%s: clear linkea overflow timestamp for the object "
3521 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
3527 * Verify the object's agent entry.
3529 * If the object claims to have agent entry but the linkEA does not contain
3530 * remote parent, then remove the agent entry. Otherwise, if the object has
3531 * no agent entry but its linkEA contains remote parent, then will generate
3532 * agent entry for it.
3534 * \param[in] env pointer to the thread context
3535 * \param[in] com pointer to the lfsck component
3536 * \param[in] obj pointer to the dt_object to be handled
3538 * \retval positive number for repaired cases
3539 * \retval 0 if nothing to be repaired
3540 * \retval negative error number on failure
3542 static int lfsck_namespace_check_agent_entry(const struct lu_env *env,
3543 struct lfsck_component *com,
3544 struct dt_object *obj)
3546 struct linkea_data ldata = { NULL };
3547 struct lfsck_thread_info *info = lfsck_env_info(env);
3548 struct lfsck_namespace *ns = com->lc_file_ram;
3549 struct lfsck_instance *lfsck = com->lc_lfsck;
3550 struct lu_fid *pfid = &info->lti_fid2;
3551 struct lu_name *cname = &info->lti_name;
3552 struct lu_seq_range *range = &info->lti_range;
3553 struct seq_server_site *ss = lfsck_dev_site(lfsck);
3554 __u32 idx = lfsck_dev_idx(lfsck);
3556 bool remote = false;
3559 if (!(lfsck->li_bookmark_ram.lb_param & LPF_ALL_TGT))
3562 rc = lfsck_links_read_with_rec(env, obj, &ldata);
3563 if (rc == -ENOENT || rc == -ENODATA)
3566 if (rc && rc != -EINVAL)
3569 /* We check the agent entry again after verifying the linkEA
3570 * successfully. So invalid linkEA should be dryrun mode. */
3571 if (rc == -EINVAL || unlikely(!ldata.ld_leh->leh_reccount))
3574 linkea_first_entry(&ldata);
3575 while (ldata.ld_lee != NULL && !remote) {
3576 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
3578 if (!linkea_entry_is_valid(&ldata, cname, pfid))
3581 fld_range_set_mdt(range);
3582 rc = fld_server_lookup(env, ss->ss_server_fld,
3583 fid_seq(pfid), range);
3585 GOTO(out, rc = (rc == -ENOENT ? 0 : rc));
3587 if (range->lsr_index != idx)
3590 linkea_next_entry(&ldata);
3593 if ((lu_object_has_agent_entry(&obj->do_lu) && !remote) ||
3594 (!lu_object_has_agent_entry(&obj->do_lu) && remote)) {
3595 struct dt_device *dev = lfsck_obj2dev(obj);
3596 struct linkea_data ldata2 = { NULL };
3597 struct lustre_handle lh = { 0 };
3598 struct lu_buf linkea_buf;
3599 struct thandle *handle;
3601 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
3604 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
3605 MDS_INODELOCK_UPDATE |
3606 MDS_INODELOCK_XATTR, LCK_EX);
3610 handle = dt_trans_create(env, dev);
3612 GOTO(unlock, rc = PTR_ERR(handle));
3614 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
3615 ldata.ld_leh->leh_len);
3616 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
3617 XATTR_NAME_LINK, LU_XATTR_REPLACE, handle);
3621 rc = dt_trans_start_local(env, dev, handle);
3625 dt_write_lock(env, obj, 0);
3626 rc = lfsck_links_read2_with_rec(env, obj, &ldata2);
3628 if (rc == -ENOENT || rc == -ENODATA)
3633 /* If someone changed linkEA by race, then the agent
3634 * entry will be updated by lower layer automatically. */
3635 if (ldata.ld_leh->leh_len != ldata2.ld_leh->leh_len ||
3636 memcmp(ldata.ld_buf->lb_buf, ldata2.ld_buf->lb_buf,
3637 ldata.ld_leh->leh_len) != 0)
3638 GOTO(unlock2, rc = 0);
3640 rc = dt_xattr_set(env, obj, &linkea_buf, XATTR_NAME_LINK,
3641 LU_XATTR_REPLACE, handle);
3648 dt_write_unlock(env, obj);
3650 dt_trans_stop(env, dev, handle);
3652 lfsck_ibits_unlock(&lh, LCK_EX);
3659 ns->ln_agent_entries_repaired++;
3661 CDEBUG(D_LFSCK, "%s: repair agent entry for "DFID": rc = %d\n",
3662 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
3667 * Double scan the MDT-object for namespace LFSCK.
3669 * If the MDT-object contains invalid or repeated linkEA entries, then drop
3670 * those entries from the linkEA; if the linkEA becomes empty or the object
3671 * has no linkEA, then it is an orphan and will be added into the directory
3672 * .lustre/lost+found/MDTxxxx/; if the remote parent is lost, then recreate
3673 * the remote parent; if the name entry corresponding to some linkEA entry
3674 * is lost, then add the name entry back to the namespace.
3676 * \param[in] env pointer to the thread context
3677 * \param[in] com pointer to the lfsck component
3678 * \param[in] child pointer to the dt_object to be handled
3679 * \param[in] flags some hints to indicate how the @child should be handled
3681 * \retval positive number for repaired cases
3682 * \retval 0 if nothing to be repaired
3683 * \retval negative error number on failure
3685 static int lfsck_namespace_double_scan_one(const struct lu_env *env,
3686 struct lfsck_component *com,
3687 struct dt_object *child, __u8 flags)
3689 struct lfsck_thread_info *info = lfsck_env_info(env);
3690 struct lu_attr *la = &info->lti_la;
3691 struct lu_name *cname = &info->lti_name;
3692 struct lu_fid *pfid = &info->lti_fid;
3693 struct lu_fid *cfid = &info->lti_fid2;
3694 struct lfsck_instance *lfsck = com->lc_lfsck;
3695 struct lfsck_namespace *ns = com->lc_file_ram;
3696 struct dt_object *parent = NULL;
3697 struct linkea_data ldata = { NULL };
3698 bool repaired = false;
3703 dt_read_lock(env, child, 0);
3704 if (unlikely(lfsck_is_dead_obj(child))) {
3705 dt_read_unlock(env, child);
3710 if (S_ISDIR(lfsck_object_type(child))) {
3711 dt_read_unlock(env, child);
3712 rc = lfsck_namespace_double_scan_dir(env, com, child, flags);
3713 if (!rc && flags & LNTF_CHECK_AGENT_ENTRY)
3714 rc = lfsck_namespace_check_agent_entry(env, com, child);
3719 rc = lfsck_links_read(env, child, &ldata);
3720 dt_read_unlock(env, child);
3722 if (rc == -EINVAL) {
3723 struct lustre_handle lh = { 0 };
3725 rc = lfsck_ibits_lock(env, com->lc_lfsck, child, &lh,
3726 MDS_INODELOCK_UPDATE |
3727 MDS_INODELOCK_XATTR, LCK_EX);
3729 rc = lfsck_namespace_links_remove(env, com, child);
3730 lfsck_ibits_unlock(&lh, LCK_EX);
3739 if (!(ns->ln_flags & LF_INCOMPLETE) &&
3740 unlikely(lfsck_namespace_linkea_stale_overflow(&ldata, ns))) {
3741 rc = lfsck_namespace_linkea_clear_overflow(env, com, &ldata,
3747 ns->ln_linkea_overflow_cleared++;
3750 linkea_first_entry(&ldata);
3751 while (ldata.ld_lee != NULL) {
3752 rc = lfsck_namespace_unpack_linkea_entry(&ldata, cname, pfid,
3754 sizeof(info->lti_key));
3755 /* Invalid PFID in the linkEA entry. */
3757 rc = lfsck_namespace_shrink_linkea(env, com, child,
3758 &ldata, cname, pfid, true);
3768 rc = lfsck_namespace_filter_linkea_entry(&ldata, cname, pfid,
3770 /* Found repeated linkEA entries */
3772 rc = lfsck_namespace_shrink_linkea(env, com, child,
3773 &ldata, cname, pfid, false);
3785 parent = lfsck_object_find_bottom(env, lfsck, pfid);
3787 GOTO(out, rc = PTR_ERR(parent));
3789 if (!dt_object_exists(parent)) {
3792 if (ldata.ld_leh->leh_reccount > 1) {
3793 /* If it is NOT the last linkEA entry, then
3794 * there is still other chance to make the
3795 * child to be visible via other parent, then
3796 * remove this linkEA entry. */
3797 rc = lfsck_namespace_shrink_linkea(env, com,
3798 child, &ldata, cname, pfid, true);
3800 /* If the LFSCK is marked as LF_INCOMPLETE,
3801 * then means some MDT has ever tried to
3802 * verify some remote MDT-object that resides
3803 * on this MDT, but this MDT failed to respond
3804 * such request. So means there may be some
3805 * remote name entry on other MDT that
3806 * references this object with another name,
3807 * so we cannot know whether this linkEA is
3808 * valid or not. So keep it there and maybe
3809 * resolved when next LFSCK run. */
3810 if (ns->ln_flags & LF_INCOMPLETE) {
3811 lfsck_object_put(env, parent);
3816 /* Create the lost parent as an orphan. */
3817 rc = lfsck_namespace_create_orphan_dir(env, com,
3820 lfsck_object_put(env, parent);
3828 /* Add the missing name entry to the parent. */
3829 rc = lfsck_namespace_insert_normal(env, com,
3830 parent, child, cname);
3831 if (unlikely(rc == -EEXIST))
3832 /* Unfortunately, someone reused the
3833 * name under the parent by race. So we
3834 * have to remove the linkEA entry from
3835 * current child object. It means that
3836 * the LFSCK cannot recover the system
3837 * totally back to its original status,
3838 * but it is necessary to make the
3839 * current system to be consistent. */
3840 rc = lfsck_namespace_shrink_linkea(env,
3844 linkea_next_entry(&ldata);
3847 lfsck_object_put(env, parent);
3855 } /* !dt_object_exists(parent) */
3857 /* The linkEA entry with bad parent will be removed. */
3858 if (unlikely(!dt_try_as_dir(env, parent))) {
3859 lfsck_object_put(env, parent);
3860 rc = lfsck_namespace_shrink_linkea(env, com, child,
3861 &ldata, cname, pfid, true);
3871 rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
3872 (const struct dt_key *)cname->ln_name);
3873 if (rc != 0 && rc != -ENOENT) {
3874 lfsck_object_put(env, parent);
3880 if (lu_fid_eq(cfid, lfsck_dto2fid(child))) {
3881 /* It is the most common case that we
3882 * find the name entry corresponding
3883 * to the linkEA entry. */
3884 lfsck_object_put(env, parent);
3885 linkea_next_entry(&ldata);
3887 /* The name entry references another
3888 * MDT-object that may be created by
3889 * the LFSCK for repairing dangling
3890 * name entry. Try to replace it. */
3891 rc = lfsck_namespace_replace_cond(env, com,
3892 parent, child, cfid, cname);
3893 lfsck_object_put(env, parent);
3899 linkea_next_entry(&ldata);
3901 rc = lfsck_namespace_shrink_linkea(env,
3915 /* The following handles -ENOENT case */
3917 rc = dt_attr_get(env, child, la);
3921 /* If there is no name entry in the parent dir and the object
3922 * link count is fewer than the linkea entries count, then the
3923 * linkea entry should be removed. */
3924 if (ldata.ld_leh->leh_reccount > la->la_nlink) {
3925 rc = lfsck_namespace_shrink_linkea_cond(env, com,
3926 parent, child, &ldata, cname, pfid);
3927 lfsck_object_put(env, parent);
3937 /* If the LFSCK is marked as LF_INCOMPLETE, then means some
3938 * MDT has ever tried to verify some remote MDT-object that
3939 * resides on this MDT, but this MDT failed to respond such
3940 * request. So means there may be some remote name entry on
3941 * other MDT that references this object with another name,
3942 * so we cannot know whether this linkEA is valid or not.
3943 * So keep it there and maybe resolved when next LFSCK run. */
3944 if (ns->ln_flags & LF_INCOMPLETE) {
3945 lfsck_object_put(env, parent);
3950 rc = lfsck_namespace_check_name(env, lfsck, parent, child,
3956 lfsck_object_put(env, parent);
3961 /* It is an invalid name entry, drop it. */
3962 if (unlikely(rc > 0)) {
3963 lfsck_object_put(env, parent);
3964 rc = lfsck_namespace_shrink_linkea(env, com, child,
3965 &ldata, cname, pfid, true);
3975 /* Add the missing name entry back to the namespace. */
3976 rc = lfsck_namespace_insert_normal(env, com, parent, child,
3978 if (unlikely(rc == -ESTALE))
3979 /* It may happen when the remote object has been
3980 * removed, but the local MDT is not aware of that. */
3983 if (unlikely(rc == -EEXIST))
3984 /* Unfortunately, someone reused the name under the
3985 * parent by race. So we have to remove the linkEA
3986 * entry from current child object. It means that the
3987 * LFSCK cannot recover the system totally back to
3988 * its original status, but it is necessary to make
3989 * the current system to be consistent.
3991 * It also may be because of the LFSCK found some
3992 * internal status of create operation. Under such
3993 * case, nothing to be done. */
3994 rc = lfsck_namespace_shrink_linkea_cond(env, com,
3995 parent, child, &ldata, cname, pfid);
3997 linkea_next_entry(&ldata);
3999 lfsck_object_put(env, parent);
4010 if (rc < 0 && rc != -ENODATA)
4013 if (rc == 0 && ldata.ld_leh != NULL)
4014 count = ldata.ld_leh->leh_reccount;
4017 /* If the LFSCK is marked as LF_INCOMPLETE, then means some
4018 * MDT has ever tried to verify some remote MDT-object that
4019 * resides on this MDT, but this MDT failed to respond such
4020 * request. So means there may be some remote name entry on
4021 * other MDT that references this object with another name,
4022 * so we cannot know whether this linkEA is valid or not.
