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_of(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 = lfsck_trans_create(env, dev, lfsck);
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_dir(env, dir, name, fid);
601 RETURN(LFSCK_NAMEENTRY_REMOVED);
606 if (!lu_fid_eq(fid, lfsck_dto2fid(obj)))
607 RETURN(LFSCK_NAMEENTRY_RECREATED);
612 static int lfsck_declare_namespace_exec_dir(const struct lu_env *env,
613 struct dt_object *obj,
614 struct thandle *handle)
618 /* For remote updating LINKEA, there may be further LFSCK action
619 * on remote MDT after the updating, so update the LINKEA ASAP. */
620 if (dt_object_remote(obj))
623 /* For destroying all invalid linkEA entries. */
624 rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, handle);
626 /* For insert new linkEA entry. */
627 rc = dt_declare_xattr_set(env, obj,
628 lfsck_buf_get_const(env, NULL, MAX_LINKEA_SIZE),
629 XATTR_NAME_LINK, 0, handle);
633 int __lfsck_links_read(const struct lu_env *env, struct dt_object *obj,
634 struct linkea_data *ldata, bool with_rec)
638 if (ldata->ld_buf->lb_buf == NULL)
641 if (!dt_object_exists(obj))
644 rc = dt_xattr_get(env, obj, ldata->ld_buf, XATTR_NAME_LINK);
646 /* Buf was too small, figure out what we need. */
647 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LINK);
648 if (unlikely(rc == 0))
654 lu_buf_realloc(ldata->ld_buf, rc);
655 if (ldata->ld_buf->lb_buf == NULL)
658 rc = dt_xattr_get(env, obj, ldata->ld_buf, XATTR_NAME_LINK);
661 if (unlikely(rc == 0))
666 rc = linkea_init_with_rec(ldata);
668 rc = linkea_init(ldata);
675 * Remove linkEA for the given object.
677 * The caller should take the ldlm lock before the calling.
679 * \param[in] env pointer to the thread context
680 * \param[in] com pointer to the lfsck component
681 * \param[in] obj pointer to the dt_object to be handled
683 * \retval 0 for repaired cases
684 * \retval negative error number on failure
686 static int lfsck_namespace_links_remove(const struct lu_env *env,
687 struct lfsck_component *com,
688 struct dt_object *obj)
690 struct lfsck_instance *lfsck = com->lc_lfsck;
691 struct dt_device *dev = lfsck_obj2dev(obj);
692 struct thandle *th = NULL;
696 LASSERT(dt_object_remote(obj) == 0);
698 th = lfsck_trans_create(env, dev, lfsck);
700 GOTO(log, rc = PTR_ERR(th));
702 rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, th);
706 rc = dt_trans_start_local(env, dev, th);
710 dt_write_lock(env, obj, 0);
711 if (unlikely(lfsck_is_dead_obj(obj)))
712 GOTO(unlock, rc = -ENOENT);
714 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
715 GOTO(unlock, rc = 0);
717 rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, th);
722 dt_write_unlock(env, obj);
725 dt_trans_stop(env, dev, th);
728 CDEBUG(D_LFSCK, "%s: namespace LFSCK remove invalid linkEA "
729 "for the object "DFID": rc = %d\n",
730 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
733 struct lfsck_namespace *ns = com->lc_file_ram;
735 ns->ln_flags |= LF_INCONSISTENT;
741 static int lfsck_links_write(const struct lu_env *env, struct dt_object *obj,
742 struct linkea_data *ldata, struct thandle *handle)
747 lfsck_buf_init(&buf, ldata->ld_buf->lb_buf, ldata->ld_leh->leh_len);
750 rc = dt_xattr_set(env, obj, &buf, XATTR_NAME_LINK, 0, handle);
751 if (unlikely(rc == -ENOSPC)) {
752 rc = linkea_overflow_shrink(ldata);
753 if (likely(rc > 0)) {
762 static inline bool linkea_reclen_is_valid(const struct linkea_data *ldata)
764 if (ldata->ld_reclen <= 0)
767 if ((char *)ldata->ld_lee + ldata->ld_reclen >
768 (char *)ldata->ld_leh + ldata->ld_leh->leh_len)
774 static inline bool linkea_entry_is_valid(const struct linkea_data *ldata,
775 const struct lu_name *cname,
776 const struct lu_fid *pfid)
778 if (!linkea_reclen_is_valid(ldata))
781 if (cname->ln_namelen <= 0 || cname->ln_namelen > NAME_MAX)
784 if (!fid_is_sane(pfid))
790 static int lfsck_namespace_unpack_linkea_entry(struct linkea_data *ldata,
791 struct lu_name *cname,
793 char *buf, const int buflen)
795 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, cname, pfid);
796 if (unlikely(!linkea_entry_is_valid(ldata, cname, pfid)))
799 /* To guarantee the 'name' is terminated with '0'. */
800 memcpy(buf, cname->ln_name, cname->ln_namelen);
801 buf[cname->ln_namelen] = 0;
802 cname->ln_name = buf;
807 static void lfsck_linkea_del_buf(struct linkea_data *ldata,
808 const struct lu_name *lname)
810 LASSERT(ldata->ld_leh != NULL && ldata->ld_lee != NULL);
812 /* If current record is corrupted, all the subsequent
813 * records will be dropped. */
814 if (unlikely(!linkea_reclen_is_valid(ldata))) {
815 void *ptr = ldata->ld_lee;
817 ldata->ld_leh->leh_len = sizeof(struct link_ea_header);
818 ldata->ld_leh->leh_reccount = 0;
819 linkea_first_entry(ldata);
820 while (ldata->ld_lee != NULL &&
821 (char *)ldata->ld_lee < (char *)ptr) {
822 int reclen = (ldata->ld_lee->lee_reclen[0] << 8) |
823 ldata->ld_lee->lee_reclen[1];
825 ldata->ld_leh->leh_len += reclen;
826 ldata->ld_leh->leh_reccount++;
827 ldata->ld_lee = (struct link_ea_entry *)
828 ((char *)ldata->ld_lee + reclen);
831 ldata->ld_lee = NULL;
833 linkea_del_buf(ldata, lname, false);
837 static int lfsck_namespace_filter_linkea_entry(struct linkea_data *ldata,
838 struct lu_name *cname,
842 struct link_ea_entry *oldlee;
846 oldlee = ldata->ld_lee;
847 oldlen = ldata->ld_reclen;
848 linkea_next_entry(ldata);
849 while (ldata->ld_lee != NULL) {
850 ldata->ld_reclen = (ldata->ld_lee->lee_reclen[0] << 8) |
851 ldata->ld_lee->lee_reclen[1];
852 if (unlikely(!linkea_reclen_is_valid(ldata))) {
853 lfsck_linkea_del_buf(ldata, NULL);
854 LASSERT(ldata->ld_lee == NULL);
855 } else if (unlikely(ldata->ld_reclen == oldlen &&
856 memcmp(ldata->ld_lee, oldlee, oldlen) == 0)) {
861 lfsck_linkea_del_buf(ldata, cname);
863 linkea_next_entry(ldata);
866 ldata->ld_lee = oldlee;
867 ldata->ld_reclen = oldlen;
873 * Insert orphan into .lustre/lost+found/MDTxxxx/ locally.
875 * Add the specified orphan MDT-object to the .lustre/lost+found/MDTxxxx/
876 * with the given type to generate the name, the detailed rules for name
877 * have been described as following.
879 * The function also generates the linkEA corresponding to the name entry
880 * under the .lustre/lost+found/MDTxxxx/ for the orphan MDT-object.
882 * \param[in] env pointer to the thread context
883 * \param[in] com pointer to the lfsck component
884 * \param[in] orphan pointer to the orphan MDT-object
885 * \param[in] infix additional information for the orphan name, such as
886 * the FID for original
887 * \param[in] type the type for describing why the orphan MDT-object is
888 * created. The rules are as following:
890 * type "D": The MDT-object is a directory, it may knows its parent
891 * but because there is no valid linkEA, the LFSCK cannot
892 * know where to put it back to the namespace.
893 * type "O": The MDT-object has no linkEA, and there is no name
894 * entry that references the MDT-object.
896 * type "S": The orphan MDT-object is a shard of a striped directory
898 * \see lfsck_layout_recreate_parent() for more types.
900 * The orphan name will be like:
901 * ${FID}-${infix}-${type}-${conflict_version}
903 * \param[out] count if some others inserted some linkEA entries by race,
904 * then return the linkEA entries count.
906 * \retval positive number for repaired cases
907 * \retval 0 if needs to repair nothing
908 * \retval negative error number on failure
910 static int lfsck_namespace_insert_orphan(const struct lu_env *env,
911 struct lfsck_component *com,
912 struct dt_object *orphan,
913 const char *infix, const char *type,
916 struct lfsck_thread_info *info = lfsck_env_info(env);
917 struct lu_name *cname = &info->lti_name;
918 struct dt_insert_rec *rec = &info->lti_dt_rec;
919 struct lu_attr *la = &info->lti_la2;
920 const struct lu_fid *cfid = lfsck_dto2fid(orphan);
921 const struct lu_fid *pfid;
923 struct lfsck_instance *lfsck = com->lc_lfsck;
924 struct dt_device *dev = lfsck_obj2dev(orphan);
925 struct dt_object *parent;
926 struct thandle *th = NULL;
927 struct lfsck_lock_handle *pllh = &info->lti_llh;
928 struct lustre_handle clh = { 0 };
929 struct linkea_data ldata2 = { NULL };
930 struct lu_buf linkea_buf;
937 cname->ln_name = NULL;
938 if (unlikely(lfsck->li_lpf_obj == NULL))
939 GOTO(log, rc = -ENXIO);
941 parent = lfsck->li_lpf_obj;
942 pfid = lfsck_dto2fid(parent);
946 namelen = snprintf(info->lti_key, NAME_MAX, DFID"%s-%s-%d",
947 PFID(cfid), infix, type, idx++);
948 rc = dt_lookup_dir(env, parent, info->lti_key, &tfid);
949 if (rc != 0 && rc != -ENOENT)
952 if (unlikely(rc == 0 && lu_fid_eq(cfid, &tfid)))
954 } while (rc == 0 && !exist);
956 rc = lfsck_lock(env, lfsck, parent, info->lti_key, pllh,
957 MDS_INODELOCK_UPDATE, LCK_PW);
961 /* Re-check whether the name conflict with othrs after taken
963 rc = dt_lookup_dir(env, parent, info->lti_key, &tfid);
965 if (!lu_fid_eq(cfid, &tfid)) {
972 } else if (rc != -ENOENT) {
978 cname->ln_name = info->lti_key;
979 cname->ln_namelen = namelen;
980 rc = linkea_links_new(&ldata2, &info->lti_linkea_buf2,
985 rc = lfsck_ibits_lock(env, lfsck, orphan, &clh,
986 MDS_INODELOCK_UPDATE | MDS_INODELOCK_LOOKUP |
987 MDS_INODELOCK_XATTR, LCK_EX);
991 lfsck_buf_init(&linkea_buf, ldata2.ld_buf->lb_buf,
992 ldata2.ld_leh->leh_len);
993 th = lfsck_trans_create(env, dev, lfsck);
995 GOTO(log, rc = PTR_ERR(th));
997 if (S_ISDIR(lfsck_object_type(orphan))) {
998 rc = dt_declare_delete(env, orphan,
999 (const struct dt_key *)dotdot, th);
1003 rec->rec_type = S_IFDIR;
1004 rec->rec_fid = pfid;
1005 rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
1006 (const struct dt_key *)dotdot, th);
1011 rc = dt_declare_xattr_set(env, orphan, &linkea_buf,
1012 XATTR_NAME_LINK, 0, th);
1017 rec->rec_type = lfsck_object_type(orphan) & S_IFMT;
1018 rec->rec_fid = cfid;
1019 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1020 (const struct dt_key *)cname->ln_name,
1025 if (S_ISDIR(rec->rec_type)) {
1026 rc = dt_declare_ref_add(env, parent, th);
1032 memset(la, 0, sizeof(*la));
1033 la->la_ctime = ktime_get_real_seconds();
1034 la->la_valid = LA_CTIME;
1035 rc = dt_declare_attr_set(env, orphan, la, th);
1039 rc = dt_trans_start_local(env, dev, th);
1043 dt_write_lock(env, orphan, 0);
1044 rc = lfsck_links_read2_with_rec(env, orphan, &ldata2);
1045 if (likely(rc == -ENODATA || rc == -EINVAL)) {
1046 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1047 GOTO(unlock, rc = 1);
1049 if (S_ISDIR(lfsck_object_type(orphan))) {
1050 rc = dt_delete(env, orphan,
1051 (const struct dt_key *)dotdot, th);
1055 rec->rec_type = S_IFDIR;
1056 rec->rec_fid = pfid;
1057 rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1058 (const struct dt_key *)dotdot, th);
1063 rc = dt_xattr_set(env, orphan, &linkea_buf, XATTR_NAME_LINK, 0,
1066 if (rc == 0 && count != NULL)
1067 *count = ldata2.ld_leh->leh_reccount;
1071 dt_write_unlock(env, orphan);
1073 if (rc == 0 && !exist) {
1074 rec->rec_type = lfsck_object_type(orphan) & S_IFMT;
1075 rec->rec_fid = cfid;
1076 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1077 (const struct dt_key *)cname->ln_name, th);
1078 if (rc == 0 && S_ISDIR(rec->rec_type)) {
1079 dt_write_lock(env, parent, 0);
1080 rc = dt_ref_add(env, parent, th);
1081 dt_write_unlock(env, parent);
1086 rc = dt_attr_set(env, orphan, la, th);
1088 GOTO(stop, rc = (rc == 0 ? 1 : rc));
1091 dt_write_unlock(env, orphan);
1094 dt_trans_stop(env, dev, th);
1097 lfsck_ibits_unlock(&clh, LCK_EX);
1099 CDEBUG(D_LFSCK, "%s: namespace LFSCK insert orphan for the "
1100 "object "DFID", name = %s: rc = %d\n",
1101 lfsck_lfsck2name(lfsck), PFID(cfid),
1102 cname->ln_name != NULL ? cname->ln_name : "<NULL>", rc);
1105 struct lfsck_namespace *ns = com->lc_file_ram;
1107 ns->ln_flags |= LF_INCONSISTENT;
1113 static int lfsck_lmv_set(const struct lu_env *env,
1114 struct lfsck_instance *lfsck,
1115 struct dt_object *obj,
1116 struct lmv_mds_md_v1 *lmv)
1118 struct dt_device *dev = lfsck->li_next;
1119 struct thandle *th = NULL;
1120 struct lu_buf buf = { lmv, sizeof(*lmv) };
1125 th = lfsck_trans_create(env, dev, lfsck);
1127 RETURN(PTR_ERR(th));
1129 rc = dt_declare_xattr_set(env, obj, &buf, XATTR_NAME_LMV, 0, th);
1133 rc = dt_trans_start_local(env, dev, th);
1137 rc = dt_xattr_set(env, obj, &buf, XATTR_NAME_LMV, 0, th);
1143 dt_trans_stop(env, dev, th);
1148 static int lfsck_lmv_delete(const struct lu_env *env,
1149 struct lfsck_instance *lfsck,
1150 struct dt_object *obj)
1152 struct dt_device *dev = lfsck->li_next;
1153 struct thandle *th = NULL;
1158 th = lfsck_trans_create(env, dev, lfsck);
1160 RETURN(PTR_ERR(th));
1162 rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LMV, th);
1166 rc = dt_trans_start_local(env, dev, th);
1170 rc = dt_xattr_del(env, obj, XATTR_NAME_LMV, th);
1176 dt_trans_stop(env, dev, th);
1181 static inline int lfsck_object_is_shard(const struct lu_env *env,
1182 struct lfsck_instance *lfsck,
1183 struct dt_object *obj,
1184 const struct lu_name *lname)
1186 struct lfsck_thread_info *info = lfsck_env_info(env);
1187 struct lmv_mds_md_v1 *lmv = &info->lti_lmv;
1190 rc = lfsck_shard_name_to_index(env, lname->ln_name, lname->ln_namelen,
1191 lfsck_object_type(obj),
1192 lfsck_dto2fid(obj));
1196 rc = lfsck_read_stripe_lmv(env, lfsck, obj, lmv);
1200 if (!rc && lmv->lmv_magic == LMV_MAGIC_STRIPE)
1207 * Add the specified name entry back to namespace.
1209 * If there is a linkEA entry that back references a name entry under
1210 * some parent directory, but such parent directory does not have the
1211 * claimed name entry. On the other hand, the linkEA entries count is
1212 * not larger than the MDT-object's hard link count. Under such case,
1213 * it is quite possible that the name entry is lost. Then the LFSCK
1214 * should add the name entry back to the namespace.
1216 * If \a child is shard, which means \a parent is a striped directory,
1217 * if \a parent has LMV, we need to delete it before insertion because
1218 * now parent's striping is broken and can't be parsed correctly.
1220 * \param[in] env pointer to the thread context
1221 * \param[in] com pointer to the lfsck component
1222 * \param[in] parent pointer to the directory under which the name entry
1223 * will be inserted into
1224 * \param[in] child pointer to the object referenced by the name entry
1225 * that to be inserted into the parent
1226 * \param[in] lname the name for the child in the parent directory
1228 * \retval positive number for repaired cases
1229 * \retval 0 if nothing to be repaired
1230 * \retval negative error number on failure
1232 static int lfsck_namespace_insert_normal(const struct lu_env *env,
1233 struct lfsck_component *com,
1234 struct dt_object *parent,
1235 struct dt_object *child,
1236 const struct lu_name *lname)
1238 struct lfsck_thread_info *info = lfsck_env_info(env);
1239 struct lu_attr *la = &info->lti_la;
1240 struct dt_insert_rec *rec = &info->lti_dt_rec;
1241 struct lfsck_instance *lfsck = com->lc_lfsck;
1242 /* The child and its name may be on different MDTs. */
1243 const struct lu_fid *pfid = lfsck_dto2fid(parent);
1244 const struct lu_fid *cfid = lfsck_dto2fid(child);
1245 struct dt_device *dev = lfsck->li_next;
1246 struct thandle *th = NULL;
1247 struct lfsck_lock_handle *llh = &info->lti_llh;
1248 struct lmv_mds_md_v1 *lmv = &info->lti_lmv;
1249 struct lu_buf buf = { lmv, sizeof(*lmv) };
1250 /* whether parent's LMV is deleted before insertion */
1251 bool parent_lmv_deleted = false;
1252 /* whether parent's LMV is missing */
1253 bool parent_lmv_lost = false;
1258 /* @parent/@child may be based on lfsck->li_bottom,
1259 * but here we need the object based on the lfsck->li_next. */
1261 parent = lfsck_object_locate(dev, parent);
1263 GOTO(log, rc = PTR_ERR(parent));
1265 child = lfsck_object_locate(dev, child);
1267 GOTO(log, rc = PTR_ERR(child));
1269 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1272 rc = lfsck_lock(env, lfsck, parent, lname->ln_name, llh,
1273 MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
1274 MDS_INODELOCK_XATTR, LCK_EX);
1278 rc = lfsck_object_is_shard(env, lfsck, child, lname);
1283 rc = lfsck_read_stripe_lmv(env, lfsck, parent, lmv);
1286 * To add a shard, we need to convert parent to a
1287 * plain directory by deleting its LMV, and after
1288 * insertion set it back.
1290 rc = lfsck_lmv_delete(env, lfsck, parent);
1293 parent_lmv_deleted = true;
1294 lmv->lmv_layout_version++;
1295 lfsck_lmv_header_cpu_to_le(lmv, lmv);
1296 } else if (rc == -ENODATA) {
1297 struct lu_seq_range *range = &info->lti_range;
1298 struct seq_server_site *ss = lfsck_dev_site(lfsck);
1300 rc = lfsck_read_stripe_lmv(env, lfsck, child, lmv);
1304 fld_range_set_mdt(range);
1305 rc = fld_server_lookup(env, ss->ss_server_fld,
1306 fid_seq(lfsck_dto2fid(parent)), range);
1310 parent_lmv_lost = true;
1311 lmv->lmv_magic = LMV_MAGIC;
1312 lmv->lmv_master_mdt_index = range->lsr_index;
1313 lmv->lmv_layout_version++;
1314 lfsck_lmv_header_cpu_to_le(lmv, lmv);
1320 if (unlikely(!dt_try_as_dir(env, parent, true)))
1321 GOTO(unlock, rc = -ENOTDIR);
1323 th = lfsck_trans_create(env, dev, lfsck);
1325 GOTO(unlock, rc = PTR_ERR(th));
1327 rec->rec_type = lfsck_object_type(child) & S_IFMT;
1328 rec->rec_fid = cfid;
1329 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1330 (const struct dt_key *)lname->ln_name, th);
1334 if (S_ISDIR(rec->rec_type)) {
1335 rc = dt_declare_ref_add(env, parent, th);
1340 if (parent_lmv_lost) {
1341 rc = dt_declare_xattr_set(env, parent, &buf, XATTR_NAME_LMV,
1347 la->la_ctime = ktime_get_real_seconds();
1348 la->la_valid = LA_CTIME;
1349 rc = dt_declare_attr_set(env, parent, la, th);
1353 rc = dt_declare_attr_set(env, child, la, th);
1357 rc = dt_trans_start_local(env, dev, th);
1361 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1362 (const struct dt_key *)lname->ln_name, th);
1366 if (S_ISDIR(rec->rec_type)) {
1367 dt_write_lock(env, parent, 0);
1368 rc = dt_ref_add(env, parent, th);
1369 dt_write_unlock(env, parent);
1374 if (parent_lmv_lost) {
1375 rc = dt_xattr_set(env, parent, &buf, XATTR_NAME_LMV, 0, th);
1380 rc = dt_attr_set(env, parent, la, th);
1384 rc = dt_attr_set(env, child, la, th);
1386 GOTO(stop, rc = (rc == 0 ? 1 : rc));
1389 dt_trans_stop(env, dev, th);
1392 if (parent_lmv_deleted)
1393 lfsck_lmv_set(env, lfsck, parent, lmv);
1398 CDEBUG(D_LFSCK, "%s: namespace LFSCK insert object "DFID" with "
1399 "the name %s and type %o to the parent "DFID": rc = %d\n",
1400 lfsck_lfsck2name(lfsck), PFID(cfid), lname->ln_name,
1401 lfsck_object_type(child) & S_IFMT, PFID(pfid), rc);
1404 struct lfsck_namespace *ns = com->lc_file_ram;
1406 ns->ln_flags |= LF_INCONSISTENT;
1408 ns->ln_lost_dirent_repaired++;
1415 * Create the specified orphan directory.
1417 * For the case that the parent MDT-object stored in some MDT-object's
1418 * linkEA entry is lost, the LFSCK will re-create the parent object as
1419 * an orphan and insert it into .lustre/lost+found/MDTxxxx/ directory
1420 * with the name ${FID}-P-${conflict_version}.
1422 * \param[in] env pointer to the thread context
1423 * \param[in] com pointer to the lfsck component
1424 * \param[in] orphan pointer to the orphan MDT-object to be created
1425 * \param[in] lmv pointer to master LMV EA that will be set to the orphan
1427 * \retval positive number for repaired cases
1428 * \retval negative error number on failure
1430 static int lfsck_namespace_create_orphan_dir(const struct lu_env *env,
1431 struct lfsck_component *com,
1432 struct dt_object *orphan,
1433 struct lmv_mds_md_v1 *lmv)
1435 struct lfsck_thread_info *info = lfsck_env_info(env);
1436 struct lu_attr *la = &info->lti_la;
1437 struct dt_allocation_hint *hint = &info->lti_hint;
1438 struct dt_object_format *dof = &info->lti_dof;
1439 struct lu_name *cname = &info->lti_name2;
1440 struct dt_insert_rec *rec = &info->lti_dt_rec;
1441 struct lmv_mds_md_v1 *lmv2 = &info->lti_lmv2;
1442 const struct lu_fid *cfid = lfsck_dto2fid(orphan);
1444 struct lfsck_instance *lfsck = com->lc_lfsck;
1445 struct lfsck_namespace *ns = com->lc_file_ram;
1446 struct dt_device *dev = lfsck_obj2dev(orphan);
1447 struct dt_object *parent = NULL;
1448 struct thandle *th = NULL;
1449 struct lfsck_lock_handle *llh = &info->lti_llh;
1450 struct linkea_data ldata = { NULL };
1451 struct lu_buf linkea_buf;
1452 struct lu_buf lmv_buf;
1460 LASSERT(!dt_object_exists(orphan));
1462 cname->ln_name = NULL;
1463 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1466 if (dt_object_remote(orphan)) {
1467 if (lfsck->li_lpf_root_obj == NULL)
1468 GOTO(log, rc = -EBADF);
1470 idx = lfsck_find_mdt_idx_by_fid(env, lfsck, cfid);
1472 GOTO(log, rc = idx);
1474 snprintf(name, 8, "MDT%04x", idx);
1475 rc = dt_lookup_dir(env, lfsck->li_lpf_root_obj, name, &tfid);
1477 GOTO(log, rc = (rc == -ENOENT ? -ENXIO : rc));
1479 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
1481 GOTO(log, rc = PTR_ERR(parent));
1483 if (unlikely(!dt_try_as_dir(env, parent, true)))
1484 GOTO(log, rc = -ENOTDIR);
1486 if (unlikely(lfsck->li_lpf_obj == NULL))
1487 GOTO(log, rc = -ENXIO);
1489 parent = lfsck->li_lpf_obj;
1492 dev = lfsck_find_dev_by_fid(env, lfsck, cfid);
1494 GOTO(log, rc = PTR_ERR(dev));
1500 namelen = snprintf(name, 31, DFID"-P-%d",
1502 rc = dt_lookup_dir(env, parent, name, &tfid);
1503 if (rc != 0 && rc != -ENOENT)
1507 rc = lfsck_lock(env, lfsck, parent, name, llh,
1508 MDS_INODELOCK_UPDATE, LCK_PW);
1512 /* Re-check whether the name conflict with othrs after taken
1514 rc = dt_lookup_dir(env, parent, name, &tfid);
1515 if (unlikely(rc == 0)) {
1523 cname->ln_name = name;
1524 cname->ln_namelen = namelen;
1526 memset(la, 0, sizeof(*la));
1527 la->la_mode = S_IFDIR | 0700;
1528 la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
1529 LA_ATIME | LA_MTIME | LA_CTIME;
1531 orphan->do_ops->do_ah_init(env, hint, parent, orphan,
1532 la->la_mode & S_IFMT);
1534 memset(dof, 0, sizeof(*dof));
1535 dof->dof_type = dt_mode_to_dft(S_IFDIR);
1537 rc = linkea_links_new(&ldata, &info->lti_linkea_buf2,
1538 cname, lfsck_dto2fid(parent));
1542 th = lfsck_trans_create(env, dev, lfsck);
1544 GOTO(unlock1, rc = PTR_ERR(th));
1546 /* Sync the remote transaction to guarantee that the subsequent
1547 * lock against the @orphan can find the @orphan in time. */
1548 if (dt_object_remote(orphan))
1551 rc = dt_declare_create(env, orphan, la, hint, dof, th);
1555 if (unlikely(!dt_try_as_dir(env, orphan, false)))
1556 GOTO(stop, rc = -ENOTDIR);
1558 rc = dt_declare_ref_add(env, orphan, th);
1562 rec->rec_type = S_IFDIR;
1563 rec->rec_fid = cfid;
1564 rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
1565 (const struct dt_key *)dot, th);
1569 rec->rec_fid = lfsck_dto2fid(parent);
1570 rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
1571 (const struct dt_key *)dotdot, th);
1576 lmv->lmv_magic = LMV_MAGIC;
1577 lmv->lmv_master_mdt_index = lfsck_dev_idx(lfsck);
1578 lfsck_lmv_header_cpu_to_le(lmv2, lmv);
1579 lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
1580 rc = dt_declare_xattr_set(env, orphan, &lmv_buf, XATTR_NAME_LMV,
1586 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
1587 ldata.ld_leh->leh_len);
1588 rc = dt_declare_xattr_set(env, orphan, &linkea_buf,
1589 XATTR_NAME_LINK, 0, th);
1593 rec->rec_fid = cfid;
1594 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1595 (const struct dt_key *)name, th);
1597 rc = dt_declare_ref_add(env, parent, th);
1602 rc = dt_trans_start_local(env, dev, th);
1606 dt_write_lock(env, orphan, 0);
1607 rc = dt_create(env, orphan, la, hint, dof, th);
1611 rc = dt_ref_add(env, orphan, th);
1615 rec->rec_fid = cfid;
1616 rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1617 (const struct dt_key *)dot, th);
1621 rec->rec_fid = lfsck_dto2fid(parent);
1622 rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1623 (const struct dt_key *)dotdot, th);
1628 rc = dt_xattr_set(env, orphan, &lmv_buf, XATTR_NAME_LMV, 0, th);
1633 rc = dt_xattr_set(env, orphan, &linkea_buf,
1634 XATTR_NAME_LINK, 0, th);
1635 dt_write_unlock(env, orphan);
1639 rec->rec_fid = cfid;
1640 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1641 (const struct dt_key *)name, th);
1643 dt_write_lock(env, parent, 0);
1644 rc = dt_ref_add(env, parent, th);
1645 dt_write_unlock(env, parent);
1648 GOTO(stop, rc = (rc == 0 ? 1 : rc));
1651 dt_write_unlock(env, orphan);
1654 rc1 = dt_trans_stop(env, dev, th);
1655 if (rc1 != 0 && rc > 0)
1662 CDEBUG(D_LFSCK, "%s: namespace LFSCK create orphan dir for "
1663 "the object "DFID", name = %s: rc = %d\n",
1664 lfsck_lfsck2name(lfsck), PFID(cfid),
1665 cname->ln_name != NULL ? cname->ln_name : "<NULL>", rc);
1667 if (parent != NULL && !IS_ERR(parent) && parent != lfsck->li_lpf_obj)
1668 lfsck_object_put(env, parent);
1671 ns->ln_flags |= LF_INCONSISTENT;
1677 * Remove the specified entry from the linkEA.