4023 * So keep it there and maybe resolved when next LFSCK run. */
4024 if (!(ns->ln_flags & LF_INCOMPLETE) &&
4025 (ldata.ld_leh == NULL ||
4026 !ldata.ld_leh->leh_overflow_time)) {
4027 /* If the child becomes orphan, then insert it into
4028 * the global .lustre/lost+found/MDTxxxx directory. */
4029 rc = lfsck_namespace_insert_orphan(env, com, child,
4035 ns->ln_mul_ref_repaired++;
4040 rc = dt_attr_get(env, child, la);
4044 if (la->la_nlink != 0 && la->la_nlink != count) {
4045 if (unlikely(!S_ISREG(lfsck_object_type(child)) &&
4046 !S_ISLNK(lfsck_object_type(child)))) {
4047 CDEBUG(D_LFSCK, "%s: namespace LFSCK finds "
4048 "the object "DFID"'s nlink count %d "
4049 "does not match linkEA count %d, "
4050 "type %o, skip it.\n",
4051 lfsck_lfsck2name(lfsck),
4052 PFID(lfsck_dto2fid(child)),
4053 la->la_nlink, count,
4054 lfsck_object_type(child));
4055 } else if (la->la_nlink < count &&
4056 likely(!ldata.ld_leh->leh_overflow_time)) {
4057 rc = lfsck_namespace_repair_nlink(env, com,
4060 ns->ln_objs_nlink_repaired++;
4068 if (la->la_nlink > 1)
4069 ns->ln_mul_linked_repaired++;
4075 if (!rc && flags & LNTF_CHECK_AGENT_ENTRY)
4076 rc = lfsck_namespace_check_agent_entry(env, com, child);
4081 static void lfsck_namespace_dump_statistics(struct seq_file *m,
4082 struct lfsck_namespace *ns,
4083 __u64 checked_phase1,
4084 __u64 checked_phase2,
4085 time64_t time_phase1,
4086 time64_t time_phase2, bool dryrun)
4088 const char *postfix = dryrun ? "inconsistent" : "repaired";
4090 seq_printf(m, "checked_phase1: %llu\n"
4091 "checked_phase2: %llu\n"
4094 "failed_phase1: %llu\n"
4095 "failed_phase2: %llu\n"
4096 "directories: %llu\n"
4100 "multiple_linked_checked: %llu\n"
4101 "multiple_linked_%s: %llu\n"
4102 "unknown_inconsistency: %llu\n"
4103 "unmatched_pairs_%s: %llu\n"
4104 "dangling_%s: %llu\n"
4105 "multiple_referenced_%s: %llu\n"
4106 "bad_file_type_%s: %llu\n"
4107 "lost_dirent_%s: %llu\n"
4108 "local_lost_found_scanned: %llu\n"
4109 "local_lost_found_moved: %llu\n"
4110 "local_lost_found_skipped: %llu\n"
4111 "local_lost_found_failed: %llu\n"
4112 "striped_dirs_scanned: %llu\n"
4113 "striped_dirs_%s: %llu\n"
4114 "striped_dirs_failed: %llu\n"
4115 "striped_dirs_disabled: %llu\n"
4116 "striped_dirs_skipped: %llu\n"
4117 "striped_shards_scanned: %llu\n"
4118 "striped_shards_%s: %llu\n"
4119 "striped_shards_failed: %llu\n"
4120 "striped_shards_skipped: %llu\n"
4121 "name_hash_%s: %llu\n"
4122 "linkea_overflow_%s: %llu\n"
4123 "agent_entries_%s: %llu\n"
4124 "success_count: %u\n"
4125 "run_time_phase1: %lld seconds\n"
4126 "run_time_phase2: %lld seconds\n",
4129 dryrun ? "inconsistent" : "updated",
4130 ns->ln_items_repaired,
4131 dryrun ? "inconsistent" : "updated",
4132 ns->ln_objs_repaired_phase2,
4133 ns->ln_items_failed,
4134 ns->ln_objs_failed_phase2,
4135 ns->ln_dirs_checked,
4136 postfix, ns->ln_dirent_repaired,
4137 postfix, ns->ln_linkea_repaired,
4138 postfix, ns->ln_objs_nlink_repaired,
4139 ns->ln_mul_linked_checked,
4140 postfix, ns->ln_mul_linked_repaired,
4141 ns->ln_unknown_inconsistency,
4142 postfix, ns->ln_unmatched_pairs_repaired,
4143 postfix, ns->ln_dangling_repaired,
4144 postfix, ns->ln_mul_ref_repaired,
4145 postfix, ns->ln_bad_type_repaired,
4146 postfix, ns->ln_lost_dirent_repaired,
4147 ns->ln_local_lpf_scanned,
4148 ns->ln_local_lpf_moved,
4149 ns->ln_local_lpf_skipped,
4150 ns->ln_local_lpf_failed,
4151 ns->ln_striped_dirs_scanned,
4152 postfix, ns->ln_striped_dirs_repaired,
4153 ns->ln_striped_dirs_failed,
4154 ns->ln_striped_dirs_disabled,
4155 ns->ln_striped_dirs_skipped,
4156 ns->ln_striped_shards_scanned,
4157 postfix, ns->ln_striped_shards_repaired,
4158 ns->ln_striped_shards_failed,
4159 ns->ln_striped_shards_skipped,
4160 postfix, ns->ln_name_hash_repaired,
4161 dryrun ? "inconsistent" : "cleared",
4162 ns->ln_linkea_overflow_cleared,
4163 postfix, ns->ln_agent_entries_repaired,
4164 ns->ln_success_count,
4169 static void lfsck_namespace_release_lmv(const struct lu_env *env,
4170 struct lfsck_component *com)
4172 struct lfsck_instance *lfsck = com->lc_lfsck;
4173 struct lfsck_namespace *ns = com->lc_file_ram;
4175 while (!list_empty(&lfsck->li_list_lmv)) {
4176 struct lfsck_lmv_unit *llu;
4177 struct lfsck_lmv *llmv;
4179 llu = list_entry(lfsck->li_list_lmv.next,
4180 struct lfsck_lmv_unit, llu_link);
4181 llmv = &llu->llu_lmv;
4183 LASSERTF(atomic_read(&llmv->ll_ref) == 1,
4184 "still in using: %u\n",
4185 atomic_read(&llmv->ll_ref));
4187 ns->ln_striped_dirs_skipped++;
4188 lfsck_lmv_put(env, llmv);
4192 static int lfsck_namespace_check_for_double_scan(const struct lu_env *env,
4193 struct lfsck_component *com,
4194 struct dt_object *obj)
4196 struct lu_attr *la = &lfsck_env_info(env)->lti_la;
4199 rc = dt_attr_get(env, obj, la);
4203 /* zero-linkEA object may be orphan, but it also maybe because
4204 * of upgrading. Currently, we cannot record it for double scan.
4205 * Because it may cause the LFSCK trace file to be too large. */
4207 /* "la_ctime" == 1 means that it has ever been removed from
4208 * backend /lost+found directory but not been added back to
4209 * the normal namespace yet. */
4211 if ((S_ISREG(lfsck_object_type(obj)) && la->la_nlink > 1) ||
4212 unlikely(la->la_ctime == 1))
4213 rc = lfsck_namespace_trace_update(env, com, lfsck_dto2fid(obj),
4214 LNTF_CHECK_LINKEA, true);
4219 /* namespace APIs */
4221 static int lfsck_namespace_reset(const struct lu_env *env,
4222 struct lfsck_component *com, bool init)
4224 struct lfsck_instance *lfsck = com->lc_lfsck;
4225 struct lfsck_namespace *ns = com->lc_file_ram;
4226 struct lfsck_assistant_data *lad = com->lc_data;
4227 struct dt_object *root;
4231 root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
4233 GOTO(log, rc = PTR_ERR(root));
4235 if (unlikely(!dt_try_as_dir(env, root)))
4236 GOTO(put, rc = -ENOTDIR);
4238 down_write(&com->lc_sem);
4240 memset(ns, 0, sizeof(*ns));
4242 __u32 count = ns->ln_success_count;
4243 time64_t last_time = ns->ln_time_last_complete;
4245 memset(ns, 0, sizeof(*ns));
4246 ns->ln_success_count = count;
4247 ns->ln_time_last_complete = last_time;
4249 ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
4250 ns->ln_status = LS_INIT;
4251 ns->ln_time_latest_reset = ktime_get_real_seconds();
4253 rc = lfsck_load_one_trace_file(env, com, root, &com->lc_obj,
4254 &dt_lfsck_namespace_features,
4255 LFSCK_NAMESPACE, true);
4259 rc = lfsck_load_sub_trace_files(env, com, &dt_lfsck_namespace_features,
4260 LFSCK_NAMESPACE, true);
4264 clear_bit(LAD_INCOMPLETE, &lad->lad_flags);
4265 CFS_RESET_BITMAP(lad->lad_bitmap);
4267 rc = lfsck_namespace_store(env, com);
4272 up_write(&com->lc_sem);
4275 lfsck_object_put(env, root);
4277 CDEBUG(D_LFSCK, "%s: namespace LFSCK reset: rc = %d\n",
4278 lfsck_lfsck2name(lfsck), rc);
4283 lfsck_namespace_fail(const struct lu_env *env, struct lfsck_component *com,
4286 struct lfsck_namespace *ns = com->lc_file_ram;
4288 down_write(&com->lc_sem);
4290 com->lc_new_checked++;
4291 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
4292 up_write(&com->lc_sem);
4295 static void lfsck_namespace_close_dir(const struct lu_env *env,
4296 struct lfsck_component *com)
4298 struct lfsck_namespace *ns = com->lc_file_ram;
4299 struct lfsck_assistant_data *lad = com->lc_data;
4300 struct lfsck_assistant_object *lso = NULL;
4301 struct lfsck_instance *lfsck = com->lc_lfsck;
4302 struct lfsck_lmv *llmv = lfsck->li_lmv;
4303 struct lfsck_namespace_req *lnr;
4304 struct lu_attr *la = &lfsck_env_info(env)->lti_la2;
4305 __u32 size = sizeof(*lnr) + LFSCK_TMPBUF_LEN;
4307 bool wakeup = false;
4313 rc = dt_attr_get(env, lfsck->li_obj_dir, la);
4317 OBD_ALLOC(lnr, size);
4319 ns->ln_striped_dirs_skipped++;
4324 lso = lfsck_assistant_object_init(env, lfsck_dto2fid(lfsck->li_obj_dir),
4325 la, lfsck->li_pos_current.lp_oit_cookie, true);
4327 OBD_FREE(lnr, size);
4328 ns->ln_striped_dirs_skipped++;
4333 /* Generate a dummy request to indicate that all shards' name entry
4334 * in this striped directory has been scanned for the first time. */
4335 INIT_LIST_HEAD(&lnr->lnr_lar.lar_list);
4336 lnr->lnr_lar.lar_parent = lso;
4337 lnr->lnr_lmv = lfsck_lmv_get(llmv);
4338 lnr->lnr_fid = *lfsck_dto2fid(lfsck->li_obj_dir);
4339 lnr->lnr_dir_cookie = MDS_DIR_END_OFF;
4340 lnr->lnr_size = size;
4341 lnr->lnr_type = lso->lso_attr.la_mode;
4343 spin_lock(&lad->lad_lock);
4344 if (lad->lad_assistant_status < 0 ||
4345 unlikely(!thread_is_running(&lfsck->li_thread) ||
4346 !thread_is_running(&lad->lad_thread))) {
4347 spin_unlock(&lad->lad_lock);
4348 lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar);
4349 ns->ln_striped_dirs_skipped++;
4354 list_add_tail(&lnr->lnr_lar.lar_list, &lad->lad_req_list);
4355 if (lad->lad_prefetched == 0)
4358 lad->lad_prefetched++;
4359 spin_unlock(&lad->lad_lock);
4361 wake_up_all(&lad->lad_thread.t_ctl_waitq);
4366 static int lfsck_namespace_open_dir(const struct lu_env *env,
4367 struct lfsck_component *com)
4369 struct lfsck_instance *lfsck = com->lc_lfsck;
4370 struct lfsck_namespace *ns = com->lc_file_ram;
4371 struct lfsck_lmv *llmv = lfsck->li_lmv;
4378 if (llmv->ll_lmv_master) {
4379 struct lmv_mds_md_v1 *lmv = &llmv->ll_lmv;
4381 if (lmv->lmv_master_mdt_index != lfsck_dev_idx(lfsck)) {
4382 lmv->lmv_master_mdt_index =
4383 lfsck_dev_idx(lfsck);
4384 ns->ln_flags |= LF_INCONSISTENT;
4385 llmv->ll_lmv_updated = 1;
4388 rc = lfsck_namespace_verify_stripe_slave(env, com,
4389 lfsck->li_obj_dir, llmv);
4392 RETURN(rc > 0 ? 0 : rc);
4395 static int lfsck_namespace_checkpoint(const struct lu_env *env,
4396 struct lfsck_component *com, bool init)
4398 struct lfsck_instance *lfsck = com->lc_lfsck;
4399 struct lfsck_namespace *ns = com->lc_file_ram;
4403 rc = lfsck_checkpoint_generic(env, com);
4408 down_write(&com->lc_sem);
4410 ns->ln_pos_latest_start = lfsck->li_pos_checkpoint;
4412 ns->ln_pos_last_checkpoint = lfsck->li_pos_checkpoint;
4413 ns->ln_run_time_phase1 += ktime_get_seconds() -
4414 lfsck->li_time_last_checkpoint;
4415 ns->ln_time_last_checkpoint = ktime_get_real_seconds();
4416 ns->ln_items_checked += com->lc_new_checked;
4417 com->lc_new_checked = 0;
4420 rc = lfsck_namespace_store(env, com);
4421 up_write(&com->lc_sem);
4424 CDEBUG(D_LFSCK, "%s: namespace LFSCK checkpoint at the pos [%llu"
4425 ", "DFID", %#llx], status = %d: rc = %d\n",
4426 lfsck_lfsck2name(lfsck), lfsck->li_pos_current.lp_oit_cookie,
4427 PFID(&lfsck->li_pos_current.lp_dir_parent),
4428 lfsck->li_pos_current.lp_dir_cookie, ns->ln_status, rc);
4430 return rc > 0 ? 0 : rc;
4433 static int lfsck_namespace_prep(const struct lu_env *env,
4434 struct lfsck_component *com,
4435 struct lfsck_start_param *lsp)
4437 struct lfsck_instance *lfsck = com->lc_lfsck;
4438 struct lfsck_namespace *ns = com->lc_file_ram;
4439 struct lfsck_position *pos = &com->lc_pos_start;
4442 rc = lfsck_namespace_load_bitmap(env, com);
4443 if (rc != 0 || ns->ln_status == LS_COMPLETED) {
4444 rc = lfsck_namespace_reset(env, com, false);