1679 * Locate the linkEA entry with the given @cname and @pfid, then
1680 * remove this entry or the other entries those are repeated with
1683 * \param[in] env pointer to the thread context
1684 * \param[in] com pointer to the lfsck component
1685 * \param[in] obj pointer to the dt_object to be handled
1686 * \param[in,out]ldata pointer to the buffer that holds the linkEA
1687 * \param[in] cname the name for the child in the parent directory
1688 * \param[in] pfid the parent directory's FID for the linkEA
1689 * \param[in] next if true, then remove the first found linkEA
1690 * entry, and move the ldata->ld_lee to next entry
1692 * \retval positive number for repaired cases
1693 * \retval 0 if nothing to be repaired
1694 * \retval negative error number on failure
1696 static int lfsck_namespace_shrink_linkea(const struct lu_env *env,
1697 struct lfsck_component *com,
1698 struct dt_object *obj,
1699 struct linkea_data *ldata,
1700 struct lu_name *cname,
1701 struct lu_fid *pfid,
1704 struct lfsck_instance *lfsck = com->lc_lfsck;
1705 struct dt_device *dev = lfsck_obj2dev(obj);
1706 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
1707 struct thandle *th = NULL;
1708 struct lustre_handle lh = { 0 };
1709 struct linkea_data ldata_new = { NULL };
1710 struct lu_buf linkea_buf;
1715 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
1716 MDS_INODELOCK_UPDATE | MDS_INODELOCK_XATTR,
1722 lfsck_linkea_del_buf(ldata, cname);
1724 lfsck_namespace_filter_linkea_entry(ldata, cname, pfid,
1726 if (ldata->ld_leh->leh_reccount > 0 ||
1727 unlikely(ldata->ld_leh->leh_overflow_time)) {
1728 lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
1729 ldata->ld_leh->leh_len);
1730 buflen = linkea_buf.lb_len;
1734 th = lfsck_trans_create(env, dev, lfsck);
1736 GOTO(unlock1, rc = PTR_ERR(th));
1739 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
1740 XATTR_NAME_LINK, 0, th);
1742 rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, th);
1746 rc = dt_trans_start_local(env, dev, th);
1750 dt_write_lock(env, obj, 0);
1751 if (unlikely(lfsck_is_dead_obj(obj)))
1752 GOTO(unlock2, rc = -ENOENT);
1754 rc = lfsck_links_read2_with_rec(env, obj, &ldata_new);
1756 GOTO(unlock2, rc = (rc == -ENODATA ? 0 : rc));
1758 /* The specified linkEA entry has been removed by race. */
1759 rc = linkea_links_find(&ldata_new, cname, pfid);
1761 GOTO(unlock2, rc = 0);
1763 if (bk->lb_param & LPF_DRYRUN)
1764 GOTO(unlock2, rc = 1);
1767 lfsck_linkea_del_buf(&ldata_new, cname);
1769 lfsck_namespace_filter_linkea_entry(&ldata_new, cname, pfid,
1773 * linkea may change because it doesn't take lock in the first read, if
1774 * it becomes larger, restart from beginning.
1776 if ((ldata_new.ld_leh->leh_reccount > 0 ||
1777 unlikely(ldata_new.ld_leh->leh_overflow_time)) &&
1778 buflen < ldata_new.ld_leh->leh_len) {
1779 dt_write_unlock(env, obj);
1780 dt_trans_stop(env, dev, th);
1781 lfsck_buf_init(&linkea_buf, ldata_new.ld_buf->lb_buf,
1782 ldata_new.ld_leh->leh_len);
1783 buflen = linkea_buf.lb_len;
1788 rc = lfsck_links_write(env, obj, &ldata_new, th);
1790 rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, th);
1792 GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
1795 dt_write_unlock(env, obj);
1798 dt_trans_stop(env, dev, th);
1801 lfsck_ibits_unlock(&lh, LCK_EX);
1804 CDEBUG(D_LFSCK, "%s: namespace LFSCK remove %s linkEA entry "
1805 "for the object: "DFID", parent "DFID", name %.*s\n",
1806 lfsck_lfsck2name(lfsck), next ? "invalid" : "redundant",
1807 PFID(lfsck_dto2fid(obj)), PFID(pfid), cname->ln_namelen,
1811 struct lfsck_namespace *ns = com->lc_file_ram;
1813 ns->ln_flags |= LF_INCONSISTENT;
1820 * Conditionally remove the specified entry from the linkEA.
1822 * Take the parent lock firstly, then check whether the specified
1823 * name entry exists or not: if yes, do nothing; otherwise, call
1824 * lfsck_namespace_shrink_linkea() to remove the linkea entry.
1826 * \param[in] env pointer to the thread context
1827 * \param[in] com pointer to the lfsck component
1828 * \param[in] parent pointer to the parent directory
1829 * \param[in] child pointer to the child object that holds the linkEA
1830 * \param[in,out]ldata pointer to the buffer that holds the linkEA
1831 * \param[in] cname the name for the child in the parent directory
1832 * \param[in] pfid the parent directory's FID for the linkEA
1834 * \retval positive number for repaired cases
1835 * \retval 0 if nothing to be repaired
1836 * \retval negative error number on failure
1838 static int lfsck_namespace_shrink_linkea_cond(const struct lu_env *env,
1839 struct lfsck_component *com,
1840 struct dt_object *parent,
1841 struct dt_object *child,
1842 struct linkea_data *ldata,
1843 struct lu_name *cname,
1844 struct lu_fid *pfid)
1846 struct lfsck_thread_info *info = lfsck_env_info(env);
1847 struct lu_fid *cfid = &info->lti_fid3;
1848 struct lfsck_lock_handle *llh = &info->lti_llh;
1852 rc = lfsck_lock(env, com->lc_lfsck, parent, cname->ln_name, llh,
1853 MDS_INODELOCK_UPDATE, LCK_PR);
1857 dt_read_lock(env, parent, 0);
1858 if (unlikely(lfsck_is_dead_obj(parent))) {
1859 dt_read_unlock(env, parent);
1861 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
1867 rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
1868 (const struct dt_key *)cname->ln_name);
1869 dt_read_unlock(env, parent);
1871 /* It is safe to release the ldlm lock, because when the logic come
1872 * here, we have got all the needed information above whether the
1873 * linkEA entry is valid or not. It is not important that others
1874 * may add new linkEA entry after the ldlm lock released. If other
1875 * has removed the specified linkEA entry by race, then it is OK,
1876 * because the subsequent lfsck_namespace_shrink_linkea() can handle
1879 if (rc == -ENOENT) {
1880 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
1889 /* The LFSCK just found some internal status of cross-MDTs
1890 * create operation. That is normal. */
1891 if (lu_fid_eq(cfid, lfsck_dto2fid(child))) {
1892 linkea_next_entry(ldata);
1897 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata, cname,
1904 * Conditionally replace name entry in the parent.
1906 * As required, the LFSCK may re-create the lost MDT-object for dangling
1907 * name entry, but such repairing may be wrong because of bad FID in the
1908 * name entry. As the LFSCK processing, the real MDT-object may be found,
1909 * then the LFSCK should check whether the former re-created MDT-object
1910 * has been modified or not, if not, then destroy it and update the name
1911 * entry in the parent to reference the real MDT-object.
1913 * \param[in] env pointer to the thread context
1914 * \param[in] com pointer to the lfsck component
1915 * \param[in] parent pointer to the parent directory
1916 * \param[in] child pointer to the MDT-object that may be the real
1917 * MDT-object corresponding to the name entry in parent
1918 * \param[in] cfid the current FID in the name entry
1919 * \param[in] cname contains the name of the child in the parent directory
1921 * \retval positive number for repaired cases
1922 * \retval 0 if nothing to be repaired
1923 * \retval negative error number on failure
1925 static int lfsck_namespace_replace_cond(const struct lu_env *env,
1926 struct lfsck_component *com,
1927 struct dt_object *parent,
1928 struct dt_object *child,
1929 const struct lu_fid *cfid,
1930 const struct lu_name *cname)
1932 struct lfsck_thread_info *info = lfsck_env_info(env);
1933 struct lu_attr *la = &info->lti_la;
1934 struct dt_insert_rec *rec = &info->lti_dt_rec;
1936 struct lfsck_instance *lfsck = com->lc_lfsck;
1937 /* The child and its name may be on different MDTs. */
1938 struct dt_device *dev = lfsck->li_next;
1939 const char *name = cname->ln_name;
1940 const struct lu_fid *pfid = lfsck_dto2fid(parent);
1941 struct dt_object *cobj = NULL;
1942 struct lfsck_lock_handle *pllh = &info->lti_llh;
1943 struct lustre_handle clh = { 0 };
1944 struct linkea_data ldata = { NULL };
1945 struct thandle *th = NULL;
1950 /* @parent/@child may be based on lfsck->li_bottom,
1951 * but here we need the object based on the lfsck->li_next. */
1953 parent = lfsck_object_locate(dev, parent);
1955 GOTO(log, rc = PTR_ERR(parent));
1957 if (unlikely(!dt_try_as_dir(env, parent, true)))
1958 GOTO(log, rc = -ENOTDIR);
1960 rc = lfsck_lock(env, lfsck, parent, name, pllh,
1961 MDS_INODELOCK_UPDATE, LCK_PW);
1965 if (!fid_is_sane(cfid)) {
1970 cobj = lfsck_object_find_by_dev(env, dev, cfid);
1973 if (rc == -ENOENT) {
1981 if (!dt_object_exists(cobj)) {
1986 rc = dt_lookup_dir(env, parent, name, &tfid);
1987 if (rc == -ENOENT) {
1995 /* Someone changed the name entry, cannot replace it. */
1996 if (!lu_fid_eq(cfid, &tfid))
1999 /* lock the object to be destroyed. */
2000 rc = lfsck_ibits_lock(env, lfsck, cobj, &clh,
2001 MDS_INODELOCK_UPDATE |
2002 MDS_INODELOCK_UPDATE | MDS_INODELOCK_XATTR,
2007 if (unlikely(lfsck_is_dead_obj(cobj))) {
2012 rc = dt_attr_get(env, cobj, la);
2016 /* The object has been modified by other(s), or it is not created by
2017 * LFSCK, the two cases are indistinguishable. So cannot replace it. */
2018 if (la->la_ctime != 0)
2021 if (S_ISREG(la->la_mode)) {
2022 rc = dt_xattr_get(env, cobj, &LU_BUF_NULL, XATTR_NAME_LOV);
2023 /* If someone has created related OST-object(s),
2025 if ((rc > 0) || (rc < 0 && rc != -ENODATA))
2026 GOTO(log, rc = (rc > 0 ? 0 : rc));
2030 dt_read_lock(env, child, 0);
2031 rc = lfsck_links_read2_with_rec(env, child, &ldata);
2032 dt_read_unlock(env, child);
2034 /* Someone changed the child, no need to replace. */
2041 rc = linkea_links_find(&ldata, cname, pfid);
2042 /* Someone moved the child, no need to replace. */
2046 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2049 th = lfsck_trans_create(env, dev, lfsck);
2051 GOTO(log, rc = PTR_ERR(th));
2054 rc = dt_declare_destroy(env, cobj, th);
2059 rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
2063 rec->rec_type = S_IFDIR;
2064 rec->rec_fid = lfsck_dto2fid(child);
2065 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
2066 (const struct dt_key *)name, th);
2070 rc = dt_trans_start_local(env, dev, th);
2075 rc = dt_destroy(env, cobj, th);
2080 /* The old name entry maybe not exist. */
2081 rc = dt_delete(env, parent, (const struct dt_key *)name, th);
2082 if (rc != 0 && rc != -ENOENT)
2085 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
2086 (const struct dt_key *)name, th);
2088 GOTO(stop, rc = (rc == 0 ? 1 : rc));
2091 dt_trans_stop(env, dev, th);
2094 lfsck_ibits_unlock(&clh, LCK_EX);
2097 if (cobj != NULL && !IS_ERR(cobj))
2098 lfsck_object_put(env, cobj);
2100 CDEBUG(D_LFSCK, "%s: namespace LFSCK conditionally destroy the "
2101 "object "DFID" because of conflict with the object "DFID
2102 " under the parent "DFID" with name %s: rc = %d\n",
2103 lfsck_lfsck2name(lfsck), PFID(cfid),
2104 PFID(lfsck_dto2fid(child)), PFID(pfid), name, rc);
2110 * Overwrite the linkEA for the object with the given ldata.
2112 * The caller should take the ldlm lock before the calling.
2114 * \param[in] env pointer to the thread context
2115 * \param[in] com pointer to the lfsck component
2116 * \param[in] obj pointer to the dt_object to be handled
2117 * \param[in] ldata pointer to the new linkEA data
2119 * \retval positive number for repaired cases
2120 * \retval 0 if nothing to be repaired
2121 * \retval negative error number on failure
2123 int lfsck_namespace_rebuild_linkea(const struct lu_env *env,
2124 struct lfsck_component *com,
2125 struct dt_object *obj,
2126 struct linkea_data *ldata)
2128 struct lfsck_instance *lfsck = com->lc_lfsck;
2129 struct dt_device *dev = lfsck_obj2dev(obj);
2130 struct thandle *th = NULL;
2131 struct lu_buf linkea_buf;
2135 th = lfsck_trans_create(env, dev, lfsck);
2137 GOTO(log, rc = PTR_ERR(th));
2139 lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
2140 ldata->ld_leh->leh_len);
2141 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
2142 XATTR_NAME_LINK, 0, th);
2146 rc = dt_trans_start_local(env, dev, th);
2150 dt_write_lock(env, obj, 0);
2151 if (unlikely(lfsck_is_dead_obj(obj)))
2152 GOTO(unlock, rc = 0);
2154 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2155 GOTO(unlock, rc = 1);
2157 rc = dt_xattr_set(env, obj, &linkea_buf,
2158 XATTR_NAME_LINK, 0, th);
2160 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2163 dt_write_unlock(env, obj);
2166 dt_trans_stop(env, dev, th);
2169 CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild linkEA for the "
2170 "object "DFID": rc = %d\n",
2171 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
2174 struct lfsck_namespace *ns = com->lc_file_ram;
2176 ns->ln_flags |= LF_INCONSISTENT;
2183 * Repair invalid name entry.
2185 * If the name entry contains invalid information, such as bad file type
2186 * or (and) corrupted object FID, then either remove the name entry or
2187 * udpate the name entry with the given (right) information.
2189 * \param[in] env pointer to the thread context
2190 * \param[in] com pointer to the lfsck component
2191 * \param[in] parent pointer to the parent directory
2192 * \param[in] child pointer to the object referenced by the name entry
2193 * \param[in] name the old name of the child under the parent directory
2194 * \param[in] name2 the new name of the child under the parent directory
2195 * \param[in] type the type claimed by the name entry
2196 * \param[in] update update the name entry if true; otherwise, remove it
2197 * \param[in] dec decrease the parent nlink count if true
2199 * \retval positive number for repaired successfully
2200 * \retval 0 if nothing to be repaired
2201 * \retval negative error number on failure
2203 int lfsck_namespace_repair_dirent(const struct lu_env *env,
2204 struct lfsck_component *com,
2205 struct dt_object *parent,
2206 struct dt_object *child,
2207 const char *name, const char *name2,
2208 __u16 type, bool update, bool dec)
2210 struct lfsck_thread_info *info = lfsck_env_info(env);
2211 struct dt_insert_rec *rec = &info->lti_dt_rec;
2212 const struct lu_fid *pfid = lfsck_dto2fid(parent);
2213 struct lu_fid cfid = {0};
2215 struct lfsck_instance *lfsck = com->lc_lfsck;
2216 struct dt_device *dev = lfsck->li_next;
2217 struct thandle *th = NULL;
2218 struct lfsck_lock_handle *llh = &info->lti_llh;
2219 struct lustre_handle lh = { 0 };
2224 cfid = *lfsck_dto2fid(child);
2225 parent = lfsck_object_locate(dev, parent);
2227 GOTO(log, rc = PTR_ERR(parent));
2229 if (unlikely(!dt_try_as_dir(env, parent, true)))
2230 GOTO(log, rc = -ENOTDIR);
2232 if (!update || strcmp(name, name2) == 0)
2233 rc = lfsck_lock(env, lfsck, parent, name, llh,
2234 MDS_INODELOCK_UPDATE, LCK_PW);
2236 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
2237 MDS_INODELOCK_UPDATE, LCK_PW);
2241 th = lfsck_trans_create(env, dev, lfsck);
2243 GOTO(unlock1, rc = PTR_ERR(th));
2245 rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
2250 rec->rec_type = lfsck_object_type(child) & S_IFMT;
2251 LASSERT(!fid_is_zero(&cfid));
2252 rec->rec_fid = &cfid;
2253 rc = dt_declare_insert(env, parent,
2254 (const struct dt_rec *)rec,
2255 (const struct dt_key *)name2, th);
2260 if (dec && S_ISDIR(type)) {
2261 rc = dt_declare_ref_del(env, parent, th);
2266 rc = dt_trans_start_local(env, dev, th);
2271 dt_write_lock(env, parent, 0);
2272 rc = dt_lookup_dir(env, dt_object_child(parent), name, &tfid);
2273 /* Someone has removed the bad name entry by race. */
2275 GOTO(unlock2, rc = 0);
2280 /* Someone has removed the bad name entry and reused it for other
2281 * object by race. */
2282 if (!lu_fid_eq(&tfid, &cfid))
2283 GOTO(unlock2, rc = 0);
2285 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2286 GOTO(unlock2, rc = 1);
2288 rc = dt_delete(env, parent, (const struct dt_key *)name, th);
2293 rc = dt_insert(env, parent,
2294 (const struct dt_rec *)rec,
2295 (const struct dt_key *)name2, th);
2300 if (dec && S_ISDIR(type)) {
2301 rc = dt_ref_del(env, parent, th);
2306 GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
2309 dt_write_unlock(env, parent);
2312 dt_trans_stop(env, dev, th);
2314 /* We are not sure whether the child will become orphan or not.
2315 * Record it in the LFSCK trace file for further checking in
2316 * the second-stage scanning. */
2317 if (!update && !dec && child && rc == 0)
2318 lfsck_namespace_trace_update(env, com, &cfid,
2319 LNTF_CHECK_LINKEA, true);
2322 /* It is harmless even if unlock the unused lock_handle */
2323 lfsck_ibits_unlock(&lh, LCK_PW);
2327 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found bad name "
2328 "entry for: parent "DFID", child "DFID", name %s, type "
2329 "in name entry %o, type claimed by child %o. repair it "
2330 "by %s with new name2 %s: rc = %d\n",
2331 lfsck_lfsck2name(lfsck), PFID(pfid), PFID(&cfid),
2332 name, type, update ? lfsck_object_type(child) : 0,
2333 update ? "updating" : "removing", name2, rc);
2336 struct lfsck_namespace *ns = com->lc_file_ram;
2338 ns->ln_flags |= LF_INCONSISTENT;
2345 * Update the ".." name entry for the given object.
2347 * The object's ".." is corrupted, this function will update the ".." name
2348 * entry with the given pfid, and the linkEA with the given ldata.
2350 * The caller should take the ldlm lock before the calling.
2352 * \param[in] env pointer to the thread context
2353 * \param[in] com pointer to the lfsck component
2354 * \param[in] obj pointer to the dt_object to be handled
2355 * \param[in] pfid the new fid for the object's ".." name entry
2356 * \param[in] cname the name for the @obj in the parent directory
2358 * \retval positive number for repaired cases
2359 * \retval 0 if nothing to be repaired
2360 * \retval negative error number on failure
2362 static int lfsck_namespace_repair_unmatched_pairs(const struct lu_env *env,
2363 struct lfsck_component *com,
2364 struct dt_object *obj,
2365 const struct lu_fid *pfid,
2366 struct lu_name *cname)
2368 struct lfsck_thread_info *info = lfsck_env_info(env);
2369 struct dt_insert_rec *rec = &info->lti_dt_rec;
2370 struct lfsck_instance *lfsck = com->lc_lfsck;
2371 struct dt_device *dev = lfsck_obj2dev(obj);
2372 struct thandle *th = NULL;
2373 struct linkea_data ldata = { NULL };
2374 struct lu_buf linkea_buf;
2378 LASSERT(!dt_object_remote(obj));
2379 LASSERT(S_ISDIR(lfsck_object_type(obj)));
2381 rc = linkea_links_new(&ldata, &info->lti_big_buf, cname, pfid);
2385 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
2386 ldata.ld_leh->leh_len);
2388 th = lfsck_trans_create(env, dev, lfsck);
2390 GOTO(log, rc = PTR_ERR(th));
2392 rc = dt_declare_delete(env, obj, (const struct dt_key *)dotdot, th);
2396 rec->rec_type = S_IFDIR;
2397 rec->rec_fid = pfid;
2398 rc = dt_declare_insert(env, obj, (const struct dt_rec *)rec,
2399 (const struct dt_key *)dotdot, th);
2403 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
2404 XATTR_NAME_LINK, 0, th);
2408 rc = dt_trans_start_local(env, dev, th);
2412 dt_write_lock(env, obj, 0);
2413 if (unlikely(lfsck_is_dead_obj(obj)))
2414 GOTO(unlock, rc = 0);
2416 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2417 GOTO(unlock, rc = 1);
2419 /* The old ".." name entry maybe not exist. */
2420 dt_delete(env, obj, (const struct dt_key *)dotdot, th);
2422 rc = dt_insert(env, obj, (const struct dt_rec *)rec,
2423 (const struct dt_key *)dotdot, th);
2427 rc = lfsck_links_write(env, obj, &ldata, th);
2429 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2432 dt_write_unlock(env, obj);
2435 dt_trans_stop(env, dev, th);
2438 CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild dotdot name entry for "
2439 "the object "DFID", new parent "DFID": rc = %d\n",
2440 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)),
2444 struct lfsck_namespace *ns = com->lc_file_ram;
2446 ns->ln_flags |= LF_INCONSISTENT;
2453 * Handle orphan @obj during Double Scan Directory.
2455 * Remove the @obj's current (invalid) linkEA entries, and insert
2456 * it in the directory .lustre/lost+found/MDTxxxx/ with the name:
2457 * ${FID}-${PFID}-D-${conflict_version}
2459 * The caller should take the ldlm lock before the calling.
2461 * \param[in] env pointer to the thread context
2462 * \param[in] com pointer to the lfsck component
2463 * \param[in] obj pointer to the orphan object to be handled
2464 * \param[in] pfid the new fid for the object's ".." name entry
2465 * \param[in,out] lh ldlm lock handler for the given @obj
2466 * \param[out] type to tell the caller what the inconsistency is
2468 * \retval positive number for repaired cases
2469 * \retval 0 if nothing to be repaired
2470 * \retval negative error number on failure
2473 lfsck_namespace_dsd_orphan(const struct lu_env *env,
2474 struct lfsck_component *com,
2475 struct dt_object *obj,
2476 const struct lu_fid *pfid,
2477 struct lustre_handle *lh,
2478 enum lfsck_namespace_inconsistency_type *type)
2480 struct lfsck_thread_info *info = lfsck_env_info(env);
2481 struct lfsck_namespace *ns = com->lc_file_ram;
2485 /* Remove the unrecognized linkEA. */
2486 rc = lfsck_namespace_links_remove(env, com, obj);
2487 lfsck_ibits_unlock(lh, LCK_EX);
2488 if (rc < 0 && rc != -ENODATA)
2491 *type = LNIT_MUL_REF;
2493 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
2494 * ever tried to verify some remote MDT-object that resides on this
2495 * MDT, but this MDT failed to respond such request. So means there
2496 * may be some remote name entry on other MDT that references this
2497 * object with another name, so we cannot know whether this linkEA
2498 * is valid or not. So keep it there and maybe resolved when next
2500 if (ns->ln_flags & LF_INCOMPLETE)
2503 /* The unique linkEA is invalid, even if the ".." name entry may be
2504 * valid, we still cannot know via which name entry this directory
2505 * will be referenced. Then handle it as pure orphan. */
2506 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2507 "-"DFID, PFID(pfid));
2508 rc = lfsck_namespace_insert_orphan(env, com, obj,
2509 info->lti_tmpbuf, "D", NULL);
2515 * Double Scan Directory object for single linkEA entry case.
2517 * The given @child has unique linkEA entry. If the linkEA entry is valid,
2518 * then check whether the name is in the namespace or not, if not, add the
2519 * missing name entry back to namespace. If the linkEA entry is invalid,
2520 * then remove it and insert the @child in the .lustre/lost+found/MDTxxxx/
2523 * \param[in] env pointer to the thread context
2524 * \param[in] com pointer to the lfsck component
2525 * \param[in] child pointer to the directory to be double scanned
2526 * \param[in] pfid the FID corresponding to the ".." entry
2527 * \param[in] ldata pointer to the linkEA data for the given @child
2528 * \param[in,out] lh ldlm lock handler for the given @child
2529 * \param[out] type to tell the caller what the inconsistency is
2530 * \param[in] retry if found inconsistency, but the caller does not hold
2531 * ldlm lock on the @child, then set @retry as true
2532 * \param[in] unknown set if does not know how to repair the inconsistency
2534 * \retval positive number for repaired cases
2535 * \retval 0 if nothing to be repaired
2536 * \retval negative error number on failure
2539 lfsck_namespace_dsd_single(const struct lu_env *env,
2540 struct lfsck_component *com,
2541 struct dt_object *child,
2542 const struct lu_fid *pfid,
2543 struct linkea_data *ldata,
2544 struct lustre_handle *lh,
2545 enum lfsck_namespace_inconsistency_type *type,
2546 bool *retry, bool *unknown)
2548 struct lfsck_thread_info *info = lfsck_env_info(env);
2549 struct lu_name *cname = &info->lti_name;
2550 const struct lu_fid *cfid = lfsck_dto2fid(child);
2552 struct lfsck_namespace *ns = com->lc_file_ram;
2553 struct lfsck_instance *lfsck = com->lc_lfsck;
2554 struct dt_object *parent = NULL;
2555 struct lmv_mds_md_v1 *lmv;
2559 rc = lfsck_namespace_unpack_linkea_entry(ldata, cname, &tfid,
2561 sizeof(info->lti_key));
2562 /* The unique linkEA entry with bad parent will be handled as orphan. */
2564 if (!lustre_handle_is_used(lh) && retry != NULL)
2567 rc = lfsck_namespace_dsd_orphan(env, com, child,
2573 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
2575 GOTO(out, rc = PTR_ERR(parent));
2577 /* We trust the unique linkEA entry in spite of whether it matches the
2578 * ".." name entry or not. Because even if the linkEA entry is wrong
2579 * and the ".." name entry is right, we still cannot know via which
2580 * name entry the child will be referenced, since all known entries
2581 * have been verified during the first-stage scanning. */
2582 if (!dt_object_exists(parent)) {
2583 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2584 * has ever tried to verify some remote MDT-object that resides
2585 * on this MDT, but this MDT failed to respond such request. So
2586 * means there may be some remote name entry on other MDT that
2587 * references this object with another name, so we cannot know
2588 * whether this linkEA is valid or not. So keep it there and
2589 * maybe resolved when next LFSCK run. */
2590 if (ns->ln_flags & LF_INCOMPLETE)
2593 if (!lustre_handle_is_used(lh) && retry != NULL) {
2599 lfsck_ibits_unlock(lh, LCK_EX);
2602 lmv = &info->lti_lmv;
2603 rc = lfsck_read_stripe_lmv(env, lfsck, child, lmv);
2604 if (rc != 0 && rc != -ENODATA)
2607 if (rc == -ENODATA || lmv->lmv_magic != LMV_MAGIC_STRIPE) {
2609 } else if (lfsck_shard_name_to_index(env,
2610 cname->ln_name, cname->ln_namelen,
2611 S_IFDIR, cfid) < 0) {
2612 /* It is an invalid name entry, we
2613 * cannot trust the parent also. */
2614 rc = lfsck_namespace_shrink_linkea(env, com, child,
2615 ldata, cname, &tfid, true);
2619 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2620 "-"DFID, PFID(pfid));
2621 rc = lfsck_namespace_insert_orphan(env, com, child,
2622 info->lti_tmpbuf, "S", NULL);
2627 /* Create the lost parent as an orphan. */
2628 rc = lfsck_namespace_create_orphan_dir(env, com, parent, lmv);
2630 /* Add the missing name entry to the parent. */
2631 rc = lfsck_namespace_insert_normal(env, com, parent,
2633 if (unlikely(rc == -EEXIST)) {
2634 /* Unfortunately, someone reused the name
2635 * under the parent by race. So we have
2636 * to remove the linkEA entry from
2637 * current child object. It means that the
2638 * LFSCK cannot recover the system
2639 * totally back to its original status,
2640 * but it is necessary to make the
2641 * current system to be consistent. */
2642 rc = lfsck_namespace_shrink_linkea(env,
2644 cname, &tfid, true);
2646 snprintf(info->lti_tmpbuf,
2647 sizeof(info->lti_tmpbuf),
2648 "-"DFID, PFID(pfid));
2649 rc = lfsck_namespace_insert_orphan(env,
2650 com, child, info->lti_tmpbuf,
2657 } /* !dt_object_exists(parent) */
2659 /* The unique linkEA entry with bad parent will be handled as orphan. */
2660 if (unlikely(!dt_try_as_dir(env, parent, true))) {
2661 if (!lustre_handle_is_used(lh) && retry != NULL)
2664 rc = lfsck_namespace_dsd_orphan(env, com, child,
2670 rc = dt_lookup_dir(env, parent, cname->ln_name, &tfid);
2671 if (rc == -ENOENT) {
2672 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2673 * has ever tried to verify some remote MDT-object that resides
2674 * on this MDT, but this MDT failed to respond such request. So
2675 * means there may be some remote name entry on other MDT that
2676 * references this object with another name, so we cannot know
2677 * whether this linkEA is valid or not. So keep it there and
2678 * maybe resolved when next LFSCK run. */
2679 if (ns->ln_flags & LF_INCOMPLETE)
2682 if (!lustre_handle_is_used(lh) && retry != NULL) {
2688 lfsck_ibits_unlock(lh, LCK_EX);
2689 rc = lfsck_namespace_check_name(env, lfsck, parent, child,
2697 /* It is an invalid name entry, drop it. */
2698 if (unlikely(rc > 0)) {
2699 rc = lfsck_namespace_shrink_linkea(env, com, child,
2700 ldata, cname, &tfid, true);
2702 snprintf(info->lti_tmpbuf,
2703 sizeof(info->lti_tmpbuf),
2704 "-"DFID, PFID(pfid));
2705 rc = lfsck_namespace_insert_orphan(env, com,
2706 child, info->lti_tmpbuf, "D", NULL);
2712 /* Add the missing name entry back to the namespace. */
2713 rc = lfsck_namespace_insert_normal(env, com, parent, child,
2715 if (unlikely(rc == -ESTALE))
2716 /* It may happen when the remote object has been
2717 * removed, but the local MDT is not aware of that. */
2720 if (unlikely(rc == -EEXIST)) {
2721 /* Unfortunately, someone reused the name under the
2722 * parent by race. So we have to remove the linkEA
2723 * entry from current child object. It means that the
2724 * LFSCK cannot recover the system totally back to
2725 * its original status, but it is necessary to make
2726 * the current system to be consistent.