4446 rc = lfsck_set_param(env, lfsck, lsp->lsp_start, true);
4449 CDEBUG(D_LFSCK, "%s: namespace LFSCK prep failed: "
4450 "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
4456 down_write(&com->lc_sem);
4457 ns->ln_time_latest_start = ktime_get_real_seconds();
4458 spin_lock(&lfsck->li_lock);
4460 if (ns->ln_flags & LF_SCANNED_ONCE) {
4461 if (!lfsck->li_drop_dryrun ||
4462 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent)) {
4463 ns->ln_status = LS_SCANNING_PHASE2;
4464 list_move_tail(&com->lc_link,
4465 &lfsck->li_list_double_scan);
4466 if (!list_empty(&com->lc_link_dir))
4467 list_del_init(&com->lc_link_dir);
4468 lfsck_pos_set_zero(pos);
4470 ns->ln_status = LS_SCANNING_PHASE1;
4471 ns->ln_run_time_phase1 = 0;
4472 ns->ln_run_time_phase2 = 0;
4473 ns->ln_items_checked = 0;
4474 ns->ln_items_repaired = 0;
4475 ns->ln_items_failed = 0;
4476 ns->ln_dirs_checked = 0;
4477 ns->ln_objs_checked_phase2 = 0;
4478 ns->ln_objs_repaired_phase2 = 0;
4479 ns->ln_objs_failed_phase2 = 0;
4480 ns->ln_objs_nlink_repaired = 0;
4481 ns->ln_dirent_repaired = 0;
4482 ns->ln_linkea_repaired = 0;
4483 ns->ln_mul_linked_checked = 0;
4484 ns->ln_mul_linked_repaired = 0;
4485 ns->ln_unknown_inconsistency = 0;
4486 ns->ln_unmatched_pairs_repaired = 0;
4487 ns->ln_dangling_repaired = 0;
4488 ns->ln_mul_ref_repaired = 0;
4489 ns->ln_bad_type_repaired = 0;
4490 ns->ln_lost_dirent_repaired = 0;
4491 ns->ln_striped_dirs_scanned = 0;
4492 ns->ln_striped_dirs_repaired = 0;
4493 ns->ln_striped_dirs_failed = 0;
4494 ns->ln_striped_dirs_disabled = 0;
4495 ns->ln_striped_dirs_skipped = 0;
4496 ns->ln_striped_shards_scanned = 0;
4497 ns->ln_striped_shards_repaired = 0;
4498 ns->ln_striped_shards_failed = 0;
4499 ns->ln_striped_shards_skipped = 0;
4500 ns->ln_name_hash_repaired = 0;
4501 fid_zero(&ns->ln_fid_latest_scanned_phase2);
4502 if (list_empty(&com->lc_link_dir))
4503 list_add_tail(&com->lc_link_dir,
4504 &lfsck->li_list_dir);
4505 *pos = ns->ln_pos_first_inconsistent;
4508 ns->ln_status = LS_SCANNING_PHASE1;
4509 if (list_empty(&com->lc_link_dir))
4510 list_add_tail(&com->lc_link_dir,
4511 &lfsck->li_list_dir);
4512 if (!lfsck->li_drop_dryrun ||
4513 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent)) {
4514 *pos = ns->ln_pos_last_checkpoint;
4515 pos->lp_oit_cookie++;
4517 *pos = ns->ln_pos_first_inconsistent;
4521 spin_unlock(&lfsck->li_lock);
4522 up_write(&com->lc_sem);
4524 rc = lfsck_start_assistant(env, com, lsp);
4526 CDEBUG(D_LFSCK, "%s: namespace LFSCK prep done, start pos [%llu, "
4527 DFID", %#llx]: rc = %d\n",
4528 lfsck_lfsck2name(lfsck), pos->lp_oit_cookie,
4529 PFID(&pos->lp_dir_parent), pos->lp_dir_cookie, rc);
4534 static int lfsck_namespace_exec_oit(const struct lu_env *env,
4535 struct lfsck_component *com,
4536 struct dt_object *obj)
4538 struct lfsck_thread_info *info = lfsck_env_info(env);
4539 struct lfsck_namespace *ns = com->lc_file_ram;
4540 struct lfsck_instance *lfsck = com->lc_lfsck;
4541 const struct lu_fid *fid = lfsck_dto2fid(obj);
4542 struct lu_fid *pfid = &info->lti_fid2;
4543 struct lu_name *cname = &info->lti_name;
4544 struct lu_seq_range *range = &info->lti_range;
4545 struct seq_server_site *ss = lfsck_dev_site(lfsck);
4546 struct linkea_data ldata = { NULL };
4547 __u32 idx = lfsck_dev_idx(lfsck);
4548 struct lu_attr la = { .la_valid = 0 };
4549 bool remote = false;
4553 rc = dt_attr_get(env, obj, &la);
4554 if (unlikely(rc || (la.la_valid & LA_FLAGS &&
4555 la.la_flags & LUSTRE_ORPHAN_FL))) {
4557 "%s: skip orphan "DFID", %llx/%x: rc = %d\n",
4558 lfsck_lfsck2name(lfsck), PFID(fid),
4559 la.la_valid, la.la_flags, rc);
4564 rc = lfsck_links_read(env, obj, &ldata);
4568 /* -EINVAL means crashed linkEA, should be verified. */
4569 if (rc == -EINVAL) {
4570 rc = lfsck_namespace_trace_update(env, com, fid,
4571 LNTF_CHECK_LINKEA, true);
4573 struct lustre_handle lh = { 0 };
4575 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
4576 MDS_INODELOCK_UPDATE |
4577 MDS_INODELOCK_XATTR, LCK_EX);
4579 rc = lfsck_namespace_links_remove(env, com,
4581 lfsck_ibits_unlock(&lh, LCK_EX);
4585 GOTO(out, rc = (rc == -ENOENT ? 0 : rc));
4588 if (rc && rc != -ENODATA)
4591 if (rc == -ENODATA || unlikely(!ldata.ld_leh->leh_reccount)) {
4592 rc = lfsck_namespace_check_for_double_scan(env, com, obj);
4597 linkea_first_entry(&ldata);
4598 while (ldata.ld_lee != NULL) {
4599 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
4601 if (!fid_is_sane(pfid)) {
4602 rc = lfsck_namespace_trace_update(env, com, fid,
4603 LNTF_CHECK_PARENT, true);
4604 } else if (!linkea_entry_is_valid(&ldata, cname, pfid)) {
4607 fld_range_set_mdt(range);
4608 rc = fld_server_lookup(env, ss->ss_server_fld,
4609 fid_seq(pfid), range);
4610 if ((rc == -ENOENT) ||
4611 (!rc && range->lsr_index != idx)) {
4619 linkea_next_entry(&ldata);
4622 if ((lu_object_has_agent_entry(&obj->do_lu) && !remote) ||
4623 (!lu_object_has_agent_entry(&obj->do_lu) && remote)) {
4624 rc = lfsck_namespace_trace_update(env, com, fid,
4625 LNTF_CHECK_AGENT_ENTRY, true);
4630 /* Record multiple-linked object. */
4631 if (ldata.ld_leh->leh_reccount > 1) {
4632 rc = lfsck_namespace_trace_update(env, com, fid,
4633 LNTF_CHECK_LINKEA, true);
4639 rc = lfsck_namespace_trace_update(env, com, fid,
4640 LNTF_CHECK_LINKEA, true);
4642 rc = lfsck_namespace_check_for_double_scan(env, com, obj);
4647 down_write(&com->lc_sem);
4648 if (S_ISDIR(lfsck_object_type(obj)))
4649 ns->ln_dirs_checked++;
4651 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
4652 up_write(&com->lc_sem);
4657 static int lfsck_namespace_exec_dir(const struct lu_env *env,
4658 struct lfsck_component *com,
4659 struct lfsck_assistant_object *lso,
4660 struct lu_dirent *ent, __u16 type)
4662 struct lfsck_assistant_data *lad = com->lc_data;
4663 struct lfsck_instance *lfsck = com->lc_lfsck;
4664 struct lfsck_namespace_req *lnr;
4665 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4666 struct ptlrpc_thread *mthread = &lfsck->li_thread;
4667 struct ptlrpc_thread *athread = &lad->lad_thread;
4668 bool wakeup = false;
4670 wait_event_idle(mthread->t_ctl_waitq,
4671 lad->lad_prefetched < bk->lb_async_windows ||
4672 !thread_is_running(mthread) ||
4673 !thread_is_running(athread));
4675 if (unlikely(!thread_is_running(mthread) ||
4676 !thread_is_running(athread)))
4679 if (unlikely(lfsck_is_dead_obj(lfsck->li_obj_dir)))
4682 lnr = lfsck_namespace_assistant_req_init(com->lc_lfsck, lso, ent, type);
4684 struct lfsck_namespace *ns = com->lc_file_ram;
4686 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
4687 return PTR_ERR(lnr);
4690 spin_lock(&lad->lad_lock);
4691 if (lad->lad_assistant_status < 0 ||
4692 unlikely(!thread_is_running(mthread) ||
4693 !thread_is_running(athread))) {
4694 spin_unlock(&lad->lad_lock);
4695 lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar);
4696 return lad->lad_assistant_status;
4699 list_add_tail(&lnr->lnr_lar.lar_list, &lad->lad_req_list);
4700 if (lad->lad_prefetched == 0)
4703 lad->lad_prefetched++;
4704 spin_unlock(&lad->lad_lock);
4706 wake_up_all(&lad->lad_thread.t_ctl_waitq);
4708 down_write(&com->lc_sem);
4709 com->lc_new_checked++;
4710 up_write(&com->lc_sem);
4715 static int lfsck_namespace_post(const struct lu_env *env,
4716 struct lfsck_component *com,
4717 int result, bool init)
4719 struct lfsck_instance *lfsck = com->lc_lfsck;
4720 struct lfsck_namespace *ns = com->lc_file_ram;
4724 lfsck_post_generic(env, com, &result);
4726 down_write(&com->lc_sem);
4727 lfsck_namespace_release_lmv(env, com);
4729 spin_lock(&lfsck->li_lock);
4731 ns->ln_pos_last_checkpoint = lfsck->li_pos_checkpoint;
4733 ns->ln_status = LS_SCANNING_PHASE2;
4734 ns->ln_flags |= LF_SCANNED_ONCE;
4735 ns->ln_flags &= ~LF_UPGRADE;
4736 list_del_init(&com->lc_link_dir);
4737 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
4738 } else if (result == 0) {
4739 if (lfsck->li_status != 0)
4740 ns->ln_status = lfsck->li_status;
4742 ns->ln_status = LS_STOPPED;
4743 if (ns->ln_status != LS_PAUSED) {
4744 list_del_init(&com->lc_link_dir);
4745 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4748 ns->ln_status = LS_FAILED;
4749 list_del_init(&com->lc_link_dir);
4750 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4752 spin_unlock(&lfsck->li_lock);
4755 ns->ln_run_time_phase1 += ktime_get_seconds() -
4756 lfsck->li_time_last_checkpoint;
4757 ns->ln_time_last_checkpoint = ktime_get_real_seconds();
4758 ns->ln_items_checked += com->lc_new_checked;
4759 com->lc_new_checked = 0;
4762 rc = lfsck_namespace_store(env, com);
4763 up_write(&com->lc_sem);
4765 CDEBUG(D_LFSCK, "%s: namespace LFSCK post done: rc = %d\n",
4766 lfsck_lfsck2name(lfsck), rc);
4772 lfsck_namespace_dump(const struct lu_env *env, struct lfsck_component *com,
4775 struct lfsck_instance *lfsck = com->lc_lfsck;
4776 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4777 struct lfsck_namespace *ns = com->lc_file_ram;
4779 down_read(&com->lc_sem);
4780 seq_printf(m, "name: lfsck_namespace\n"
4786 lfsck_status2name(ns->ln_status));
4788 lfsck_bits_dump(m, ns->ln_flags, lfsck_flags_names, "flags");
4790 lfsck_bits_dump(m, bk->lb_param, lfsck_param_names, "param");
4792 lfsck_time_dump(m, ns->ln_time_last_complete, "last_completed");
4794 lfsck_time_dump(m, ns->ln_time_latest_start, "latest_start");
4796 lfsck_time_dump(m, ns->ln_time_last_checkpoint, "last_checkpoint");
4798 lfsck_pos_dump(m, &ns->ln_pos_latest_start, "latest_start_position");
4800 lfsck_pos_dump(m, &ns->ln_pos_last_checkpoint,
4801 "last_checkpoint_position");
4803 lfsck_pos_dump(m, &ns->ln_pos_first_inconsistent,
4804 "first_failure_position");
4806 if (ns->ln_status == LS_SCANNING_PHASE1) {
4807 struct lfsck_position pos;
4808 time64_t duration = ktime_get_seconds() -
4809 lfsck->li_time_last_checkpoint;
4810 u64 checked = ns->ln_items_checked + com->lc_new_checked;
4811 u64 speed = checked;
4812 u64 new_checked = com->lc_new_checked;
4813 time64_t rtime = ns->ln_run_time_phase1 + duration;
4816 new_checked = div64_s64(new_checked, duration);
4819 speed = div64_s64(speed, rtime);
4821 lfsck_namespace_dump_statistics(m, ns, checked, 0, rtime, 0,
4822 bk->lb_param & LPF_DRYRUN);
4823 seq_printf(m, "average_speed_phase1: %llu items/sec\n"
4824 "average_speed_phase2: N/A\n"
4825 "average_speed_total: %llu items/sec\n"
4826 "real_time_speed_phase1: %llu items/sec\n"
4827 "real_time_speed_phase2: N/A\n",
4832 if (likely(lfsck->li_di_oit)) {
4833 const struct dt_it_ops *iops =
4834 &lfsck->li_obj_oit->do_index_ops->dio_it;
4836 /* The low layer otable-based iteration position may NOT
4837 * exactly match the namespace-based directory traversal
4838 * cookie. Generally, it is not a serious issue. But the
4839 * caller should NOT make assumption on that. */
4840 pos.lp_oit_cookie = iops->store(env, lfsck->li_di_oit);
4841 if (!lfsck->li_current_oit_processed)
4842 pos.lp_oit_cookie--;
4844 spin_lock(&lfsck->li_lock);
4845 if (lfsck->li_di_dir) {
4846 pos.lp_dir_cookie = lfsck->li_cookie_dir;
4847 if (pos.lp_dir_cookie >= MDS_DIR_END_OFF) {
4848 fid_zero(&pos.lp_dir_parent);
4849 pos.lp_dir_cookie = 0;
4852 *lfsck_dto2fid(lfsck->li_obj_dir);
4855 fid_zero(&pos.lp_dir_parent);
4856 pos.lp_dir_cookie = 0;
4858 spin_unlock(&lfsck->li_lock);
4860 pos = ns->ln_pos_last_checkpoint;
4863 lfsck_pos_dump(m, &pos, "current_position");
4864 } else if (ns->ln_status == LS_SCANNING_PHASE2) {
4865 time64_t duration = ktime_get_seconds() -
4866 com->lc_time_last_checkpoint;
4867 __u64 checked = ns->ln_objs_checked_phase2 +
4868 com->lc_new_checked;
4869 __u64 speed1 = ns->ln_items_checked;
4870 __u64 speed2 = checked;
4871 __u64 speed0 = speed1 + speed2;
4872 __u64 new_checked = com->lc_new_checked;
4873 time64_t rtime = ns->ln_run_time_phase2 + duration;
4874 time64_t time0 = ns->ln_run_time_phase1 + rtime;
4877 new_checked = div64_s64(new_checked, duration);
4879 if (ns->ln_run_time_phase1 != 0)
4880 speed1 = div64_s64(speed1, ns->ln_run_time_phase1);
4881 else if (ns->ln_items_checked != 0)
4885 speed2 = div64_s64(speed2, rtime);
4886 else if (checked != 0)
4890 speed0 = div64_s64(speed0, time0);
4892 lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked,
4894 ns->ln_run_time_phase1, rtime,
4895 bk->lb_param & LPF_DRYRUN);
4896 seq_printf(m, "average_speed_phase1: %llu items/sec\n"
4897 "average_speed_phase2: %llu objs/sec\n"
4898 "average_speed_total: %llu items/sec\n"
4899 "real_time_speed_phase1: N/A\n"
4900 "real_time_speed_phase2: %llu objs/sec\n"
4901 "current_position: "DFID"\n",
4906 PFID(&ns->ln_fid_latest_scanned_phase2));
4908 __u64 speed1 = ns->ln_items_checked;
4909 __u64 speed2 = ns->ln_objs_checked_phase2;
4910 __u64 speed0 = speed1 + speed2;
4911 time64_t time0 = ns->ln_run_time_phase1 + ns->ln_run_time_phase2;
4913 if (ns->ln_run_time_phase1 != 0)
4914 speed1 = div64_s64(speed1, ns->ln_run_time_phase1);
4915 else if (ns->ln_items_checked != 0)
4918 if (ns->ln_run_time_phase2 != 0)
4919 speed2 = div64_s64(speed2, ns->ln_run_time_phase2);
4920 else if (ns->ln_objs_checked_phase2 != 0)
4924 speed0 = div64_s64(speed0, time0);
4926 lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked,
4927 ns->ln_objs_checked_phase2,
4928 ns->ln_run_time_phase1,
4929 ns->ln_run_time_phase2,
4930 bk->lb_param & LPF_DRYRUN);
4931 seq_printf(m, "average_speed_phase1: %llu items/sec\n"
4932 "average_speed_phase2: %llu objs/sec\n"
4933 "average_speed_total: %llu items/sec\n"
4934 "real_time_speed_phase1: N/A\n"
4935 "real_time_speed_phase2: N/A\n"
4936 "current_position: N/A\n",
4942 up_read(&com->lc_sem);
4945 static int lfsck_namespace_double_scan(const struct lu_env *env,
4946 struct lfsck_component *com)
4948 struct lfsck_namespace *ns = com->lc_file_ram;
4949 struct lfsck_assistant_data *lad = com->lc_data;
4950 struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
4951 struct lfsck_tgt_desc *ltd;
4952 struct lfsck_tgt_desc *next;
4955 rc = lfsck_double_scan_generic(env, com, ns->ln_status);
4956 if (thread_is_stopped(&lad->lad_thread)) {
4957 LASSERT(list_empty(&lad->lad_req_list));
4958 LASSERT(list_empty(&lad->lad_mdt_phase1_list));
4960 spin_lock(<ds->ltd_lock);
4961 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4962 ltd_namespace_phase_list) {
4963 list_del_init(<d->ltd_namespace_phase_list);
4965 spin_unlock(<ds->ltd_lock);
4971 static void lfsck_namespace_data_release(const struct lu_env *env,
4972 struct lfsck_component *com)
4974 struct lfsck_assistant_data *lad = com->lc_data;
4975 struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
4976 struct lfsck_tgt_desc *ltd;
4977 struct lfsck_tgt_desc *next;
4979 LASSERT(lad != NULL);
4980 LASSERT(thread_is_init(&lad->lad_thread) ||
4981 thread_is_stopped(&lad->lad_thread));
4982 LASSERT(list_empty(&lad->lad_req_list));
4984 com->lc_data = NULL;
4985 lfsck_namespace_release_lmv(env, com);
4987 spin_lock(<ds->ltd_lock);
4988 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
4989 ltd_namespace_phase_list) {
4990 list_del_init(<d->ltd_namespace_phase_list);
4992 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4993 ltd_namespace_phase_list) {
4994 list_del_init(<d->ltd_namespace_phase_list);
4996 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_list,
4997 ltd_namespace_list) {
4998 list_del_init(<d->ltd_namespace_list);
5000 spin_unlock(<ds->ltd_lock);
5002 if (likely(lad->lad_bitmap != NULL))
5003 CFS_FREE_BITMAP(lad->lad_bitmap);
5008 static void lfsck_namespace_quit(const struct lu_env *env,
5009 struct lfsck_component *com)
5011 struct lfsck_assistant_data *lad = com->lc_data;
5012 struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
5013 struct lfsck_tgt_desc *ltd;
5014 struct lfsck_tgt_desc *next;
5016 LASSERT(lad != NULL);
5018 lfsck_quit_generic(env, com);
5020 LASSERT(thread_is_init(&lad->lad_thread) ||
5021 thread_is_stopped(&lad->lad_thread));
5022 LASSERT(list_empty(&lad->lad_req_list));
5024 lfsck_namespace_release_lmv(env, com);
5026 spin_lock(<ds->ltd_lock);
5027 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
5028 ltd_namespace_phase_list) {
5029 list_del_init(<d->ltd_namespace_phase_list);
5031 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
5032 ltd_namespace_phase_list) {
5033 list_del_init(<d->ltd_namespace_phase_list);
5035 spin_unlock(<ds->ltd_lock);
5038 static int lfsck_namespace_in_notify(const struct lu_env *env,
5039 struct lfsck_component *com,
5040 struct lfsck_request *lr)
5042 struct lfsck_instance *lfsck = com->lc_lfsck;
5043 struct lfsck_namespace *ns = com->lc_file_ram;
5044 struct lfsck_assistant_data *lad = com->lc_data;
5045 struct lfsck_tgt_descs *ltds = &lfsck->li_mdt_descs;
5046 struct lfsck_tgt_desc *ltd;
5051 switch (lr->lr_event) {
5052 case LE_SET_LMV_MASTER: {
5053 struct dt_object *obj;
5055 obj = lfsck_object_find_bottom(env, lfsck, &lr->lr_fid);
5057 RETURN(PTR_ERR(obj));
5059 if (likely(dt_object_exists(obj)))
5060 rc = lfsck_namespace_notify_lmv_master_local(env, com,
5063 lfsck_object_put(env, obj);
5065 RETURN(rc > 0 ? 0 : rc);
5067 case LE_SET_LMV_SLAVE: {
5068 if (!(lr->lr_flags & LEF_RECHECK_NAME_HASH))
5069 ns->ln_striped_shards_repaired++;
5071 rc = lfsck_namespace_trace_update(env, com, &lr->lr_fid,
5072 LNTF_RECHECK_NAME_HASH, true);
5074 RETURN(rc > 0 ? 0 : rc);
5076 case LE_PHASE1_DONE:
5077 case LE_PHASE2_DONE:
5084 CDEBUG(D_LFSCK, "%s: namespace LFSCK handles notify %u from MDT %x, "
5085 "status %d, flags %x\n", lfsck_lfsck2name(lfsck), lr->lr_event,
5086 lr->lr_index, lr->lr_status, lr->lr_flags2);
5088 spin_lock(<ds->ltd_lock);
5089 ltd = lfsck_ltd2tgt(ltds, lr->lr_index);
5091 spin_unlock(<ds->ltd_lock);
5096 list_del_init(<d->ltd_namespace_phase_list);
5097 switch (lr->lr_event) {
5098 case LE_PHASE1_DONE:
5099 if (lr->lr_status <= 0) {
5100 ltd->ltd_namespace_done = 1;
5101 list_del_init(<d->ltd_namespace_list);
5102 CDEBUG(D_LFSCK, "%s: MDT %x failed/stopped at "
5103 "phase1 for namespace LFSCK: rc = %d.\n",
5104 lfsck_lfsck2name(lfsck),
5105 ltd->ltd_index, lr->lr_status);
5106 ns->ln_flags |= LF_INCOMPLETE;
5111 if (lr->lr_flags2 & LF_INCOMPLETE)
5112 ns->ln_flags |= LF_INCOMPLETE;
5114 if (list_empty(<d->ltd_namespace_list))
5115 list_add_tail(<d->ltd_namespace_list,
5116 &lad->lad_mdt_list);
5117 list_add_tail(<d->ltd_namespace_phase_list,
5118 &lad->lad_mdt_phase2_list);
5120 case LE_PHASE2_DONE:
5121 ltd->ltd_namespace_done = 1;
5122 list_del_init(<d->ltd_namespace_list);
5126 ltd->ltd_namespace_done = 1;
5127 list_del_init(<d->ltd_namespace_list);
5128 if (!(lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT)) {
5130 "%s: the peer MDT %x exit namespace LFSCK\n",
5131 lfsck_lfsck2name(lfsck), ltd->ltd_index);
5132 ns->ln_flags |= LF_INCOMPLETE;
5138 spin_unlock(<ds->ltd_lock);
5140 if (fail && lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) {
5141 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
5143 memset(stop, 0, sizeof(*stop));
5144 stop->ls_status = lr->lr_status;
5145 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
5146 lfsck_stop(env, lfsck->li_bottom, stop);
5147 } else if (lfsck_phase2_next_ready(lad)) {
5148 wake_up_all(&lad->lad_thread.t_ctl_waitq);
5154 static void lfsck_namespace_repaired(struct lfsck_namespace *ns, __u64 *count)
5156 *count += ns->ln_objs_nlink_repaired;
5157 *count += ns->ln_dirent_repaired;
5158 *count += ns->ln_linkea_repaired;
5159 *count += ns->ln_mul_linked_repaired;
5160 *count += ns->ln_unmatched_pairs_repaired;
5161 *count += ns->ln_dangling_repaired;
5162 *count += ns->ln_mul_ref_repaired;
5163 *count += ns->ln_bad_type_repaired;
5164 *count += ns->ln_lost_dirent_repaired;
5165 *count += ns->ln_striped_dirs_disabled;
5166 *count += ns->ln_striped_dirs_repaired;
5167 *count += ns->ln_striped_shards_repaired;
5168 *count += ns->ln_name_hash_repaired;
5169 *count += ns->ln_local_lpf_moved;
5172 static int lfsck_namespace_query_all(const struct lu_env *env,
5173 struct lfsck_component *com,
5174 __u32 *mdts_count, __u64 *repaired)
5176 struct lfsck_namespace *ns = com->lc_file_ram;
5177 struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
5178 struct lfsck_tgt_desc *ltd;
5183 rc = lfsck_query_all(env, com);
5187 down_read(<ds->ltd_rw_sem);
5188 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
5189 ltd = lfsck_ltd2tgt(ltds, idx);
5190 LASSERT(ltd != NULL);
5192 mdts_count[ltd->ltd_namespace_status]++;
5193 *repaired += ltd->ltd_namespace_repaired;
5195 up_read(<ds->ltd_rw_sem);
5197 down_read(&com->lc_sem);
5198 mdts_count[ns->ln_status]++;
5199 lfsck_namespace_repaired(ns, repaired);
5200 up_read(&com->lc_sem);
5205 static int lfsck_namespace_query(const struct lu_env *env,
5206 struct lfsck_component *com,
5207 struct lfsck_request *req,
5208 struct lfsck_reply *rep,
5209 struct lfsck_query *que, int idx)
5211 struct lfsck_namespace *ns = com->lc_file_ram;
5215 LASSERT(com->lc_lfsck->li_master);
5217 rc = lfsck_namespace_query_all(env, com,
5218 que->lu_mdts_count[idx],
5219 &que->lu_repaired[idx]);
5221 down_read(&com->lc_sem);
5222 rep->lr_status = ns->ln_status;
5223 if (req->lr_flags & LEF_QUERY_ALL)
5224 lfsck_namespace_repaired(ns, &rep->lr_repaired);
5225 up_read(&com->lc_sem);
5231 static struct lfsck_operations lfsck_namespace_ops = {
5232 .lfsck_reset = lfsck_namespace_reset,
5233 .lfsck_fail = lfsck_namespace_fail,
5234 .lfsck_close_dir = lfsck_namespace_close_dir,
5235 .lfsck_open_dir = lfsck_namespace_open_dir,
5236 .lfsck_checkpoint = lfsck_namespace_checkpoint,
5237 .lfsck_prep = lfsck_namespace_prep,
5238 .lfsck_exec_oit = lfsck_namespace_exec_oit,
5239 .lfsck_exec_dir = lfsck_namespace_exec_dir,
5240 .lfsck_post = lfsck_namespace_post,
5241 .lfsck_dump = lfsck_namespace_dump,
5242 .lfsck_double_scan = lfsck_namespace_double_scan,
5243 .lfsck_data_release = lfsck_namespace_data_release,
5244 .lfsck_quit = lfsck_namespace_quit,
5245 .lfsck_in_notify = lfsck_namespace_in_notify,
5246 .lfsck_query = lfsck_namespace_query,
5250 * Repair dangling name entry.