2728 * It also may be because of the LFSCK found some
2729 * internal status of create operation. Under such
2730 * case, nothing to be done. */
2731 rc = lfsck_namespace_shrink_linkea_cond(env, com,
2732 parent, child, ldata, cname, &tfid);
2734 snprintf(info->lti_tmpbuf,
2735 sizeof(info->lti_tmpbuf),
2736 "-"DFID, PFID(pfid));
2737 rc = lfsck_namespace_insert_orphan(env, com,
2738 child, info->lti_tmpbuf, "D", NULL);
2743 } /* rc == -ENOENT */
2748 if (!lu_fid_eq(&tfid, cfid)) {
2749 if (!lustre_handle_is_used(lh) && retry != NULL) {
2755 lfsck_ibits_unlock(lh, LCK_EX);
2756 /* The name entry references another MDT-object that
2757 * may be created by the LFSCK for repairing dangling
2758 * name entry. Try to replace it. */
2759 rc = lfsck_namespace_replace_cond(env, com, parent, child,
2762 rc = lfsck_namespace_dsd_orphan(env, com, child,
2768 /* Zero FID may because the remote directroy object has invalid linkEA,
2769 * or lost linkEA. Under such case, the LFSCK on this MDT does not know
2770 * how to repair the inconsistency, but the namespace LFSCK on the MDT
2771 * where its name entry resides may has more information (name, FID) to
2772 * repair such inconsistency. So here, keep the inconsistency to avoid
2773 * some imporper repairing. */
2774 if (fid_is_zero(pfid)) {
2781 /* The ".." name entry is wrong, update it. */
2782 if (!lu_fid_eq(pfid, lfsck_dto2fid(parent))) {
2783 if (!lustre_handle_is_used(lh) && retry != NULL) {
2789 *type = LNIT_UNMATCHED_PAIRS;
2790 rc = lfsck_namespace_repair_unmatched_pairs(env, com, child,
2791 lfsck_dto2fid(parent), cname);
2797 if (parent != NULL && !IS_ERR(parent))
2798 lfsck_object_put(env, parent);
2804 * Double Scan Directory object for multiple linkEA entries case.
2806 * The given @child has multiple linkEA entries. There is at most one linkEA
2807 * entry will be valid, all the others will be removed. Firstly, the function
2808 * will try to find out the linkEA entry for which the name entry exists under
2809 * the given parent (@pfid). If there is no linkEA entry that matches the given
2810 * ".." name entry, then tries to find out the first linkEA entry that both the
2811 * parent and the name entry exist to rebuild a new ".." name entry.
2813 * \param[in] env pointer to the thread context
2814 * \param[in] com pointer to the lfsck component
2815 * \param[in] child pointer to the directory to be double scanned
2816 * \param[in] pfid the FID corresponding to the ".." entry
2817 * \param[in] ldata pointer to the linkEA data for the given @child
2818 * \param[in,out] lh ldlm lock handler for the given @child
2819 * \param[out] type to tell the caller what the inconsistency is
2820 * \param[in] lpf true if the ".." entry is under lost+found/MDTxxxx/
2821 * \param[in] unknown set if does not know how to repair the inconsistency
2823 * \retval positive number for repaired cases
2824 * \retval 0 if nothing to be repaired
2825 * \retval negative error number on failure
2828 lfsck_namespace_dsd_multiple(const struct lu_env *env,
2829 struct lfsck_component *com,
2830 struct dt_object *child,
2831 const struct lu_fid *pfid,
2832 struct linkea_data *ldata,
2833 struct lustre_handle *lh,
2834 enum lfsck_namespace_inconsistency_type *type,
2835 bool lpf, bool *unknown)
2837 struct lfsck_thread_info *info = lfsck_env_info(env);
2838 struct lu_name *cname = &info->lti_name;
2839 const struct lu_fid *cfid = lfsck_dto2fid(child);
2840 struct lu_fid *pfid2 = &info->lti_fid3;
2842 struct lfsck_namespace *ns = com->lc_file_ram;
2843 struct lfsck_instance *lfsck = com->lc_lfsck;
2844 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
2845 struct dt_object *parent = NULL;
2846 struct linkea_data ldata_new = { NULL };
2847 int dirent_count = 0;
2853 while (ldata->ld_lee != NULL) {
2854 rc = lfsck_namespace_unpack_linkea_entry(ldata, cname, &tfid,
2856 sizeof(info->lti_key));
2857 /* Drop invalid linkEA entry. */
2859 lfsck_linkea_del_buf(ldata, cname);
2863 /* Drop repeated linkEA entries. */
2864 lfsck_namespace_filter_linkea_entry(ldata, cname, &tfid, true);
2866 /* If current dotdot is the .lustre/lost+found/MDTxxxx/,
2867 * then it is possible that: the directry object has ever
2868 * been lost, but its name entry was there. In the former
2869 * LFSCK run, during the first-stage scanning, the LFSCK
2870 * found the dangling name entry, but it did not recreate
2871 * the lost object, and when moved to the second-stage
2872 * scanning, some children objects of the lost directory
2873 * object were found, then the LFSCK recreated such lost
2874 * directory object as an orphan.
2876 * When the LFSCK runs again, if the dangling name is still
2877 * there, the LFSCK should move the orphan directory object
2878 * back to the normal namespace. */
2879 if (!lpf && !fid_is_zero(pfid) &&
2880 !lu_fid_eq(pfid, &tfid) && once) {
2881 linkea_next_entry(ldata);
2885 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
2886 if (IS_ERR(parent)) {
2887 rc = PTR_ERR(parent);
2888 /* if @pfid doesn't have a valid OI mapping, it will
2889 * trigger OI scrub, and -ENONET is is returned if it's
2890 * remote, -EINPROGRESS if local.
2892 if ((rc == -ENOENT || rc == -EINPROGRESS) &&
2893 ldata->ld_leh->leh_reccount > 1) {
2894 lfsck_linkea_del_buf(ldata, cname);
2901 if (!dt_object_exists(parent)) {
2902 lfsck_object_put(env, parent);
2903 if (ldata->ld_leh->leh_reccount > 1) {
2904 /* If it is NOT the last linkEA entry, then
2905 * there is still other chance to make the
2906 * child to be visible via other parent, then
2907 * remove this linkEA entry. */
2908 lfsck_linkea_del_buf(ldata, cname);
2915 /* The linkEA entry with bad parent will be removed. */
2916 if (unlikely(!dt_try_as_dir(env, parent, true))) {
2917 lfsck_object_put(env, parent);
2918 lfsck_linkea_del_buf(ldata, cname);
2922 rc = dt_lookup_dir(env, parent, cname->ln_name, &tfid);
2923 *pfid2 = *lfsck_dto2fid(parent);
2924 if (rc == -ENOENT) {
2925 lfsck_object_put(env, parent);
2926 linkea_next_entry(ldata);
2931 lfsck_object_put(env, parent);
2936 if (lu_fid_eq(&tfid, cfid)) {
2937 lfsck_object_put(env, parent);
2938 /* If the parent (that is declared via linkEA entry)
2939 * directory contains the specified child, but such
2940 * parent does not match the dotdot name entry, then
2941 * trust the linkEA. */
2942 if (!fid_is_zero(pfid) && !lu_fid_eq(pfid, pfid2)) {
2943 *type = LNIT_UNMATCHED_PAIRS;
2944 rc = lfsck_namespace_repair_unmatched_pairs(env,
2945 com, child, pfid2, cname);
2951 /* It is the most common case that we find the
2952 * name entry corresponding to the linkEA entry
2953 * that matches the ".." name entry. */
2954 rc = linkea_links_new(&ldata_new, &info->lti_big_buf,
2959 rc = lfsck_namespace_rebuild_linkea(env, com, child,
2964 lfsck_linkea_del_buf(ldata, cname);
2965 linkea_first_entry(ldata);
2966 /* There may be some invalid dangling name entries under
2967 * other parent directories, remove all of them. */
2968 while (ldata->ld_lee != NULL) {
2969 rc = lfsck_namespace_unpack_linkea_entry(ldata,
2970 cname, &tfid, info->lti_key,
2971 sizeof(info->lti_key));
2975 parent = lfsck_object_find_bottom(env, lfsck,
2977 if (IS_ERR(parent)) {
2978 rc = PTR_ERR(parent);
2979 if (rc != -ENOENT &&
2980 bk->lb_param & LPF_FAILOUT)
2986 if (!dt_object_exists(parent)) {
2987 lfsck_object_put(env, parent);
2991 rc = lfsck_namespace_repair_dirent(env, com,
2992 parent, child, cname->ln_name,
2993 cname->ln_name, S_IFDIR, false, true);
2994 lfsck_object_put(env, parent);
2996 if (bk->lb_param & LPF_FAILOUT)
3005 lfsck_linkea_del_buf(ldata, cname);
3008 ns->ln_dirent_repaired += dirent_count;
3011 } /* lu_fid_eq(&tfid, lfsck_dto2fid(child)) */
3013 lfsck_ibits_unlock(lh, LCK_EX);
3014 /* The name entry references another MDT-object that may be
3015 * created by the LFSCK for repairing dangling name entry.
3016 * Try to replace it. */
3017 rc = lfsck_namespace_replace_cond(env, com, parent, child,
3019 lfsck_object_put(env, parent);
3026 lfsck_linkea_del_buf(ldata, cname);
3027 } /* while (ldata->ld_lee != NULL) */
3029 /* If there is still linkEA overflow, return. */
3030 if (unlikely(ldata->ld_leh->leh_overflow_time))
3033 linkea_first_entry(ldata);
3034 if (ldata->ld_leh->leh_reccount == 1) {
3035 rc = lfsck_namespace_dsd_single(env, com, child, pfid, ldata,
3036 lh, type, NULL, unknown);
3041 /* All linkEA entries are invalid and removed, then handle the @child
3043 if (ldata->ld_leh->leh_reccount == 0) {
3044 rc = lfsck_namespace_dsd_orphan(env, com, child, pfid, lh,
3050 /* If the dangling name entry for the orphan directory object has
3051 * been remvoed, then just check whether the directory object is
3052 * still under the .lustre/lost+found/MDTxxxx/ or not. */
3058 /* There is no linkEA entry that matches the ".." name entry. Find
3059 * the first linkEA entry that both parent and name entry exist to
3060 * rebuild a new ".." name entry. */
3070 * Repair the object's nlink attribute.
3072 * If all the known name entries have been verified, then the object's hard
3073 * link attribute should match the object's linkEA entries count unless the
3074 * object's has too many hard link to be recorded in the linkEA. Such cases
3075 * should have been marked in the LFSCK trace file. Otherwise, trust the
3076 * linkEA to update the object's nlink attribute.
3078 * \param[in] env pointer to the thread context
3079 * \param[in] com pointer to the lfsck component
3080 * \param[in] obj pointer to the dt_object to be handled
3081 * \param[in,out] la pointer to buffer to object's attribute before
3082 * and after the repairing
3084 * \retval positive number for repaired cases
3085 * \retval 0 if nothing to be repaired
3086 * \retval negative error number on failure
3088 static int lfsck_namespace_repair_nlink(const struct lu_env *env,
3089 struct lfsck_component *com,
3090 struct dt_object *obj,
3093 struct lfsck_namespace *ns = com->lc_file_ram;
3094 struct lfsck_instance *lfsck = com->lc_lfsck;
3095 struct dt_device *dev = lfsck_obj2dev(obj);
3096 const struct lu_fid *cfid = lfsck_dto2fid(obj);
3097 struct thandle *th = NULL;
3098 struct linkea_data ldata = { NULL };
3099 struct lustre_handle lh = { 0 };
3100 __u32 old = la->la_nlink;
3104 LASSERT(!dt_object_remote(obj));
3106 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
3107 MDS_INODELOCK_UPDATE, LCK_PW);
3111 th = lfsck_trans_create(env, dev, lfsck);
3113 GOTO(log, rc = PTR_ERR(th));
3115 la->la_valid = LA_NLINK;
3116 rc = dt_declare_attr_set(env, obj, la, th);
3120 rc = dt_trans_start_local(env, dev, th);
3124 dt_write_lock(env, obj, 0);
3125 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
3126 * ever tried to verify some remote MDT-object that resides on this
3127 * MDT, but this MDT failed to respond such request. So means there
3128 * may be some remote name entry on other MDT that references this
3129 * object with another name, so we cannot know whether this linkEA
3130 * is valid or not. So keep it there and maybe resolved when next
3132 if (ns->ln_flags & LF_INCOMPLETE)
3133 GOTO(unlock, rc = 0);
3135 rc = dt_attr_get(env, obj, la);
3137 GOTO(unlock, rc = (rc == -ENOENT ? 0 : rc));
3139 rc = lfsck_links_read2_with_rec(env, obj, &ldata);
3141 GOTO(unlock, rc = (rc == -ENODATA ? 0 : rc));
3143 /* XXX: Currently, we only update the nlink attribute if the known
3144 * linkEA entries is larger than the nlink attribute. That is
3146 if (la->la_nlink >= ldata.ld_leh->leh_reccount ||
3147 unlikely(la->la_nlink == 0 ||
3148 ldata.ld_leh->leh_overflow_time))
3149 GOTO(unlock, rc = 0);
3151 la->la_nlink = ldata.ld_leh->leh_reccount;
3152 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
3153 GOTO(unlock, rc = 1);
3155 rc = dt_attr_set(env, obj, la, th);
3157 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
3160 dt_write_unlock(env, obj);
3163 dt_trans_stop(env, dev, th);
3166 lfsck_ibits_unlock(&lh, LCK_PW);
3167 CDEBUG(D_LFSCK, "%s: namespace LFSCK repaired the object "DFID"'s "
3168 "nlink count from %u to %u: rc = %d\n",
3169 lfsck_lfsck2name(lfsck), PFID(cfid), old, la->la_nlink, rc);
3172 ns->ln_flags |= LF_INCONSISTENT;
3178 * Double scan the directory object for namespace LFSCK.
3180 * This function will verify the <parent, child> pairs in the namespace tree:
3181 * the parent references the child via some name entry that should be in the
3182 * child's linkEA entry, the child should back references the parent via its
3185 * The LFSCK will scan every linkEA entry in turn until find out the first
3186 * matched pairs. If found, then all other linkEA entries will be dropped.
3187 * If all the linkEA entries cannot match the ".." name entry, then there
3188 * are serveral possible cases:
3190 * 1) If there is only one linkEA entry, then trust it as long as the PFID
3191 * in the linkEA entry is valid.
3193 * 2) If there are multiple linkEA entries, then try to find the linkEA
3194 * that matches the ".." name entry. If found, then all other entries
3195 * are invalid; otherwise, it is quite possible that the ".." name entry
3196 * is corrupted. Under such case, the LFSCK will rebuild the ".." name
3197 * entry according to the first valid linkEA entry (both the parent and
3198 * the name entry should exist).