5252 * For the name entry with dangling reference, we need to repare the
5253 * inconsistency according to the LFSCK sponsor's requirement:
5255 * 1) Keep the inconsistency there and report the inconsistency case,
5256 * then give the chance to the application to find related issues,
5257 * and the users can make the decision about how to handle it with
5258 * more human knownledge. (by default)
5260 * 2) Re-create the missing MDT-object with the FID information.
5262 * \param[in] env pointer to the thread context
5263 * \param[in] com pointer to the lfsck component
5264 * \param[in] parent pointer to the dir object that contains the dangling
5266 * \param[in] child pointer to the object corresponding to the dangling
5268 * \param[in] lnr pointer to the namespace request that contains the
5269 * name's name, parent object, parent's LMV, and ect.
5271 * \retval positive number if no need to repair
5272 * \retval zero for repaired successfully
5273 * \retval negative error number on failure
5275 int lfsck_namespace_repair_dangling(const struct lu_env *env,
5276 struct lfsck_component *com,
5277 struct dt_object *parent,
5278 struct dt_object *child,
5279 struct lfsck_namespace_req *lnr)
5281 struct lfsck_thread_info *info = lfsck_env_info(env);
5282 struct lu_attr *la = &info->lti_la;
5283 struct dt_allocation_hint *hint = &info->lti_hint;
5284 struct dt_object_format *dof = &info->lti_dof;
5285 struct dt_insert_rec *rec = &info->lti_dt_rec;
5286 struct lmv_mds_md_v1 *lmv2 = &info->lti_lmv2;
5287 const struct lu_name *cname;
5288 const struct lu_fid *pfid = lfsck_dto2fid(parent);
5289 const struct lu_fid *cfid = lfsck_dto2fid(child);
5290 struct linkea_data ldata = { NULL };
5291 struct lfsck_lock_handle *llh = &info->lti_llh;
5292 struct lustre_handle rlh = { 0 };
5293 struct lustre_handle clh = { 0 };
5294 struct lu_buf linkea_buf;
5295 struct lu_buf lmv_buf;
5296 struct lfsck_instance *lfsck = com->lc_lfsck;
5297 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5298 struct dt_device *dev = lfsck->li_next;
5299 struct thandle *th = NULL;
5301 __u16 type = lnr->lnr_type;
5305 cname = lfsck_name_get_const(env, lnr->lnr_name, lnr->lnr_namelen);
5306 if (bk->lb_param & LPF_CREATE_MDTOBJ)
5311 if (!create || bk->lb_param & LPF_DRYRUN)
5314 /* We may need to create the sub-objects of the @child via LOD,
5315 * so make the modification based on lfsck->li_next. */
5317 parent = lfsck_object_locate(dev, parent);
5319 GOTO(log, rc = PTR_ERR(parent));
5321 if (unlikely(!dt_try_as_dir(env, parent)))
5322 GOTO(log, rc = -ENOTDIR);
5324 child = lfsck_object_locate(dev, child);
5326 GOTO(log, rc = PTR_ERR(child));
5328 rc = linkea_links_new(&ldata, &info->lti_linkea_buf2,
5333 rc = lfsck_lock(env, lfsck, parent, lnr->lnr_name, llh,
5334 MDS_INODELOCK_UPDATE, LCK_PW);
5338 rc = lfsck_namespace_check_exist(env, parent, child, lnr->lnr_name);
5342 if (dt_object_remote(child)) {
5343 rc = lfsck_remote_lookup_lock(env, lfsck, parent, child, &rlh,
5349 rc = lfsck_ibits_lock(env, lfsck, child, &clh,
5350 MDS_INODELOCK_UPDATE | MDS_INODELOCK_LOOKUP |
5351 MDS_INODELOCK_XATTR, LCK_EX);
5353 GOTO(unlock_remote_lookup, rc);
5355 /* Set the ctime as zero, then others can know it is created for
5356 * repairing dangling name entry by LFSCK. And if the LFSCK made
5357 * wrong decision and the real MDT-object has been found later,
5358 * then the LFSCK has chance to fix the incosistency properly. */
5359 memset(la, 0, sizeof(*la));
5361 la->la_mode = (type & S_IFMT) | 0700;
5363 la->la_mode = (type & S_IFMT) | 0600;
5364 la->la_valid = LA_TYPE | LA_MODE | LA_CTIME;
5367 * if it's directory, skip do_ah_init() to create a plain directory
5368 * because it may have shards already, which will be inserted back
5369 * later, besides, it may be remote, and creating stripe directory
5370 * remotely is not supported.
5373 child->do_ops->do_ah_init(env, hint, parent, child, type);
5375 memset(dof, 0, sizeof(*dof));
5376 dof->dof_type = dt_mode_to_dft(type);
5377 /* If the target is a regular file, then the LFSCK will only create
5378 * the MDT-object without stripes (dof->dof_reg.striped = 0). related
5379 * OST-objects will be created when write open. */
5381 th = dt_trans_create(env, dev);
5383 GOTO(unlock_child, rc = PTR_ERR(th));
5385 /* 1a. create child. */
5386 rc = dt_declare_create(env, child, la, hint, dof, th);
5390 if (S_ISDIR(type)) {
5391 if (unlikely(!dt_try_as_dir(env, child)))
5392 GOTO(stop, rc = -ENOTDIR);
5394 /* 2a. increase child nlink */
5395 rc = dt_declare_ref_add(env, child, th);
5399 /* 3a. insert dot into child dir */
5400 rec->rec_type = S_IFDIR;
5401 rec->rec_fid = cfid;
5402 rc = dt_declare_insert(env, child,
5403 (const struct dt_rec *)rec,
5404 (const struct dt_key *)dot, th);
5408 /* 4a. insert dotdot into child dir */
5409 rec->rec_fid = pfid;
5410 rc = dt_declare_insert(env, child,
5411 (const struct dt_rec *)rec,
5412 (const struct dt_key *)dotdot, th);
5416 /* 5a. generate slave LMV EA. */
5417 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
5420 idx = lfsck_shard_name_to_index(env,
5421 lnr->lnr_name, lnr->lnr_namelen,
5423 if (unlikely(idx < 0))
5424 GOTO(stop, rc = idx);
5426 *lmv2 = lnr->lnr_lmv->ll_lmv;
5427 lmv2->lmv_magic = LMV_MAGIC_STRIPE;
5428 lmv2->lmv_master_mdt_index = idx;
5430 lfsck_lmv_header_cpu_to_le(lmv2, lmv2);
5431 lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
5432 rc = dt_declare_xattr_set(env, child, &lmv_buf,
5433 XATTR_NAME_LMV".set", 0, th);
5439 /* 6a. insert linkEA for child */
5440 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
5441 ldata.ld_leh->leh_len);
5442 rc = dt_declare_xattr_set(env, child, &linkea_buf,
5443 XATTR_NAME_LINK, 0, th);
5447 /* 7a. if child is remote, delete and insert to generate local agent */
5448 if (dt_object_remote(child)) {
5449 rc = dt_declare_delete(env, parent,
5450 (const struct dt_key *)lnr->lnr_name,
5455 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
5456 (const struct dt_key *)lnr->lnr_name,
5462 rc = dt_trans_start_local(env, dev, th);
5464 GOTO(stop, rc = (rc == -EEXIST ? 1 : rc));
5466 dt_write_lock(env, child, 0);
5467 /* 1b. create child */
5468 rc = dt_create(env, child, la, hint, dof, th);
5470 GOTO(unlock, rc = (rc == -EEXIST ? 1 : rc));
5472 if (S_ISDIR(type)) {
5473 /* 2b. increase child nlink */
5474 rc = dt_ref_add(env, child, th);
5478 /* 3b. insert dot into child dir */
5479 rec->rec_type = S_IFDIR;
5480 rec->rec_fid = cfid;
5481 rc = dt_insert(env, child, (const struct dt_rec *)rec,
5482 (const struct dt_key *)dot, th);
5486 /* 4b. insert dotdot into child dir */
5487 rec->rec_fid = pfid;
5488 rc = dt_insert(env, child, (const struct dt_rec *)rec,
5489 (const struct dt_key *)dotdot, th);
5493 /* 5b. generate slave LMV EA. */
5494 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
5495 rc = dt_xattr_set(env, child, &lmv_buf,
5496 XATTR_NAME_LMV".set", 0, th);
5502 /* 6b. insert linkEA for child. */
5503 rc = dt_xattr_set(env, child, &linkea_buf,
5504 XATTR_NAME_LINK, 0, th);
5508 /* 7b. if child is remote, delete and insert to generate local agent */
5509 if (dt_object_remote(child)) {
5510 rc = dt_delete(env, parent,
5511 (const struct dt_key *)lnr->lnr_name, th);
5515 rec->rec_type = type;
5516 rec->rec_fid = cfid;
5517 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
5518 (const struct dt_key *)lnr->lnr_name, th);
5526 dt_write_unlock(env, child);
5529 dt_trans_stop(env, dev, th);
5532 lfsck_ibits_unlock(&clh, LCK_EX);
5533 unlock_remote_lookup:
5534 if (dt_object_remote(child))
5535 lfsck_ibits_unlock(&rlh, LCK_EX);
5538 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found dangling "
5539 "reference for: parent "DFID", child "DFID", type %u, "
5540 "name %s. %s: rc = %d\n", lfsck_lfsck2name(lfsck),
5541 PFID(pfid), PFID(cfid), type, cname->ln_name,
5542 create ? "Create the lost MDT-object as required" :
5543 "Keep the MDT-object there by default", rc);
5546 struct lfsck_namespace *ns = com->lc_file_ram;
5548 ns->ln_flags |= LF_INCONSISTENT;
5554 static int lfsck_namespace_assistant_handler_p1(const struct lu_env *env,
5555 struct lfsck_component *com,
5556 struct lfsck_assistant_req *lar)
5558 struct lfsck_thread_info *info = lfsck_env_info(env);
5559 struct lu_attr *la = &info->lti_la;
5560 struct lfsck_instance *lfsck = com->lc_lfsck;
5561 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5562 struct lfsck_namespace *ns = com->lc_file_ram;
5563 struct lfsck_assistant_data *lad = com->lc_data;
5564 struct linkea_data ldata = { NULL };
5565 const struct lu_name *cname;
5566 struct thandle *handle = NULL;
5567 struct lfsck_namespace_req *lnr =
5568 container_of0(lar, struct lfsck_namespace_req, lnr_lar);
5569 struct dt_object *dir = NULL;
5570 struct dt_object *obj = NULL;
5571 struct lfsck_assistant_object *lso = lar->lar_parent;
5572 const struct lu_fid *pfid = &lso->lso_fid;
5573 struct dt_device *dev = NULL;
5574 struct lustre_handle lh = { 0 };
5575 bool repaired = false;
5576 bool dtlocked = false;
5580 bool bad_hash = false;
5581 bool bad_linkea = false;
5585 enum lfsck_namespace_inconsistency_type type = LNIT_NONE;
5592 if (lnr->lnr_attr & LUDA_UPGRADE) {
5593 ns->ln_flags |= LF_UPGRADE;
5594 ns->ln_dirent_repaired++;
5596 } else if (lnr->lnr_attr & LUDA_REPAIR) {
5597 ns->ln_flags |= LF_INCONSISTENT;
5598 ns->ln_dirent_repaired++;
5602 if (unlikely(fid_is_zero(&lnr->lnr_fid))) {
5603 if (strcmp(lnr->lnr_name, dotdot) != 0)
5606 rc = lfsck_namespace_trace_update(env, com, pfid,
5607 LNTF_CHECK_PARENT, true);
5612 if (unlikely(!fid_is_sane(&lnr->lnr_fid))) {
5613 CDEBUG(D_LFSCK, "%s: dir scan find invalid FID "DFID
5614 " for the name entry %.*s under "DFID"\n",
5615 lfsck_lfsck2name(lfsck), PFID(&lnr->lnr_fid),
5616 lnr->lnr_namelen, lnr->lnr_name, PFID(pfid));
5618 if (strcmp(lnr->lnr_name, dotdot) != 0)
5619 /* invalid FID means bad name entry, remove it. */
5620 type = LNIT_BAD_DIRENT;
5622 /* If the parent FID is invalid, we cannot remove
5623 * the ".." entry directly. */
5624 rc = lfsck_namespace_trace_update(env, com, pfid,
5625 LNTF_CHECK_PARENT, true);
5630 if (unlikely(lnr->lnr_dir_cookie == MDS_DIR_END_OFF)) {
5631 rc = lfsck_namespace_striped_dir_rescan(env, com, lnr);
5636 if (fid_seq_is_dot(fid_seq(&lnr->lnr_fid)))
5639 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
5640 rc = lfsck_namespace_handle_striped_master(env, com, lnr);
5645 idx = lfsck_find_mdt_idx_by_fid(env, lfsck, &lnr->lnr_fid);
5647 GOTO(out, rc = idx);
5649 if (idx == lfsck_dev_idx(lfsck)) {
5650 if (unlikely(strcmp(lnr->lnr_name, dotdot) == 0))
5653 dev = lfsck->li_bottom;
5655 struct lfsck_tgt_desc *ltd;
5657 /* Usually, some local filesystem consistency verification
5658 * tools can guarantee the local namespace tree consistenct.
5659 * So the LFSCK will only verify the remote directory. */
5660 if (unlikely(strcmp(lnr->lnr_name, dotdot) == 0)) {
5661 rc = lfsck_namespace_trace_update(env, com, pfid,
5662 LNTF_CHECK_PARENT, true);
5667 ltd = lfsck_ltd2tgt(&lfsck->li_mdt_descs, idx);
5668 if (unlikely(ltd == NULL)) {
5669 CDEBUG(D_LFSCK, "%s: cannot talk with MDT %x which "
5670 "did not join the namespace LFSCK\n",
5671 lfsck_lfsck2name(lfsck), idx);
5672 lfsck_lad_set_bitmap(env, com, idx);
5674 GOTO(out, rc = -ENODEV);
5680 obj = lfsck_object_find_by_dev(env, dev, &lnr->lnr_fid);
5682 GOTO(out, rc = PTR_ERR(obj));
5684 cname = lfsck_name_get_const(env, lnr->lnr_name, lnr->lnr_namelen);
5685 if (dt_object_exists(obj) == 0) {
5689 dir = lfsck_assistant_object_load(env, lfsck, lso);
5693 GOTO(trace, rc == -ENOENT ? 0 : rc);
5697 rc = lfsck_namespace_check_exist(env, dir, obj, lnr->lnr_name);
5699 if (!lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5700 lnr->lnr_name, lnr->lnr_namelen)) {
5701 type = LNIT_BAD_DIRENT;
5706 type = LNIT_DANGLING;
5707 rc = lfsck_namespace_repair_dangling(env, com, dir,
5716 if (!(bk->lb_param & LPF_DRYRUN) && lad->lad_advance_lock) {
5719 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
5720 MDS_INODELOCK_UPDATE |
5721 MDS_INODELOCK_XATTR, LCK_EX);
5725 handle = dt_trans_create(env, dev);
5727 GOTO(out, rc = PTR_ERR(handle));
5729 rc = lfsck_declare_namespace_exec_dir(env, obj, handle);
5733 rc = dt_trans_start_local(env, dev, handle);
5737 dt_write_lock(env, obj, 0);
5741 rc = lfsck_links_read(env, obj, &ldata);
5742 if (unlikely(rc == -ENOENT)) {
5743 if (handle != NULL) {
5744 dt_write_unlock(env, obj);
5747 dt_trans_stop(env, dev, handle);
5750 lfsck_ibits_unlock(&lh, LCK_EX);
5753 /* It may happen when the remote object has been removed,
5754 * but the local MDT is not aware of that. */
5756 } else if (rc == 0) {
5757 count = ldata.ld_leh->leh_reccount;
5758 rc = linkea_links_find(&ldata, cname, pfid);
5760 (count == 1 || !S_ISDIR(lfsck_object_type(obj)))) {
5761 if ((lfsck_object_type(obj) & S_IFMT) !=
5763 ns->ln_flags |= LF_INCONSISTENT;
5764 type = LNIT_BAD_TYPE;
5770 /* If the name entry hash does not match the slave striped
5771 * directory, and the name entry does not match also, then
5772 * it is quite possible that name entry is corrupted. */
5773 if (!lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5774 lnr->lnr_name, lnr->lnr_namelen)) {
5775 ns->ln_flags |= LF_INCONSISTENT;
5776 type = LNIT_BAD_DIRENT;
5781 /* If the file type stored in the name entry does not match
5782 * the file type claimed by the object, and the object does
5783 * not recognize the name entry, then it is quite possible
5784 * that the name entry is corrupted. */
5785 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type) {
5786 ns->ln_flags |= LF_INCONSISTENT;
5787 type = LNIT_BAD_DIRENT;
5792 /* For sub-dir object, we cannot make sure whether the sub-dir
5793 * back references the parent via ".." name entry correctly or
5794 * not in the LFSCK first-stage scanning. It may be that the
5795 * (remote) sub-dir ".." name entry has no parent FID after
5796 * file-level backup/restore and its linkEA may be wrong.
5797 * So under such case, we should replace the linkEA according
5798 * to current name entry. But this needs to be done during the
5799 * LFSCK second-stage scanning. The LFSCK will record the name
5800 * entry for further possible using. */
5804 } else if (unlikely(rc == -EINVAL)) {
5805 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type)
5806 type = LNIT_BAD_TYPE;
5809 /* The magic crashed, we are not sure whether there are more
5810 * corrupt data in the linkea, so remove all linkea entries. */
5814 } else if (rc == -ENODATA) {
5815 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type)
5816 type = LNIT_BAD_TYPE;
5823 if (bk->lb_param & LPF_DRYRUN) {
5825 ns->ln_flags |= LF_UPGRADE;
5827 ns->ln_flags |= LF_INCONSISTENT;
5828 ns->ln_linkea_repaired++;
5834 if (!lustre_handle_is_used(&lh)) {
5842 LASSERT(handle != NULL);
5845 dir = lfsck_assistant_object_load(env, lfsck, lso);
5849 GOTO(stop, rc == -ENOENT ? 0 : rc);
5853 rc = lfsck_namespace_check_exist(env, dir, obj, lnr->lnr_name);
5858 if (!remove && newdata)
5859 ns->ln_flags |= LF_UPGRADE;
5860 else if (remove || !(ns->ln_flags & LF_UPGRADE))
5861 ns->ln_flags |= LF_INCONSISTENT;
5866 rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, handle);
5867 if (rc != 0 && rc != -ENOENT && rc != -ENODATA)
5872 rc = linkea_data_new(&ldata,
5873 &lfsck_env_info(env)->lti_linkea_buf);
5878 rc = linkea_add_buf(&ldata, cname, pfid);
5880 rc = lfsck_links_write(env, obj, &ldata, handle);
5884 count = ldata.ld_leh->leh_reccount;
5885 if (!S_ISDIR(lfsck_object_type(obj)) ||
5886 !dt_object_remote(obj)) {
5887 ns->ln_linkea_repaired++;
5897 dt_write_unlock(env, obj);
5899 if (handle != NULL && !IS_ERR(handle))
5900 dt_trans_stop(env, dev, handle);
5903 lfsck_ibits_unlock(&lh, LCK_EX);
5905 if (!name_is_dot_or_dotdot(lnr->lnr_name, lnr->lnr_namelen) &&
5906 !lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5907 lnr->lnr_name, lnr->lnr_namelen) &&
5908 type != LNIT_BAD_DIRENT) {
5909 ns->ln_flags |= LF_INCONSISTENT;
5913 dir = lfsck_assistant_object_load(env, lfsck, lso);
5917 GOTO(trace, rc == -ENOENT ? 0 : rc);
5921 rc = lfsck_namespace_repair_bad_name_hash(env, com, dir,
5922 lnr->lnr_lmv, lnr->lnr_name);
5928 if (type != LNIT_NONE && dir == NULL) {
5929 dir = lfsck_assistant_object_load(env, lfsck, lso);
5933 GOTO(trace, rc == -ENOENT ? 0 : rc);
5940 rc = lfsck_namespace_repair_dirent(env, com, dir,
5941 obj, lnr->lnr_name, lnr->lnr_name,
5942 lnr->lnr_type, true, false);
5946 case LNIT_BAD_DIRENT:
5948 /* XXX: This is a bad dirent, we do not know whether
5949 * the original name entry reference a regular
5950 * file or a directory, then keep the parent's
5951 * nlink count unchanged here. */
5952 rc = lfsck_namespace_repair_dirent(env, com, dir,
5953 obj, lnr->lnr_name, lnr->lnr_name,
5954 lnr->lnr_type, false, false);
5962 if (obj != NULL && count == 1 &&
5963 S_ISREG(lfsck_object_type(obj)))
5964 dt_attr_get(env, obj, la);
5968 down_write(&com->lc_sem);
5970 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail to handle "
5971 "the entry: "DFID", parent "DFID", name %.*s: rc = %d\n",
5972 lfsck_lfsck2name(lfsck), PFID(&lnr->lnr_fid), PFID(pfid),
5973 lnr->lnr_namelen, lnr->lnr_name, rc);
5975 lfsck_namespace_record_failure(env, lfsck, ns);
5976 if ((rc == -ENOTCONN || rc == -ESHUTDOWN || rc == -EREMCHG ||
5977 rc == -ETIMEDOUT || rc == -EHOSTDOWN ||
5978 rc == -EHOSTUNREACH || rc == -EINPROGRESS) &&
5979 dev != NULL && dev != lfsck->li_bottom)
5980 lfsck_lad_set_bitmap(env, com, idx);
5982 if (!(bk->lb_param & LPF_FAILOUT))
5986 ns->ln_items_repaired++;
5988 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant "
5989 "repaired the entry: "DFID", parent "DFID
5990 ", name %.*s, type %d\n",
5991 lfsck_lfsck2name(lfsck),
5992 PFID(&lnr->lnr_fid), PFID(pfid),
5993 lnr->lnr_namelen, lnr->lnr_name, type);
5997 ns->ln_dangling_repaired++;
6000 ns->ln_bad_type_repaired++;
6002 case LNIT_BAD_DIRENT:
6003 ns->ln_dirent_repaired++;
6009 if (bk->lb_param & LPF_DRYRUN &&
6010 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent))
6011 lfsck_pos_fill(env, lfsck,
6012 &ns->ln_pos_first_inconsistent,
6017 ns->ln_name_hash_repaired++;
6019 /* Not count repeatedly. */
6021 ns->ln_items_repaired++;
6023 CDEBUG(D_LFSCK, "%s: namespace LFSCK "
6024 "assistant repaired the entry: "
6027 lfsck_lfsck2name(lfsck),
6028 PFID(&lnr->lnr_fid), PFID(pfid),
6029 lnr->lnr_namelen, lnr->lnr_name);
6032 if (bk->lb_param & LPF_DRYRUN &&
6033 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent))
6034 lfsck_pos_fill(env, lfsck,
6035 &ns->ln_pos_first_inconsistent,
6042 if (count > 1 || la->la_nlink > 1)
6043 ns->ln_mul_linked_checked++;
6045 up_write(&com->lc_sem);
6047 if (obj != NULL && !IS_ERR(obj))
6048 lfsck_object_put(env, obj);
6050 if (dir != NULL && !IS_ERR(dir))
6051 lfsck_object_put(env, dir);
6053 lad->lad_advance_lock = bad_linkea;
6059 * Handle one orphan under the backend /lost+found directory
6061 * Insert the orphan FID into the namespace LFSCK trace file for further
6062 * processing (via the subsequent namespace LFSCK second-stage scanning).
6063 * At the same time, remove the orphan name entry from backend /lost+found
6064 * directory. There is an interval between the orphan name entry removed
6065 * from the backend /lost+found directory and the orphan FID in the LFSCK
6066 * trace file handled. In such interval, the LFSCK can be reset, then
6067 * all the FIDs recorded in the namespace LFSCK trace file will be dropped.
6068 * To guarantee that the orphans can be found when LFSCK run next time
6069 * without e2fsck again, when remove the orphan name entry, the LFSCK
6070 * will set the orphan's ctime attribute as 1. Since normal applications
6071 * cannot change the object's ctime attribute as 1. Then when LFSCK run
6072 * next time, it can record the object (that ctime is 1) in the namespace
6073 * LFSCK trace file during the first-stage scanning.
6075 * \param[in] env pointer to the thread context
6076 * \param[in] com pointer to the lfsck component
6077 * \param[in] parent pointer to the object for the backend /lost+found
6078 * \param[in] ent pointer to the name entry for the target under the
6079 * backend /lost+found
6081 * \retval positive for repaired
6082 * \retval 0 if needs to repair nothing
6083 * \retval negative error number on failure
6085 static int lfsck_namespace_scan_local_lpf_one(const struct lu_env *env,
6086 struct lfsck_component *com,
6087 struct dt_object *parent,
6088 struct lu_dirent *ent)
6090 struct lfsck_thread_info *info = lfsck_env_info(env);
6091 struct lu_fid *key = &info->lti_fid;
6092 struct lu_attr *la = &info->lti_la;
6093 struct lfsck_instance *lfsck = com->lc_lfsck;
6094 struct dt_object *obj;
6095 struct dt_device *dev = lfsck->li_bottom;
6096 struct dt_object *child = NULL;
6097 struct thandle *th = NULL;
6104 child = lfsck_object_find_by_dev(env, dev, &ent->lde_fid);
6106 RETURN(PTR_ERR(child));
6108 LASSERT(dt_object_exists(child));
6109 LASSERT(!dt_object_remote(child));
6111 idx = lfsck_sub_trace_file_fid2idx(&ent->lde_fid);
6112 obj = com->lc_sub_trace_objs[idx].lsto_obj;
6113 fid_cpu_to_be(key, &ent->lde_fid);
6114 rc = dt_lookup(env, obj, (struct dt_rec *)&flags,
6115 (const struct dt_key *)key);
6118 flags |= LNTF_CHECK_ORPHAN;
6119 } else if (rc == -ENOENT) {
6120 flags = LNTF_CHECK_ORPHAN;
6125 th = dt_trans_create(env, dev);
6127 GOTO(out, rc = PTR_ERR(th));
6129 /* a1. remove name entry from backend /lost+found */
6130 rc = dt_declare_delete(env, parent,
6131 (const struct dt_key *)ent->lde_name, th);
6135 if (S_ISDIR(lfsck_object_type(child))) {
6136 /* a2. decrease parent's nlink */
6137 rc = dt_declare_ref_del(env, parent, th);
6143 /* a3. remove child's FID from the LFSCK trace file. */
6144 rc = dt_declare_delete(env, obj,
6145 (const struct dt_key *)key, th);
6149 /* a4. set child's ctime as 1 */
6150 memset(la, 0, sizeof(*la));
6152 la->la_valid = LA_CTIME;
6153 rc = dt_declare_attr_set(env, child, la, th);
6158 /* a5. insert child's FID into the LFSCK trace file. */
6159 rc = dt_declare_insert(env, obj, (const struct dt_rec *)&flags,
6160 (const struct dt_key *)key, th);
6164 rc = dt_trans_start_local(env, dev, th);
6168 /* b1. remove name entry from backend /lost+found */
6169 rc = dt_delete(env, parent, (const struct dt_key *)ent->lde_name, th);
6173 if (S_ISDIR(lfsck_object_type(child))) {
6174 /* b2. decrease parent's nlink */
6175 dt_write_lock(env, parent, 0);
6176 rc = dt_ref_del(env, parent, th);
6177 dt_write_unlock(env, parent);
6183 /* a3. remove child's FID from the LFSCK trace file. */
6184 rc = dt_delete(env, obj, (const struct dt_key *)key, th);
6188 /* b4. set child's ctime as 1 */
6189 rc = dt_attr_set(env, child, la, th);
6194 /* b5. insert child's FID into the LFSCK trace file. */
6195 rc = dt_insert(env, obj, (const struct dt_rec *)&flags,
6196 (const struct dt_key *)key, th);
6198 GOTO(stop, rc = (rc == 0 ? 1 : rc));
6201 dt_trans_stop(env, dev, th);
6204 lfsck_object_put(env, child);
6210 * Handle orphans under the backend /lost+found directory
6212 * Some backend checker, such as e2fsck for ldiskfs may find some orphans
6213 * and put them under the backend /lost+found directory that is invisible
6214 * to client. The LFSCK will scan such directory, for the original client
6215 * visible orphans, add their fids into the namespace LFSCK trace file,
6216 * then the subsenquent namespace LFSCK second-stage scanning can handle
6217 * them as other objects to be double scanned: either move back to normal
6218 * namespace, or to the global visible orphan directory:
6219 * /ROOT/.lustre/lost+found/MDTxxxx/
6221 * \param[in] env pointer to the thread context
6222 * \param[in] com pointer to the lfsck component
6224 static void lfsck_namespace_scan_local_lpf(const struct lu_env *env,
6225 struct lfsck_component *com)
6227 struct lfsck_thread_info *info = lfsck_env_info(env);
6228 struct lu_dirent *ent =
6229 (struct lu_dirent *)info->lti_key;
6230 struct lu_seq_range *range = &info->lti_range;
6231 struct lfsck_instance *lfsck = com->lc_lfsck;
6232 struct ptlrpc_thread *thread = &lfsck->li_thread;
6233 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
6234 struct lfsck_namespace *ns = com->lc_file_ram;
6235 struct dt_object *parent;
6236 const struct dt_it_ops *iops;
6238 struct seq_server_site *ss = lfsck_dev_site(lfsck);
6240 __u32 idx = lfsck_dev_idx(lfsck);
6245 parent = lfsck_object_find_by_dev(env, lfsck->li_bottom,
6246 &LU_BACKEND_LPF_FID);
6247 if (IS_ERR(parent)) {
6248 CERROR("%s: fail to find backend /lost+found: rc = %ld\n",
6249 lfsck_lfsck2name(lfsck), PTR_ERR(parent));
6253 /* It is normal that the /lost+found does not exist for ZFS backend. */
6254 if (!dt_object_exists(parent))
6257 if (unlikely(!dt_try_as_dir(env, parent)))
6258 GOTO(out, rc = -ENOTDIR);
6260 CDEBUG(D_LFSCK, "%s: start to scan backend /lost+found\n",
6261 lfsck_lfsck2name(lfsck));
6263 com->lc_new_scanned = 0;
6264 iops = &parent->do_index_ops->dio_it;
6265 di = iops->init(env, parent, LUDA_64BITHASH | LUDA_TYPE);
6267 GOTO(out, rc = PTR_ERR(di));
6269 rc = iops->load(env, di, 0);
6271 rc = iops->next(env, di);
6276 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
6277 unlikely(!thread_is_running(thread)))
6280 rc = iops->rec(env, di, (struct dt_rec *)ent,
6281 LUDA_64BITHASH | LUDA_TYPE);
6283 rc = lfsck_unpack_ent(ent, &cookie, &type);
6285 if (unlikely(rc != 0)) {
6286 CDEBUG(D_LFSCK, "%s: fail to iterate backend "
6287 "/lost+found: rc = %d\n",
6288 lfsck_lfsck2name(lfsck), rc);
6293 /* skip dot and dotdot entries */
6294 if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen))
6297 if (!fid_seq_in_fldb(fid_seq(&ent->lde_fid)))
6300 if (fid_is_norm(&ent->lde_fid)) {
6301 fld_range_set_mdt(range);
6302 rc = fld_local_lookup(env, ss->ss_server_fld,
6303 fid_seq(&ent->lde_fid), range);
6306 } else if (idx != 0) {
6307 /* If the returned FID is IGIF, then there are three
6310 * 1) The object is upgraded from old Lustre-1.8 with
6311 * IGIF assigned to such object.