3200 * 3) If the directory object has no (valid) linkEA entry, then the
3201 * directory object will be handled as pure orphan and inserted
3202 * in the .lustre/lost+found/MDTxxxx/ with the name:
3203 * ${self_FID}-${PFID}-D-${conflict_version}
3205 * \param[in] env pointer to the thread context
3206 * \param[in] com pointer to the lfsck component
3207 * \param[in] child pointer to the directory object to be handled
3208 * \param[in] flags to indicate the specical checking on the @child
3210 * \retval positive number for repaired cases
3211 * \retval 0 if nothing to be repaired
3212 * \retval negative error number on failure
3214 static int lfsck_namespace_double_scan_dir(const struct lu_env *env,
3215 struct lfsck_component *com,
3216 struct dt_object *child, __u8 flags)
3218 struct lfsck_thread_info *info = lfsck_env_info(env);
3219 const struct lu_fid *cfid = lfsck_dto2fid(child);
3220 struct lu_fid *pfid = &info->lti_fid2;
3221 struct lfsck_namespace *ns = com->lc_file_ram;
3222 struct lfsck_instance *lfsck = com->lc_lfsck;
3223 struct lustre_handle lh = { 0 };
3224 struct linkea_data ldata = { NULL };
3225 bool unknown = false;
3228 enum lfsck_namespace_inconsistency_type type = LNIT_BAD_LINKEA;
3232 LASSERT(!dt_object_remote(child));
3234 if (flags & LNTF_UNCERTAIN_LMV) {
3235 if (flags & LNTF_RECHECK_NAME_HASH) {
3236 rc = lfsck_namespace_scan_shard(env, com, child);
3240 ns->ln_striped_shards_scanned++;
3242 ns->ln_striped_shards_skipped++;
3246 flags &= ~(LNTF_RECHECK_NAME_HASH | LNTF_UNCERTAIN_LMV);
3250 if (flags & (LNTF_CHECK_LINKEA | LNTF_CHECK_PARENT) &&
3251 !(lfsck->li_bookmark_ram.lb_param & LPF_ALL_TGT)) {
3253 "%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",
3254 lfsck_lfsck2name(lfsck), PFID(cfid));
3259 if (unlikely(!dt_try_as_dir(env, child, true)))
3260 GOTO(out, rc = -ENOTDIR);
3262 /* We only take ldlm lock on the @child when required. When the
3263 * logic comes here for the first time, it is always false. */
3267 rc = lfsck_ibits_lock(env, lfsck, child, &lh,
3268 MDS_INODELOCK_UPDATE |
3269 MDS_INODELOCK_XATTR, LCK_EX);
3274 dt_read_lock(env, child, 0);
3275 if (unlikely(lfsck_is_dead_obj(child))) {
3276 dt_read_unlock(env, child);
3281 rc = dt_lookup_dir(env, child, dotdot, pfid);
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 = lfsck_trans_create(env, dev, lfsck);
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 lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
3498 ldata->ld_leh->leh_len);
3499 rc = dt_xattr_set(env, obj, &linkea_buf, XATTR_NAME_LINK, 0, th);
3500 if (unlikely(rc == -ENOSPC))
3508 dt_write_unlock(env, obj);
3511 dt_trans_stop(env, dev, th);
3514 lfsck_ibits_unlock(&lh, LCK_PW);
3515 CDEBUG(D_LFSCK, "%s: clear linkea overflow timestamp for the object "
3517 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
3523 * Verify the object's agent entry.
3525 * If the object claims to have agent entry but the linkEA does not contain
3526 * remote parent, then remove the agent entry. Otherwise, if the object has
3527 * no agent entry but its linkEA contains remote parent, then will generate
3528 * agent entry for it.
3530 * \param[in] env pointer to the thread context
3531 * \param[in] com pointer to the lfsck component
3532 * \param[in] obj pointer to the dt_object to be handled
3534 * \retval positive number for repaired cases
3535 * \retval 0 if nothing to be repaired
3536 * \retval negative error number on failure
3538 static int lfsck_namespace_check_agent_entry(const struct lu_env *env,
3539 struct lfsck_component *com,
3540 struct dt_object *obj)
3542 struct linkea_data ldata = { NULL };
3543 struct lfsck_thread_info *info = lfsck_env_info(env);
3544 struct lfsck_namespace *ns = com->lc_file_ram;
3545 struct lfsck_instance *lfsck = com->lc_lfsck;
3546 struct lu_fid *pfid = &info->lti_fid2;
3547 struct lu_name *cname = &info->lti_name;
3548 struct lu_seq_range *range = &info->lti_range;
3549 struct seq_server_site *ss = lfsck_dev_site(lfsck);
3550 __u32 idx = lfsck_dev_idx(lfsck);
3552 bool remote = false;
3555 if (!(lfsck->li_bookmark_ram.lb_param & LPF_ALL_TGT))
3558 rc = lfsck_links_read_with_rec(env, obj, &ldata);
3559 if (rc == -ENOENT || rc == -ENODATA)
3562 if (rc && rc != -EINVAL)
3565 /* We check the agent entry again after verifying the linkEA
3566 * successfully. So invalid linkEA should be dryrun mode. */
3567 if (rc == -EINVAL || unlikely(!ldata.ld_leh->leh_reccount))
3570 linkea_first_entry(&ldata);
3571 while (ldata.ld_lee != NULL && !remote) {
3572 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
3574 if (!linkea_entry_is_valid(&ldata, cname, pfid))
3577 fld_range_set_mdt(range);
3578 rc = fld_server_lookup(env, ss->ss_server_fld,
3579 fid_seq(pfid), range);
3581 GOTO(out, rc = (rc == -ENOENT ? 0 : rc));
3583 if (range->lsr_index != idx)
3586 linkea_next_entry(&ldata);
3589 if ((lu_object_has_agent_entry(&obj->do_lu) && !remote) ||
3590 (!lu_object_has_agent_entry(&obj->do_lu) && remote)) {
3591 struct dt_device *dev = lfsck_obj2dev(obj);
3592 struct linkea_data ldata2 = { NULL };
3593 struct lustre_handle lh = { 0 };
3594 struct lu_buf linkea_buf;
3595 struct thandle *handle;
3597 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
3600 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
3601 MDS_INODELOCK_UPDATE |
3602 MDS_INODELOCK_XATTR, LCK_EX);
3606 handle = lfsck_trans_create(env, dev, lfsck);
3608 GOTO(unlock, rc = PTR_ERR(handle));
3610 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
3611 ldata.ld_leh->leh_len);
3612 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
3613 XATTR_NAME_LINK, LU_XATTR_REPLACE, handle);
3617 rc = dt_trans_start_local(env, dev, handle);
3621 dt_write_lock(env, obj, 0);
3622 rc = lfsck_links_read2_with_rec(env, obj, &ldata2);
3624 if (rc == -ENOENT || rc == -ENODATA)
3629 /* If someone changed linkEA by race, then the agent
3630 * entry will be updated by lower layer automatically. */
3631 if (ldata.ld_leh->leh_len != ldata2.ld_leh->leh_len ||
3632 memcmp(ldata.ld_buf->lb_buf, ldata2.ld_buf->lb_buf,
3633 ldata.ld_leh->leh_len) != 0)
3634 GOTO(unlock2, rc = 0);
3636 rc = dt_xattr_set(env, obj, &linkea_buf, XATTR_NAME_LINK,
3637 LU_XATTR_REPLACE, handle);
3644 dt_write_unlock(env, obj);
3646 dt_trans_stop(env, dev, handle);
3648 lfsck_ibits_unlock(&lh, LCK_EX);
3655 ns->ln_agent_entries_repaired++;
3657 CDEBUG(D_LFSCK, "%s: repair agent entry for "DFID": rc = %d\n",
3658 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
3663 * Double scan the MDT-object for namespace LFSCK.
3665 * If the MDT-object contains invalid or repeated linkEA entries, then drop
3666 * those entries from the linkEA; if the linkEA becomes empty or the object
3667 * has no linkEA, then it is an orphan and will be added into the directory
3668 * .lustre/lost+found/MDTxxxx/; if the remote parent is lost, then recreate
3669 * the remote parent; if the name entry corresponding to some linkEA entry
3670 * is lost, then add the name entry back to the namespace.
3672 * \param[in] env pointer to the thread context
3673 * \param[in] com pointer to the lfsck component
3674 * \param[in] child pointer to the dt_object to be handled
3675 * \param[in] flags some hints to indicate how the @child should be handled
3677 * \retval positive number for repaired cases
3678 * \retval 0 if nothing to be repaired
3679 * \retval negative error number on failure
3681 static int lfsck_namespace_double_scan_one(const struct lu_env *env,
3682 struct lfsck_component *com,
3683 struct dt_object *child, __u8 flags)
3685 struct lfsck_thread_info *info = lfsck_env_info(env);
3686 struct lu_attr *la = &info->lti_la;
3687 struct lu_name *cname = &info->lti_name;
3688 struct lu_fid *pfid = &info->lti_fid;
3689 struct lu_fid *cfid = &info->lti_fid2;
3690 struct lfsck_instance *lfsck = com->lc_lfsck;
3691 struct lfsck_namespace *ns = com->lc_file_ram;
3692 struct dt_object *parent = NULL;
3693 struct linkea_data ldata = { NULL };
3694 bool repaired = false;
3699 dt_read_lock(env, child, 0);
3700 if (unlikely(lfsck_is_dead_obj(child))) {
3701 dt_read_unlock(env, child);
3706 if (S_ISDIR(lfsck_object_type(child))) {
3707 dt_read_unlock(env, child);
3708 rc = lfsck_namespace_double_scan_dir(env, com, child, flags);
3709 if (!rc && flags & LNTF_CHECK_AGENT_ENTRY)
3710 rc = lfsck_namespace_check_agent_entry(env, com, child);
3715 rc = lfsck_links_read(env, child, &ldata);
3716 dt_read_unlock(env, child);
3718 if (rc == -EINVAL) {
3719 struct lustre_handle lh = { 0 };
3721 rc = lfsck_ibits_lock(env, com->lc_lfsck, child, &lh,
3722 MDS_INODELOCK_UPDATE |
3723 MDS_INODELOCK_XATTR, LCK_EX);
3725 rc = lfsck_namespace_links_remove(env, com, child);
3726 lfsck_ibits_unlock(&lh, LCK_EX);
3735 if (!(ns->ln_flags & LF_INCOMPLETE) &&
3736 unlikely(lfsck_namespace_linkea_stale_overflow(&ldata, ns))) {
3737 rc = lfsck_namespace_linkea_clear_overflow(env, com, &ldata,
3743 ns->ln_linkea_overflow_cleared++;
3746 linkea_first_entry(&ldata);
3747 while (ldata.ld_lee != NULL) {
3748 rc = lfsck_namespace_unpack_linkea_entry(&ldata, cname, pfid,
3750 sizeof(info->lti_key));
3751 /* Invalid PFID in the linkEA entry. */
3753 rc = lfsck_namespace_shrink_linkea(env, com, child,
3754 &ldata, cname, pfid, true);
3764 rc = lfsck_namespace_filter_linkea_entry(&ldata, cname, pfid,
3766 /* Found repeated linkEA entries */
3768 rc = lfsck_namespace_shrink_linkea(env, com, child,
3769 &ldata, cname, pfid, false);
3781 parent = lfsck_object_find_bottom(env, lfsck, pfid);
3782 if (IS_ERR(parent)) {
3783 rc = PTR_ERR(parent);
3784 /* if @pfid doesn't have a valid OI mapping, it will
3785 * trigger OI scrub, and -ENONET is is returned if it's
3786 * remote, -EINPROGRESS if local.
3788 if ((rc == -ENOENT || rc == -EINPROGRESS) &&
3789 ldata.ld_leh->leh_reccount > 1)
3790 rc = lfsck_namespace_shrink_linkea(env, com,
3791 child, &ldata, cname, pfid, true);
3795 if (!dt_object_exists(parent)) {
3798 if (ldata.ld_leh->leh_reccount > 1) {
3799 /* If it is NOT the last linkEA entry, then
3800 * there is still other chance to make the
3801 * child to be visible via other parent, then
3802 * remove this linkEA entry. */
3803 rc = lfsck_namespace_shrink_linkea(env, com,
3804 child, &ldata, cname, pfid, true);
3806 /* If the LFSCK is marked as LF_INCOMPLETE,
3807 * then means some MDT has ever tried to
3808 * verify some remote MDT-object that resides
3809 * on this MDT, but this MDT failed to respond
3810 * such request. So means there may be some
3811 * remote name entry on other MDT that
3812 * references this object with another name,
3813 * so we cannot know whether this linkEA is
3814 * valid or not. So keep it there and maybe
3815 * resolved when next LFSCK run. */
3816 if (ns->ln_flags & LF_INCOMPLETE) {
3817 lfsck_object_put(env, parent);
3822 /* Create the lost parent as an orphan. */
3823 rc = lfsck_namespace_create_orphan_dir(env, com,
3826 lfsck_object_put(env, parent);
3834 /* Add the missing name entry to the parent. */
3835 rc = lfsck_namespace_insert_normal(env, com,
3836 parent, child, cname);
3837 if (unlikely(rc == -EEXIST))
3838 /* Unfortunately, someone reused the
3839 * name under the parent by race. So we
3840 * have to remove the linkEA entry from
3841 * current child object. It means that
3842 * the LFSCK cannot recover the system
3843 * totally back to its original status,
3844 * but it is necessary to make the
3845 * current system to be consistent. */
3846 rc = lfsck_namespace_shrink_linkea(env,
3850 linkea_next_entry(&ldata);
3853 lfsck_object_put(env, parent);
3861 } /* !dt_object_exists(parent) */
3863 /* The linkEA entry with bad parent will be removed. */
3864 if (unlikely(!dt_try_as_dir(env, parent, true))) {
3865 lfsck_object_put(env, parent);
3866 rc = lfsck_namespace_shrink_linkea(env, com, child,
3867 &ldata, cname, pfid, true);
3877 rc = dt_lookup_dir(env, parent, cname->ln_name, cfid);
3878 if (rc != 0 && rc != -ENOENT) {
3879 lfsck_object_put(env, parent);
3885 if (lu_fid_eq(cfid, lfsck_dto2fid(child))) {
3886 /* It is the most common case that we
3887 * find the name entry corresponding
3888 * to the linkEA entry. */
3889 lfsck_object_put(env, parent);
3890 linkea_next_entry(&ldata);
3892 /* The name entry references another
3893 * MDT-object that may be created by
3894 * the LFSCK for repairing dangling
3895 * name entry. Try to replace it. */
3896 rc = lfsck_namespace_replace_cond(env, com,
3897 parent, child, cfid, cname);
3898 lfsck_object_put(env, parent);
3904 linkea_next_entry(&ldata);
3906 rc = lfsck_namespace_shrink_linkea(env,
3920 /* The following handles -ENOENT case */
3922 rc = dt_attr_get(env, child, la);
3926 /* If there is no name entry in the parent dir and the object
3927 * link count is fewer than the linkea entries count, then the
3928 * linkea entry should be removed. */
3929 if (ldata.ld_leh->leh_reccount > la->la_nlink) {
3930 rc = lfsck_namespace_shrink_linkea_cond(env, com,
3931 parent, child, &ldata, cname, pfid);
3932 lfsck_object_put(env, parent);
3942 /* If the LFSCK is marked as LF_INCOMPLETE, then means some
3943 * MDT has ever tried to verify some remote MDT-object that
3944 * resides on this MDT, but this MDT failed to respond such
3945 * request. So means there may be some remote name entry on
3946 * other MDT that references this object with another name,
3947 * so we cannot know whether this linkEA is valid or not.