6312 * 2) The object is a backend local object and is
6313 * invisible to client.
6314 * 3) The object lost its LMV EA, and since there is
6315 * no FID-in-dirent for the orphan in the backend
6316 * /lost+found directory, then the low layer will
6317 * return IGIF for such object.
6319 * For MDTx (x != 0), it is either case 2) or case 3),
6320 * but from the LFSCK view, they are indistinguishable.
6321 * To be safe, the LFSCK will keep it there and report
6322 * some message, then the adminstrator can handle that
6325 * For MDT0, it is more possible the case 1). The LFSCK
6326 * will handle the orphan as an upgraded object. */
6327 CDEBUG(D_LFSCK, "%s: the orphan %.*s with IGIF "DFID
6328 "in the backend /lost+found on the MDT %04x, "
6329 "to be safe, skip it.\n",
6330 lfsck_lfsck2name(lfsck), ent->lde_namelen,
6331 ent->lde_name, PFID(&ent->lde_fid), idx);
6335 rc = lfsck_namespace_scan_local_lpf_one(env, com, parent, ent);
6338 down_write(&com->lc_sem);
6339 com->lc_new_scanned++;
6340 ns->ln_local_lpf_scanned++;
6342 ns->ln_local_lpf_moved++;
6344 ns->ln_local_lpf_skipped++;
6346 ns->ln_local_lpf_failed++;
6347 up_write(&com->lc_sem);
6349 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
6353 lfsck_control_speed_by_self(com);
6354 if (unlikely(!thread_is_running(thread))) {
6359 rc = iops->next(env, di);
6363 iops->fini(env, di);
6368 CDEBUG(D_LFSCK, "%s: stop to scan backend /lost+found: rc = %d\n",
6369 lfsck_lfsck2name(lfsck), rc);
6371 lfsck_object_put(env, parent);
6375 * Rescan the striped directory after the master LMV EA reset.
6377 * Sometimes, the master LMV EA of the striped directory maybe lost, so when
6378 * the namespace LFSCK engine scan the striped directory for the first time,
6379 * it will be regarded as a normal directory. As the LFSCK processing, some
6380 * other LFSCK instance on other MDT will find the shard of this striped dir,
6381 * and find that the master MDT-object of the striped directory lost its LMV
6382 * EA, then such remote LFSCK instance will regenerate the master LMV EA and
6383 * notify the LFSCK instance on this MDT to rescan the striped directory.
6385 * \param[in] env pointer to the thread context
6386 * \param[in] com pointer to the lfsck component
6387 * \param[in] llu the lfsck_lmv_unit that contains the striped directory
6390 * \retval positive number for success
6391 * \retval 0 for LFSCK stopped/paused
6392 * \retval negative error number on failure
6394 static int lfsck_namespace_rescan_striped_dir(const struct lu_env *env,
6395 struct lfsck_component *com,
6396 struct lfsck_lmv_unit *llu)
6398 struct lfsck_thread_info *info = lfsck_env_info(env);
6399 struct lfsck_instance *lfsck = com->lc_lfsck;
6400 struct lfsck_assistant_data *lad = com->lc_data;
6401 struct dt_object *dir;
6402 const struct dt_it_ops *iops;
6404 struct lu_dirent *ent =
6405 (struct lu_dirent *)info->lti_key;
6406 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
6407 struct ptlrpc_thread *thread = &lfsck->li_thread;
6408 struct lfsck_assistant_object *lso = NULL;
6409 struct lfsck_namespace_req *lnr;
6410 struct lfsck_assistant_req *lar;
6415 LASSERT(list_empty(&lad->lad_req_list));
6417 lfsck->li_lmv = &llu->llu_lmv;
6418 lfsck->li_obj_dir = lfsck_object_get(llu->llu_obj);
6419 rc = lfsck_open_dir(env, lfsck, 0);
6423 dir = lfsck->li_obj_dir;
6424 di = lfsck->li_di_dir;
6425 iops = &dir->do_index_ops->dio_it;
6427 rc = iops->rec(env, di, (struct dt_rec *)ent,
6428 lfsck->li_args_dir);
6430 rc = lfsck_unpack_ent(ent, &lfsck->li_cookie_dir,
6434 if (bk->lb_param & LPF_FAILOUT)
6440 if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen))
6444 lso = lfsck_assistant_object_init(env,
6445 lfsck_dto2fid(dir), NULL,
6446 lfsck->li_pos_current.lp_oit_cookie, true);
6448 if (bk->lb_param & LPF_FAILOUT)
6449 GOTO(out, rc = PTR_ERR(lso));
6456 lnr = lfsck_namespace_assistant_req_init(lfsck, lso, ent, type);
6458 if (bk->lb_param & LPF_FAILOUT)
6459 GOTO(out, rc = PTR_ERR(lnr));
6464 lar = &lnr->lnr_lar;
6465 rc = lfsck_namespace_assistant_handler_p1(env, com, lar);
6466 lfsck_namespace_assistant_req_fini(env, lar);
6467 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
6470 if (unlikely(!thread_is_running(thread)))
6474 rc = iops->next(env, di);
6478 if (lso != NULL && !IS_ERR(lso))
6479 lfsck_assistant_object_put(env, lso);
6481 lfsck_close_dir(env, lfsck, rc);
6485 /* The close_dir() may insert a dummy lnr in the lad->lad_req_list. */
6486 if (list_empty(&lad->lad_req_list))
6489 spin_lock(&lad->lad_lock);
6490 lar = list_entry(lad->lad_req_list.next, struct lfsck_assistant_req,
6492 list_del_init(&lar->lar_list);
6493 spin_unlock(&lad->lad_lock);
6495 rc = lfsck_namespace_assistant_handler_p1(env, com, lar);
6496 lfsck_namespace_assistant_req_fini(env, lar);
6498 RETURN(rc == 0 ? 1 : rc);
6502 lfsck_namespace_double_scan_one_trace_file(const struct lu_env *env,
6503 struct lfsck_component *com,
6504 struct dt_object *obj, bool first)
6506 struct lfsck_instance *lfsck = com->lc_lfsck;
6507 struct ptlrpc_thread *thread = &lfsck->li_thread;
6508 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
6509 struct lfsck_namespace *ns = com->lc_file_ram;
6510 const struct dt_it_ops *iops = &obj->do_index_ops->dio_it;
6511 struct dt_object *target;
6519 di = iops->init(env, obj, 0);
6521 RETURN(PTR_ERR(di));
6524 fid_cpu_to_be(&fid, &ns->ln_fid_latest_scanned_phase2);
6527 rc = iops->get(env, di, (const struct dt_key *)&fid);
6532 /* The start one either has been processed or does not exist,
6534 rc = iops->next(env, di);
6540 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
6541 unlikely(!thread_is_running(thread)))
6544 key = iops->key(env, di);
6553 fid_be_to_cpu(&fid, (const struct lu_fid *)key);
6554 if (!fid_is_sane(&fid)) {
6559 target = lfsck_object_find_bottom(env, lfsck, &fid);
6560 if (IS_ERR(target)) {
6561 rc = PTR_ERR(target);
6565 if (dt_object_exists(target)) {
6566 rc = iops->rec(env, di, (struct dt_rec *)&flags, 0);
6568 rc = lfsck_namespace_double_scan_one(env, com,
6575 lfsck_object_put(env, target);
6578 down_write(&com->lc_sem);
6579 com->lc_new_checked++;
6580 com->lc_new_scanned++;
6582 ns->ln_fid_latest_scanned_phase2 = fid;
6585 ns->ln_objs_repaired_phase2++;
6587 ns->ln_objs_failed_phase2++;
6588 up_write(&com->lc_sem);
6590 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
6593 if (unlikely(com->lc_time_next_checkpoint <=
6594 ktime_get_seconds()) &&
6595 com->lc_new_checked != 0) {
6596 down_write(&com->lc_sem);
6597 ns->ln_run_time_phase2 += ktime_get_seconds() -
6598 com->lc_time_last_checkpoint;
6599 ns->ln_time_last_checkpoint = ktime_get_real_seconds();
6600 ns->ln_objs_checked_phase2 += com->lc_new_checked;
6601 com->lc_new_checked = 0;
6602 lfsck_namespace_store(env, com);
6603 up_write(&com->lc_sem);
6605 com->lc_time_last_checkpoint = ktime_get_seconds();
6606 com->lc_time_next_checkpoint =
6607 com->lc_time_last_checkpoint +
6608 LFSCK_CHECKPOINT_INTERVAL;
6611 lfsck_control_speed_by_self(com);
6612 if (unlikely(!thread_is_running(thread)))
6615 rc = iops->next(env, di);
6624 iops->fini(env, di);
6629 static int lfsck_namespace_assistant_handler_p2(const struct lu_env *env,
6630 struct lfsck_component *com)
6632 struct lfsck_instance *lfsck = com->lc_lfsck;
6633 struct lfsck_namespace *ns = com->lc_file_ram;
6638 while (!list_empty(&lfsck->li_list_lmv)) {
6639 struct lfsck_lmv_unit *llu;
6641 spin_lock(&lfsck->li_lock);
6642 llu = list_entry(lfsck->li_list_lmv.next,
6643 struct lfsck_lmv_unit, llu_link);
6644 list_del_init(&llu->llu_link);
6645 spin_unlock(&lfsck->li_lock);
6647 rc = lfsck_namespace_rescan_striped_dir(env, com, llu);
6652 CDEBUG(D_LFSCK, "%s: namespace LFSCK phase2 scan start\n",
6653 lfsck_lfsck2name(lfsck));
6655 lfsck_namespace_scan_local_lpf(env, com);
6657 com->lc_new_checked = 0;
6658 com->lc_new_scanned = 0;
6659 com->lc_time_last_checkpoint = ktime_get_seconds();
6660 com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
6661 LFSCK_CHECKPOINT_INTERVAL;
6663 i = lfsck_sub_trace_file_fid2idx(&ns->ln_fid_latest_scanned_phase2);
6664 rc = lfsck_namespace_double_scan_one_trace_file(env, com,
6665 com->lc_sub_trace_objs[i].lsto_obj, true);
6666 while (rc > 0 && ++i < LFSCK_STF_COUNT)
6667 rc = lfsck_namespace_double_scan_one_trace_file(env, com,
6668 com->lc_sub_trace_objs[i].lsto_obj, false);
6670 CDEBUG(D_LFSCK, "%s: namespace LFSCK phase2 scan stop at the No. %d "
6671 "trace file: rc = %d\n", lfsck_lfsck2name(lfsck), i, rc);
6676 static void lfsck_namespace_assistant_fill_pos(const struct lu_env *env,
6677 struct lfsck_component *com,
6678 struct lfsck_position *pos)
6680 struct lfsck_assistant_data *lad = com->lc_data;
6681 struct lfsck_namespace_req *lnr;
6683 if (((struct lfsck_namespace *)(com->lc_file_ram))->ln_status !=
6687 if (list_empty(&lad->lad_req_list))
6690 lnr = list_entry(lad->lad_req_list.next,
6691 struct lfsck_namespace_req,
6693 pos->lp_oit_cookie = lnr->lnr_lar.lar_parent->lso_oit_cookie;
6694 pos->lp_dir_cookie = lnr->lnr_dir_cookie - 1;
6695 pos->lp_dir_parent = lnr->lnr_lar.lar_parent->lso_fid;
6698 static int lfsck_namespace_double_scan_result(const struct lu_env *env,
6699 struct lfsck_component *com,
6702 struct lfsck_instance *lfsck = com->lc_lfsck;
6703 struct lfsck_namespace *ns = com->lc_file_ram;
6705 down_write(&com->lc_sem);
6706 ns->ln_run_time_phase2 += ktime_get_seconds() -
6707 com->lc_time_last_checkpoint;
6708 ns->ln_time_last_checkpoint = ktime_get_real_seconds();
6709 ns->ln_objs_checked_phase2 += com->lc_new_checked;
6710 com->lc_new_checked = 0;
6713 if (ns->ln_flags & LF_INCOMPLETE)
6714 ns->ln_status = LS_PARTIAL;
6716 ns->ln_status = LS_COMPLETED;
6717 ns->ln_flags &= ~LF_SCANNED_ONCE;
6718 if (!(lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN))
6719 ns->ln_flags &= ~LF_INCONSISTENT;
6720 ns->ln_time_last_complete = ns->ln_time_last_checkpoint;
6721 ns->ln_success_count++;
6722 } else if (rc == 0) {
6723 if (lfsck->li_status != 0)
6724 ns->ln_status = lfsck->li_status;
6726 ns->ln_status = LS_STOPPED;
6728 ns->ln_status = LS_FAILED;
6731 rc = lfsck_namespace_store(env, com);
6732 up_write(&com->lc_sem);
6738 lfsck_namespace_assistant_sync_failures_interpret(const struct lu_env *env,
6739 struct ptlrpc_request *req,
6743 struct lfsck_async_interpret_args *laia = args;
6744 struct lfsck_tgt_desc *ltd = laia->laia_ltd;
6746 ltd->ltd_synced_failures = 1;
6753 * Notify remote LFSCK instances about former failures.
6755 * The local LFSCK instance has recorded which MDTs have ever failed to respond
6756 * some LFSCK verification requests (maybe because of network issues or the MDT
6757 * itself trouble). During the respond gap the MDT may missed some name entries
6758 * verification, then the MDT cannot know whether related MDT-objects have been
6759 * referenced by related name entries or not, then in the second-stage scanning,
6760 * these MDT-objects will be regarded as orphan, if the MDT-object contains bad
6761 * linkEA for back reference, then it will misguide the LFSCK to generate wrong
6762 * name entry for repairing the orphan.
6764 * To avoid above trouble, when layout LFSCK finishes the first-stage scanning,
6765 * it will scan the bitmap for the ever failed MDTs, and notify them that they
6766 * have ever missed some name entries verification and should skip the handling
6767 * for orphan MDT-objects.
6769 * \param[in] env pointer to the thread context
6770 * \param[in] com pointer to the lfsck component
6771 * \param[in] lr pointer to the lfsck request
6773 static void lfsck_namespace_assistant_sync_failures(const struct lu_env *env,
6774 struct lfsck_component *com,
6775 struct lfsck_request *lr)
6777 struct lfsck_async_interpret_args *laia =
6778 &lfsck_env_info(env)->lti_laia2;
6779 struct lfsck_assistant_data *lad = com->lc_data;
6780 struct lfsck_namespace *ns = com->lc_file_ram;
6781 struct lfsck_instance *lfsck = com->lc_lfsck;
6782 struct lfsck_tgt_descs *ltds = &lfsck->li_mdt_descs;
6783 struct lfsck_tgt_desc *ltd;
6784 struct ptlrpc_request_set *set;
6789 if (!test_bit(LAD_INCOMPLETE, &lad->lad_flags))
6792 set = ptlrpc_prep_set();
6794 GOTO(out, rc = -ENOMEM);
6796 lr->lr_flags2 = ns->ln_flags | LF_INCOMPLETE;
6797 memset(laia, 0, sizeof(*laia));
6798 lad->lad_touch_gen++;
6800 down_read(<ds->ltd_rw_sem);
6801 cfs_foreach_bit(lad->lad_bitmap, idx) {
6802 ltd = lfsck_ltd2tgt(ltds, idx);
6806 laia->laia_ltd = ltd;
6807 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
6808 lfsck_namespace_assistant_sync_failures_interpret,
6809 laia, LFSCK_NOTIFY);
6811 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail "
6812 "to sync failure with MDT %x: rc = %d\n",
6813 lfsck_lfsck2name(lfsck), ltd->ltd_index, rc);
6815 up_read(<ds->ltd_rw_sem);
6817 rc = ptlrpc_set_wait(env, set);
6818 ptlrpc_set_destroy(set);
6824 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail "
6825 "to sync failure with MDTs, and related MDTs "
6826 "may handle orphan improperly: rc = %d\n",
6827 lfsck_lfsck2name(lfsck), rc);
6832 struct lfsck_assistant_operations lfsck_namespace_assistant_ops = {
6833 .la_handler_p1 = lfsck_namespace_assistant_handler_p1,
6834 .la_handler_p2 = lfsck_namespace_assistant_handler_p2,
6835 .la_fill_pos = lfsck_namespace_assistant_fill_pos,
6836 .la_double_scan_result = lfsck_namespace_double_scan_result,
6837 .la_req_fini = lfsck_namespace_assistant_req_fini,
6838 .la_sync_failures = lfsck_namespace_assistant_sync_failures,
6842 * Verify the specified linkEA entry for the given directory object.
6843 * If the object has no such linkEA entry or it has more other linkEA
6844 * entries, then re-generate the linkEA with the given information.
6846 * \param[in] env pointer to the thread context
6847 * \param[in] obj pointer to the dt_object to be handled
6848 * \param[in] cname the name for the child in the parent directory
6849 * \param[in] pfid the parent directory's FID for the linkEA
6851 * \retval 0 for success
6852 * \retval negative error number on failure
6854 int lfsck_verify_linkea(const struct lu_env *env, struct dt_object *obj,
6855 const struct lu_name *cname, const struct lu_fid *pfid)
6857 struct dt_device *dev = lfsck_obj2dev(obj);
6858 struct linkea_data ldata = { NULL };
6859 struct lu_buf linkea_buf;
6862 int fl = LU_XATTR_CREATE;
6866 LASSERT(S_ISDIR(lfsck_object_type(obj)));
6868 rc = lfsck_links_read_with_rec(env, obj, &ldata);
6869 if (rc == -ENODATA) {
6871 } else if (rc == 0) {
6872 fl = LU_XATTR_REPLACE;
6873 if (ldata.ld_leh->leh_reccount != 1) {
6876 rc = linkea_links_find(&ldata, cname, pfid);
6885 rc = linkea_links_new(&ldata, &lfsck_env_info(env)->lti_linkea_buf,
6890 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
6891 ldata.ld_leh->leh_len);
6892 th = dt_trans_create(env, dev);
6894 RETURN(PTR_ERR(th));
6896 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
6897 XATTR_NAME_LINK, fl, th);
6901 rc = dt_trans_start_local(env, dev, th);
6905 dt_write_lock(env, obj, 0);
6906 rc = dt_xattr_set(env, obj, &linkea_buf,
6907 XATTR_NAME_LINK, fl, th);
6908 dt_write_unlock(env, obj);
6913 dt_trans_stop(env, dev, th);
6918 * Get the name and parent directory's FID from the first linkEA entry.
6920 * \param[in] env pointer to the thread context
6921 * \param[in] obj pointer to the object which get linkEA from
6922 * \param[out] name pointer to the buffer to hold the name
6923 * in the first linkEA entry
6924 * \param[out] pfid pointer to the buffer to hold the parent
6925 * directory's FID in the first linkEA entry
6927 * \retval 0 for success
6928 * \retval negative error number on failure
6930 int lfsck_links_get_first(const struct lu_env *env, struct dt_object *obj,
6931 char *name, struct lu_fid *pfid)
6933 struct lu_name *cname = &lfsck_env_info(env)->lti_name;
6934 struct linkea_data ldata = { NULL };
6937 rc = lfsck_links_read_with_rec(env, obj, &ldata);
6941 linkea_first_entry(&ldata);
6942 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, cname, pfid);
6943 if (!linkea_entry_is_valid(&ldata, cname, pfid))
6946 /* To guarantee the 'name' is terminated with '0'. */
6947 memcpy(name, cname->ln_name, cname->ln_namelen);
6948 name[cname->ln_namelen] = 0;
6954 * Update the object's name entry with the given FID.
6956 * \param[in] env pointer to the thread context
6957 * \param[in] lfsck pointer to the lfsck instance
6958 * \param[in] dir pointer to the directory that holds
6960 * \param[in] name the name for the entry to be updated
6961 * \param[in] fid the new FID for the name entry referenced
6962 * \param[in] type the type for the name entry to be updated
6964 * \retval 0 for success
6965 * \retval negative error number on failure
6967 int lfsck_update_name_entry(const struct lu_env *env,
6968 struct lfsck_instance *lfsck,
6969 struct dt_object *dir, const char *name,
6970 const struct lu_fid *fid, __u32 type)
6972 struct lfsck_thread_info *info = lfsck_env_info(env);
6973 struct dt_insert_rec *rec = &info->lti_dt_rec;
6974 struct lfsck_lock_handle *llh = &info->lti_llh;
6975 struct dt_device *dev = lfsck_obj2dev(dir);
6981 rc = lfsck_lock(env, lfsck, dir, name, llh,
6982 MDS_INODELOCK_UPDATE, LCK_PW);
6986 th = dt_trans_create(env, dev);
6988 GOTO(unlock, rc = PTR_ERR(th));
6990 rc = dt_declare_delete(env, dir, (const struct dt_key *)name, th);
6994 rec->rec_type = type;
6996 rc = dt_declare_insert(env, dir, (const struct dt_rec *)rec,
6997 (const struct dt_key *)name, th);
7001 rc = dt_declare_ref_add(env, dir, th);
7005 rc = dt_trans_start_local(env, dev, th);
7009 rc = dt_delete(env, dir, (const struct dt_key *)name, th);
7010 if (rc == -ENOENT) {
7018 rc = dt_insert(env, dir, (const struct dt_rec *)rec,
7019 (const struct dt_key *)name, th);
7020 if (rc == 0 && S_ISDIR(type) && !exists) {
7021 dt_write_lock(env, dir, 0);
7022 rc = dt_ref_add(env, dir, th);
7023 dt_write_unlock(env, dir);
7029 dt_trans_stop(env, dev, th);
7033 CDEBUG(D_LFSCK, "%s: update name entry "DFID"/%s with the FID "DFID
7034 " and the type %o: rc = %d\n", lfsck_lfsck2name(lfsck),
7035 PFID(lfsck_dto2fid(dir)), name, PFID(fid), type, rc);
7040 int lfsck_namespace_setup(const struct lu_env *env,
7041 struct lfsck_instance *lfsck)
7043 struct lfsck_component *com;
7044 struct lfsck_namespace *ns;
7045 struct dt_object *root = NULL;
7046 struct dt_object *obj;
7051 LASSERT(lfsck->li_master);
7057 INIT_LIST_HEAD(&com->lc_link);
7058 INIT_LIST_HEAD(&com->lc_link_dir);
7059 init_rwsem(&com->lc_sem);
7060 atomic_set(&com->lc_ref, 1);
7061 com->lc_lfsck = lfsck;
7062 com->lc_type = LFSCK_TYPE_NAMESPACE;
7063 com->lc_ops = &lfsck_namespace_ops;
7064 com->lc_data = lfsck_assistant_data_init(
7065 &lfsck_namespace_assistant_ops,
7067 if (com->lc_data == NULL)
7068 GOTO(out, rc = -ENOMEM);
7070 com->lc_file_size = sizeof(struct lfsck_namespace);
7071 OBD_ALLOC(com->lc_file_ram, com->lc_file_size);
7072 if (com->lc_file_ram == NULL)
7073 GOTO(out, rc = -ENOMEM);
7075 OBD_ALLOC(com->lc_file_disk, com->lc_file_size);
7076 if (com->lc_file_disk == NULL)
7077 GOTO(out, rc = -ENOMEM);
7079 for (i = 0; i < LFSCK_STF_COUNT; i++)
7080 mutex_init(&com->lc_sub_trace_objs[i].lsto_mutex);
7082 root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
7084 GOTO(out, rc = PTR_ERR(root));
7086 if (unlikely(!dt_try_as_dir(env, root)))
7087 GOTO(out, rc = -ENOTDIR);
7089 obj = local_index_find_or_create(env, lfsck->li_los, root,
7091 S_IFREG | S_IRUGO | S_IWUSR,
7092 &dt_lfsck_namespace_features);
7094 GOTO(out, rc = PTR_ERR(obj));
7097 rc = lfsck_namespace_load(env, com);
7098 if (rc == -ENODATA) {
7099 rc = lfsck_namespace_init(env, com);
7100 } else if (rc < 0) {
7101 rc = lfsck_namespace_reset(env, com, true);
7103 rc = lfsck_load_sub_trace_files(env, com,
7104 &dt_lfsck_namespace_features, LFSCK_NAMESPACE, false);
7106 rc = lfsck_namespace_reset(env, com, true);
7111 ns = com->lc_file_ram;
7112 switch (ns->ln_status) {
7117 spin_lock(&lfsck->li_lock);
7118 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
7119 spin_unlock(&lfsck->li_lock);
7122 CERROR("%s: unknown lfsck_namespace status %d\n",
7123 lfsck_lfsck2name(lfsck), ns->ln_status);
7125 case LS_SCANNING_PHASE1:
7126 case LS_SCANNING_PHASE2:
7127 /* No need to store the status to disk right now.
7128 * If the system crashed before the status stored,
7129 * it will be loaded back when next time. */
7130 ns->ln_status = LS_CRASHED;
7134 spin_lock(&lfsck->li_lock);
7135 list_add_tail(&com->lc_link, &lfsck->li_list_scan);
7136 list_add_tail(&com->lc_link_dir, &lfsck->li_list_dir);
7137 spin_unlock(&lfsck->li_lock);
7144 if (root != NULL && !IS_ERR(root))
7145 lfsck_object_put(env, root);
7147 lfsck_component_cleanup(env, com);
7148 CERROR("%s: fail to init namespace LFSCK component: rc = %d\n",
7149 lfsck_lfsck2name(lfsck), rc);