3948 * So keep it there and maybe resolved when next LFSCK run. */
3949 if (ns->ln_flags & LF_INCOMPLETE) {
3950 lfsck_object_put(env, parent);
3955 rc = lfsck_namespace_check_name(env, lfsck, parent, child,
3961 lfsck_object_put(env, parent);
3966 /* It is an invalid name entry, drop it. */
3967 if (unlikely(rc > 0)) {
3968 lfsck_object_put(env, parent);
3969 rc = lfsck_namespace_shrink_linkea(env, com, child,
3970 &ldata, cname, pfid, true);
3980 /* Add the missing name entry back to the namespace. */
3981 rc = lfsck_namespace_insert_normal(env, com, parent, child,
3983 if (unlikely(rc == -ESTALE))
3984 /* It may happen when the remote object has been
3985 * removed, but the local MDT is not aware of that. */
3988 if (unlikely(rc == -EEXIST))
3989 /* Unfortunately, someone reused the name under the
3990 * parent by race. So we have to remove the linkEA
3991 * entry from current child object. It means that the
3992 * LFSCK cannot recover the system totally back to
3993 * its original status, but it is necessary to make
3994 * the current system to be consistent.
3996 * It also may be because of the LFSCK found some
3997 * internal status of create operation. Under such
3998 * case, nothing to be done. */
3999 rc = lfsck_namespace_shrink_linkea_cond(env, com,
4000 parent, child, &ldata, cname, pfid);
4002 linkea_next_entry(&ldata);
4004 lfsck_object_put(env, parent);
4015 if (rc < 0 && rc != -ENODATA)
4018 if (rc == 0 && ldata.ld_leh != NULL)
4019 count = ldata.ld_leh->leh_reccount;
4022 /* If the LFSCK is marked as LF_INCOMPLETE, then means some
4023 * MDT has ever tried to verify some remote MDT-object that
4024 * resides on this MDT, but this MDT failed to respond such
4025 * request. So means there may be some remote name entry on
4026 * other MDT that references this object with another name,
4027 * so we cannot know whether this linkEA is valid or not.
4028 * So keep it there and maybe resolved when next LFSCK run. */
4029 if (!(ns->ln_flags & LF_INCOMPLETE) &&
4030 (ldata.ld_leh == NULL ||
4031 !ldata.ld_leh->leh_overflow_time)) {
4032 /* If the child becomes orphan, then insert it into
4033 * the global .lustre/lost+found/MDTxxxx directory. */
4034 rc = lfsck_namespace_insert_orphan(env, com, child,
4040 ns->ln_mul_ref_repaired++;
4045 rc = dt_attr_get(env, child, la);
4049 if (la->la_nlink != 0 && la->la_nlink != count) {
4050 if (unlikely(!S_ISREG(lfsck_object_type(child)) &&
4051 !S_ISLNK(lfsck_object_type(child)))) {
4052 CDEBUG(D_LFSCK, "%s: namespace LFSCK finds "
4053 "the object "DFID"'s nlink count %d "
4054 "does not match linkEA count %d, "
4055 "type %o, skip it.\n",
4056 lfsck_lfsck2name(lfsck),
4057 PFID(lfsck_dto2fid(child)),
4058 la->la_nlink, count,
4059 lfsck_object_type(child));
4060 } else if (la->la_nlink < count &&
4061 likely(!ldata.ld_leh->leh_overflow_time)) {
4062 rc = lfsck_namespace_repair_nlink(env, com,
4065 ns->ln_objs_nlink_repaired++;
4073 if (la->la_nlink > 1)
4074 ns->ln_mul_linked_repaired++;
4080 if (!rc && flags & LNTF_CHECK_AGENT_ENTRY)
4081 rc = lfsck_namespace_check_agent_entry(env, com, child);
4086 static void lfsck_namespace_dump_statistics(struct seq_file *m,
4087 struct lfsck_namespace *ns,
4088 __u64 checked_phase1,
4089 __u64 checked_phase2,
4090 time64_t time_phase1,
4091 time64_t time_phase2, bool dryrun)
4093 const char *postfix = dryrun ? "inconsistent" : "repaired";
4095 seq_printf(m, "checked_phase1: %llu\n"
4096 "checked_phase2: %llu\n"
4099 "failed_phase1: %llu\n"
4100 "failed_phase2: %llu\n"
4101 "directories: %llu\n"
4105 "multiple_linked_checked: %llu\n"
4106 "multiple_linked_%s: %llu\n"
4107 "unknown_inconsistency: %llu\n"
4108 "unmatched_pairs_%s: %llu\n"
4109 "dangling_%s: %llu\n"
4110 "multiple_referenced_%s: %llu\n"
4111 "bad_file_type_%s: %llu\n"
4112 "lost_dirent_%s: %llu\n"
4113 "local_lost_found_scanned: %llu\n"
4114 "local_lost_found_moved: %llu\n"
4115 "local_lost_found_skipped: %llu\n"
4116 "local_lost_found_failed: %llu\n"
4117 "striped_dirs_scanned: %llu\n"
4118 "striped_dirs_%s: %llu\n"
4119 "striped_dirs_failed: %llu\n"
4120 "striped_dirs_disabled: %llu\n"
4121 "striped_dirs_skipped: %llu\n"
4122 "striped_shards_scanned: %llu\n"
4123 "striped_shards_%s: %llu\n"
4124 "striped_shards_failed: %llu\n"
4125 "striped_shards_skipped: %llu\n"
4126 "name_hash_%s: %llu\n"
4127 "linkea_overflow_%s: %llu\n"
4128 "agent_entries_%s: %llu\n"
4129 "success_count: %u\n"
4130 "run_time_phase1: %lld seconds\n"
4131 "run_time_phase2: %lld seconds\n",
4134 dryrun ? "inconsistent" : "updated",
4135 ns->ln_items_repaired,
4136 dryrun ? "inconsistent" : "updated",
4137 ns->ln_objs_repaired_phase2,
4138 ns->ln_items_failed,
4139 ns->ln_objs_failed_phase2,
4140 ns->ln_dirs_checked,
4141 postfix, ns->ln_dirent_repaired,
4142 postfix, ns->ln_linkea_repaired,
4143 postfix, ns->ln_objs_nlink_repaired,
4144 ns->ln_mul_linked_checked,
4145 postfix, ns->ln_mul_linked_repaired,
4146 ns->ln_unknown_inconsistency,
4147 postfix, ns->ln_unmatched_pairs_repaired,
4148 postfix, ns->ln_dangling_repaired,
4149 postfix, ns->ln_mul_ref_repaired,
4150 postfix, ns->ln_bad_type_repaired,
4151 postfix, ns->ln_lost_dirent_repaired,
4152 ns->ln_local_lpf_scanned,
4153 ns->ln_local_lpf_moved,
4154 ns->ln_local_lpf_skipped,
4155 ns->ln_local_lpf_failed,
4156 ns->ln_striped_dirs_scanned,
4157 postfix, ns->ln_striped_dirs_repaired,
4158 ns->ln_striped_dirs_failed,
4159 ns->ln_striped_dirs_disabled,
4160 ns->ln_striped_dirs_skipped,
4161 ns->ln_striped_shards_scanned,
4162 postfix, ns->ln_striped_shards_repaired,
4163 ns->ln_striped_shards_failed,
4164 ns->ln_striped_shards_skipped,
4165 postfix, ns->ln_name_hash_repaired,
4166 dryrun ? "inconsistent" : "cleared",
4167 ns->ln_linkea_overflow_cleared,
4168 postfix, ns->ln_agent_entries_repaired,
4169 ns->ln_success_count,
4174 static void lfsck_namespace_release_lmv(const struct lu_env *env,
4175 struct lfsck_component *com)
4177 struct lfsck_instance *lfsck = com->lc_lfsck;
4178 struct lfsck_namespace *ns = com->lc_file_ram;
4180 while (!list_empty(&lfsck->li_list_lmv)) {
4181 struct lfsck_lmv_unit *llu;
4182 struct lfsck_lmv *llmv;
4184 llu = list_entry(lfsck->li_list_lmv.next,
4185 struct lfsck_lmv_unit, llu_link);
4186 llmv = &llu->llu_lmv;
4188 LASSERTF(atomic_read(&llmv->ll_ref) == 1,
4189 "still in using: %u\n",
4190 atomic_read(&llmv->ll_ref));
4192 ns->ln_striped_dirs_skipped++;
4193 lfsck_lmv_put(env, llmv);
4197 static int lfsck_namespace_check_for_double_scan(const struct lu_env *env,
4198 struct lfsck_component *com,
4199 struct dt_object *obj)
4201 struct lu_attr *la = &lfsck_env_info(env)->lti_la;
4204 rc = dt_attr_get(env, obj, la);
4208 /* zero-linkEA object may be orphan, but it also maybe because
4209 * of upgrading. Currently, we cannot record it for double scan.
4210 * Because it may cause the LFSCK trace file to be too large. */
4212 /* "la_ctime" == 1 means that it has ever been removed from
4213 * backend /lost+found directory but not been added back to
4214 * the normal namespace yet. */
4216 if ((S_ISREG(lfsck_object_type(obj)) && la->la_nlink > 1) ||
4217 unlikely(la->la_ctime == 1))
4218 rc = lfsck_namespace_trace_update(env, com, lfsck_dto2fid(obj),
4219 LNTF_CHECK_LINKEA, true);
4224 /* namespace APIs */
4226 static int lfsck_namespace_reset(const struct lu_env *env,
4227 struct lfsck_component *com, bool init)
4229 struct lfsck_instance *lfsck = com->lc_lfsck;
4230 struct lfsck_namespace *ns = com->lc_file_ram;
4231 struct lfsck_assistant_data *lad = com->lc_data;
4232 struct dt_object *root;
4236 root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
4238 GOTO(log, rc = PTR_ERR(root));
4240 if (unlikely(!dt_try_as_dir(env, root, true)))
4241 GOTO(put, rc = -ENOTDIR);
4243 down_write(&com->lc_sem);
4245 memset(ns, 0, sizeof(*ns));
4247 __u32 count = ns->ln_success_count;
4248 time64_t last_time = ns->ln_time_last_complete;
4250 memset(ns, 0, sizeof(*ns));
4251 ns->ln_success_count = count;
4252 ns->ln_time_last_complete = last_time;
4254 ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
4255 ns->ln_status = LS_INIT;
4256 ns->ln_time_latest_reset = ktime_get_real_seconds();
4258 rc = lfsck_load_one_trace_file(env, com, root, &com->lc_obj,
4259 &dt_lfsck_namespace_features,
4260 LFSCK_NAMESPACE, true);
4264 rc = lfsck_load_sub_trace_files(env, com, &dt_lfsck_namespace_features,
4265 LFSCK_NAMESPACE, true);
4269 clear_bit(LAD_INCOMPLETE, &lad->lad_flags);
4270 CFS_RESET_BITMAP(lad->lad_bitmap);
4272 rc = lfsck_namespace_store(env, com);
4277 up_write(&com->lc_sem);
4280 lfsck_object_put(env, root);
4282 CDEBUG(D_LFSCK, "%s: namespace LFSCK reset: rc = %d\n",
4283 lfsck_lfsck2name(lfsck), rc);
4288 lfsck_namespace_fail(const struct lu_env *env, struct lfsck_component *com,
4291 struct lfsck_namespace *ns = com->lc_file_ram;
4293 down_write(&com->lc_sem);
4295 com->lc_new_checked++;
4296 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
4297 up_write(&com->lc_sem);
4300 static void lfsck_namespace_close_dir(const struct lu_env *env,
4301 struct lfsck_component *com)
4303 struct lfsck_namespace *ns = com->lc_file_ram;
4304 struct lfsck_assistant_data *lad = com->lc_data;
4305 struct lfsck_assistant_object *lso = NULL;
4306 struct lfsck_instance *lfsck = com->lc_lfsck;
4307 struct lfsck_lmv *llmv = lfsck->li_lmv;
4308 struct lfsck_namespace_req *lnr;
4309 struct lu_attr *la = &lfsck_env_info(env)->lti_la2;
4310 __u32 size = sizeof(*lnr) + LFSCK_TMPBUF_LEN;
4312 bool wakeup = false;
4318 rc = dt_attr_get(env, lfsck->li_obj_dir, la);
4322 OBD_ALLOC(lnr, size);
4324 ns->ln_striped_dirs_skipped++;
4329 lso = lfsck_assistant_object_init(env, lfsck_dto2fid(lfsck->li_obj_dir),
4330 la, lfsck->li_pos_current.lp_oit_cookie, true);
4332 OBD_FREE(lnr, size);
4333 ns->ln_striped_dirs_skipped++;
4338 /* Generate a dummy request to indicate that all shards' name entry
4339 * in this striped directory has been scanned for the first time. */
4340 INIT_LIST_HEAD(&lnr->lnr_lar.lar_list);
4341 lnr->lnr_lar.lar_parent = lso;
4342 lnr->lnr_lmv = lfsck_lmv_get(llmv);
4343 lnr->lnr_fid = *lfsck_dto2fid(lfsck->li_obj_dir);
4344 lnr->lnr_dir_cookie = MDS_DIR_END_OFF;
4345 lnr->lnr_size = size;
4346 lnr->lnr_type = lso->lso_attr.la_mode;
4348 spin_lock(&lad->lad_lock);
4349 if (lad->lad_assistant_status < 0 ||
4350 unlikely(!thread_is_running(&lfsck->li_thread) ||
4351 !thread_is_running(&lad->lad_thread))) {
4352 spin_unlock(&lad->lad_lock);
4353 lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar);
4354 ns->ln_striped_dirs_skipped++;
4359 list_add_tail(&lnr->lnr_lar.lar_list, &lad->lad_req_list);
4360 if (lad->lad_prefetched == 0)
4363 lad->lad_prefetched++;
4364 spin_unlock(&lad->lad_lock);
4366 wake_up(&lad->lad_thread.t_ctl_waitq);
4371 static int lfsck_namespace_open_dir(const struct lu_env *env,
4372 struct lfsck_component *com)
4374 struct lfsck_instance *lfsck = com->lc_lfsck;
4375 struct lfsck_namespace *ns = com->lc_file_ram;
4376 struct lfsck_lmv *llmv = lfsck->li_lmv;
4383 if (llmv->ll_lmv_master) {
4384 struct lmv_mds_md_v1 *lmv = &llmv->ll_lmv;
4386 if (lmv->lmv_master_mdt_index != lfsck_dev_idx(lfsck)) {
4387 lmv->lmv_master_mdt_index =
4388 lfsck_dev_idx(lfsck);
4389 ns->ln_flags |= LF_INCONSISTENT;
4390 llmv->ll_lmv_updated = 1;
4393 rc = lfsck_namespace_verify_stripe_slave(env, com,
4394 lfsck->li_obj_dir, llmv);
4397 RETURN(rc > 0 ? 0 : rc);
4400 static int lfsck_namespace_checkpoint(const struct lu_env *env,
4401 struct lfsck_component *com, bool init)
4403 struct lfsck_instance *lfsck = com->lc_lfsck;
4404 struct lfsck_namespace *ns = com->lc_file_ram;
4408 rc = lfsck_checkpoint_generic(env, com);
4413 down_write(&com->lc_sem);
4415 ns->ln_pos_latest_start = lfsck->li_pos_checkpoint;
4417 ns->ln_pos_last_checkpoint = lfsck->li_pos_checkpoint;
4418 ns->ln_run_time_phase1 += ktime_get_seconds() -
4419 lfsck->li_time_last_checkpoint;
4420 ns->ln_time_last_checkpoint = ktime_get_real_seconds();
4421 ns->ln_items_checked += com->lc_new_checked;
4422 com->lc_new_checked = 0;
4425 rc = lfsck_namespace_store(env, com);
4426 up_write(&com->lc_sem);
4429 CDEBUG(D_LFSCK, "%s: namespace LFSCK checkpoint at the pos [%llu"
4430 ", "DFID", %#llx], status = %d: rc = %d\n",
4431 lfsck_lfsck2name(lfsck), lfsck->li_pos_current.lp_oit_cookie,
4432 PFID(&lfsck->li_pos_current.lp_dir_parent),
4433 lfsck->li_pos_current.lp_dir_cookie, ns->ln_status, rc);
4435 return rc > 0 ? 0 : rc;
4438 static int lfsck_namespace_prep(const struct lu_env *env,
4439 struct lfsck_component *com,
4440 struct lfsck_start_param *lsp)
4442 struct lfsck_instance *lfsck = com->lc_lfsck;
4443 struct lfsck_namespace *ns = com->lc_file_ram;
4444 struct lfsck_position *pos = &com->lc_pos_start;
4447 rc = lfsck_namespace_load_bitmap(env, com);
4448 if (rc != 0 || ns->ln_status == LS_COMPLETED) {
4449 rc = lfsck_namespace_reset(env, com, false);
4451 rc = lfsck_set_param(env, lfsck, lsp->lsp_start, true);
4454 CDEBUG(D_LFSCK, "%s: namespace LFSCK prep failed: "
4455 "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
4461 down_write(&com->lc_sem);
4462 ns->ln_time_latest_start = ktime_get_real_seconds();
4463 spin_lock(&lfsck->li_lock);
4465 if (ns->ln_flags & LF_SCANNED_ONCE) {
4466 if (!lfsck->li_drop_dryrun ||
4467 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent)) {
4468 ns->ln_status = LS_SCANNING_PHASE2;
4469 list_move_tail(&com->lc_link,
4470 &lfsck->li_list_double_scan);
4471 if (!list_empty(&com->lc_link_dir))
4472 list_del_init(&com->lc_link_dir);
4473 lfsck_pos_set_zero(pos);
4475 ns->ln_status = LS_SCANNING_PHASE1;
4476 ns->ln_run_time_phase1 = 0;
4477 ns->ln_run_time_phase2 = 0;
4478 ns->ln_items_checked = 0;
4479 ns->ln_items_repaired = 0;
4480 ns->ln_items_failed = 0;
4481 ns->ln_dirs_checked = 0;
4482 ns->ln_objs_checked_phase2 = 0;
4483 ns->ln_objs_repaired_phase2 = 0;
4484 ns->ln_objs_failed_phase2 = 0;
4485 ns->ln_objs_nlink_repaired = 0;
4486 ns->ln_dirent_repaired = 0;
4487 ns->ln_linkea_repaired = 0;
4488 ns->ln_mul_linked_checked = 0;
4489 ns->ln_mul_linked_repaired = 0;
4490 ns->ln_unknown_inconsistency = 0;
4491 ns->ln_unmatched_pairs_repaired = 0;
4492 ns->ln_dangling_repaired = 0;
4493 ns->ln_mul_ref_repaired = 0;
4494 ns->ln_bad_type_repaired = 0;
4495 ns->ln_lost_dirent_repaired = 0;
4496 ns->ln_striped_dirs_scanned = 0;
4497 ns->ln_striped_dirs_repaired = 0;
4498 ns->ln_striped_dirs_failed = 0;
4499 ns->ln_striped_dirs_disabled = 0;
4500 ns->ln_striped_dirs_skipped = 0;
4501 ns->ln_striped_shards_scanned = 0;
4502 ns->ln_striped_shards_repaired = 0;
4503 ns->ln_striped_shards_failed = 0;
4504 ns->ln_striped_shards_skipped = 0;
4505 ns->ln_name_hash_repaired = 0;
4506 fid_zero(&ns->ln_fid_latest_scanned_phase2);
4507 if (list_empty(&com->lc_link_dir))
4508 list_add_tail(&com->lc_link_dir,
4509 &lfsck->li_list_dir);
4510 *pos = ns->ln_pos_first_inconsistent;
4513 ns->ln_status = LS_SCANNING_PHASE1;
4514 if (list_empty(&com->lc_link_dir))
4515 list_add_tail(&com->lc_link_dir,
4516 &lfsck->li_list_dir);
4517 if (!lfsck->li_drop_dryrun ||
4518 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent)) {
4519 *pos = ns->ln_pos_last_checkpoint;
4520 pos->lp_oit_cookie++;
4522 *pos = ns->ln_pos_first_inconsistent;
4526 spin_unlock(&lfsck->li_lock);
4527 up_write(&com->lc_sem);
4529 rc = lfsck_start_assistant(env, com, lsp);
4531 CDEBUG(D_LFSCK, "%s: namespace LFSCK prep done, start pos [%llu, "
4532 DFID", %#llx]: rc = %d\n",
4533 lfsck_lfsck2name(lfsck), pos->lp_oit_cookie,
4534 PFID(&pos->lp_dir_parent), pos->lp_dir_cookie, rc);
4539 static int lfsck_namespace_exec_oit(const struct lu_env *env,
4540 struct lfsck_component *com,
4541 struct dt_object *obj)
4543 struct lfsck_thread_info *info = lfsck_env_info(env);
4544 struct lfsck_namespace *ns = com->lc_file_ram;
4545 struct lfsck_instance *lfsck = com->lc_lfsck;
4546 const struct lu_fid *fid = lfsck_dto2fid(obj);
4547 struct lu_fid *pfid = &info->lti_fid2;
4548 struct lu_name *cname = &info->lti_name;
4549 struct lu_seq_range *range = &info->lti_range;
4550 struct seq_server_site *ss = lfsck_dev_site(lfsck);
4551 struct linkea_data ldata = { NULL };
4552 __u32 idx = lfsck_dev_idx(lfsck);
4553 struct lu_attr la = { .la_valid = 0 };
4554 bool remote = false;
4558 rc = dt_attr_get(env, obj, &la);
4559 if (unlikely(rc || (la.la_valid & LA_FLAGS &&
4560 la.la_flags & LUSTRE_ORPHAN_FL))) {
4562 "%s: skip orphan "DFID", %llx/%x: rc = %d\n",
4563 lfsck_lfsck2name(lfsck), PFID(fid),
4564 la.la_valid, la.la_flags, rc);
4569 rc = lfsck_links_read(env, obj, &ldata);
4573 /* -EINVAL means crashed linkEA, should be verified. */
4574 if (rc == -EINVAL) {
4575 rc = lfsck_namespace_trace_update(env, com, fid,
4576 LNTF_CHECK_LINKEA, true);
4578 struct lustre_handle lh = { 0 };
4580 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
4581 MDS_INODELOCK_UPDATE |
4582 MDS_INODELOCK_XATTR, LCK_EX);
4584 rc = lfsck_namespace_links_remove(env, com,
4586 lfsck_ibits_unlock(&lh, LCK_EX);
4590 GOTO(out, rc = (rc == -ENOENT ? 0 : rc));
4593 if (rc && rc != -ENODATA)
4596 if (rc == -ENODATA || unlikely(!ldata.ld_leh->leh_reccount)) {
4597 rc = lfsck_namespace_check_for_double_scan(env, com, obj);
4602 linkea_first_entry(&ldata);
4603 while (ldata.ld_lee != NULL) {
4604 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
4606 if (!fid_is_sane(pfid)) {
4607 rc = lfsck_namespace_trace_update(env, com, fid,
4608 LNTF_CHECK_PARENT, true);
4609 } else if (!linkea_entry_is_valid(&ldata, cname, pfid)) {
4612 fld_range_set_mdt(range);
4613 rc = fld_server_lookup(env, ss->ss_server_fld,
4614 fid_seq(pfid), range);
4615 if ((rc == -ENOENT) ||
4616 (!rc && range->lsr_index != idx)) {
4624 linkea_next_entry(&ldata);
4627 if ((lu_object_has_agent_entry(&obj->do_lu) && !remote) ||
4628 (!lu_object_has_agent_entry(&obj->do_lu) && remote)) {
4629 rc = lfsck_namespace_trace_update(env, com, fid,
4630 LNTF_CHECK_AGENT_ENTRY, true);
4635 /* Record multiple-linked object. */
4636 if (ldata.ld_leh->leh_reccount > 1) {
4637 rc = lfsck_namespace_trace_update(env, com, fid,
4638 LNTF_CHECK_LINKEA, true);
4644 rc = lfsck_namespace_trace_update(env, com, fid,
4645 LNTF_CHECK_LINKEA, true);
4647 rc = lfsck_namespace_check_for_double_scan(env, com, obj);
4652 down_write(&com->lc_sem);
4653 if (S_ISDIR(lfsck_object_type(obj)))
4654 ns->ln_dirs_checked++;
4656 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
4657 up_write(&com->lc_sem);
4662 static int lfsck_namespace_exec_dir(const struct lu_env *env,
4663 struct lfsck_component *com,
4664 struct lfsck_assistant_object *lso,
4665 struct lu_dirent *ent, __u16 type)
4667 struct lfsck_assistant_data *lad = com->lc_data;
4668 struct lfsck_instance *lfsck = com->lc_lfsck;
4669 struct lfsck_namespace_req *lnr;
4670 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4671 struct ptlrpc_thread *mthread = &lfsck->li_thread;
4672 struct ptlrpc_thread *athread = &lad->lad_thread;
4673 bool wakeup = false;
4675 wait_event_idle(mthread->t_ctl_waitq,
4676 lad->lad_prefetched < bk->lb_async_windows ||
4677 !thread_is_running(mthread) ||
4678 !thread_is_running(athread));
4680 if (unlikely(!thread_is_running(mthread) ||
4681 !thread_is_running(athread)))
4684 if (unlikely(lfsck_is_dead_obj(lfsck->li_obj_dir)))
4687 lnr = lfsck_namespace_assistant_req_init(com->lc_lfsck, lso, ent, type);
4689 struct lfsck_namespace *ns = com->lc_file_ram;
4691 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
4692 return PTR_ERR(lnr);
4695 spin_lock(&lad->lad_lock);
4696 if (lad->lad_assistant_status < 0 ||
4697 unlikely(!thread_is_running(mthread) ||
4698 !thread_is_running(athread))) {
4699 spin_unlock(&lad->lad_lock);
4700 lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar);
4701 return lad->lad_assistant_status;
4704 list_add_tail(&lnr->lnr_lar.lar_list, &lad->lad_req_list);
4705 if (lad->lad_prefetched == 0)
4708 lad->lad_prefetched++;
4709 spin_unlock(&lad->lad_lock);
4711 wake_up(&lad->lad_thread.t_ctl_waitq);
4713 down_write(&com->lc_sem);
4714 com->lc_new_checked++;
4715 up_write(&com->lc_sem);
4720 static int lfsck_namespace_post(const struct lu_env *env,
4721 struct lfsck_component *com,
4722 int result, bool init)
4724 struct lfsck_instance *lfsck = com->lc_lfsck;
4725 struct lfsck_namespace *ns = com->lc_file_ram;
4729 lfsck_post_generic(env, com, &result);
4731 down_write(&com->lc_sem);
4732 lfsck_namespace_release_lmv(env, com);
4734 spin_lock(&lfsck->li_lock);
4736 ns->ln_pos_last_checkpoint = lfsck->li_pos_checkpoint;
4738 ns->ln_status = LS_SCANNING_PHASE2;
4739 ns->ln_flags |= LF_SCANNED_ONCE;
4740 ns->ln_flags &= ~LF_UPGRADE;
4741 list_del_init(&com->lc_link_dir);
4742 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
4743 } else if (result == 0) {
4744 if (lfsck->li_status != 0)
4745 ns->ln_status = lfsck->li_status;
4747 ns->ln_status = LS_STOPPED;
4748 if (ns->ln_status != LS_PAUSED) {
4749 list_del_init(&com->lc_link_dir);
4750 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4753 ns->ln_status = LS_FAILED;
4754 list_del_init(&com->lc_link_dir);
4755 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4757 spin_unlock(&lfsck->li_lock);
4760 ns->ln_run_time_phase1 += ktime_get_seconds() -
4761 lfsck->li_time_last_checkpoint;
4762 ns->ln_time_last_checkpoint = ktime_get_real_seconds();
4763 ns->ln_items_checked += com->lc_new_checked;
4764 com->lc_new_checked = 0;
4767 rc = lfsck_namespace_store(env, com);
4768 up_write(&com->lc_sem);
4770 CDEBUG(D_LFSCK, "%s: namespace LFSCK post done: rc = %d\n",
4771 lfsck_lfsck2name(lfsck), rc);
4777 lfsck_namespace_dump(const struct lu_env *env, struct lfsck_component *com,
4780 struct lfsck_instance *lfsck = com->lc_lfsck;
4781 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4782 struct lfsck_namespace *ns = com->lc_file_ram;
4784 down_read(&com->lc_sem);
4785 seq_printf(m, "name: lfsck_namespace\n"
4791 lfsck_status2name(ns->ln_status));
4793 lfsck_bits_dump(m, ns->ln_flags, lfsck_flags_names, "flags");
4795 lfsck_bits_dump(m, bk->lb_param, lfsck_param_names, "param");
4797 lfsck_time_dump(m, ns->ln_time_last_complete, "last_completed");
4799 lfsck_time_dump(m, ns->ln_time_latest_start, "latest_start");
4801 lfsck_time_dump(m, ns->ln_time_last_checkpoint, "last_checkpoint");
4803 lfsck_pos_dump(m, &ns->ln_pos_latest_start, "latest_start_position");
4805 lfsck_pos_dump(m, &ns->ln_pos_last_checkpoint,
4806 "last_checkpoint_position");
4808 lfsck_pos_dump(m, &ns->ln_pos_first_inconsistent,
4809 "first_failure_position");
4811 if (ns->ln_status == LS_SCANNING_PHASE1) {
4812 struct lfsck_position pos;
4813 time64_t duration = ktime_get_seconds() -
4814 lfsck->li_time_last_checkpoint;
4815 u64 checked = ns->ln_items_checked + com->lc_new_checked;
4816 u64 speed = checked;
4817 u64 new_checked = com->lc_new_checked;
4818 time64_t rtime = ns->ln_run_time_phase1 + duration;
4821 new_checked = div64_s64(new_checked, duration);
4824 speed = div64_s64(speed, rtime);
4826 lfsck_namespace_dump_statistics(m, ns, checked, 0, rtime, 0,
4827 bk->lb_param & LPF_DRYRUN);
4828 seq_printf(m, "average_speed_phase1: %llu items/sec\n"
4829 "average_speed_phase2: N/A\n"
4830 "average_speed_total: %llu items/sec\n"
4831 "real_time_speed_phase1: %llu items/sec\n"
4832 "real_time_speed_phase2: N/A\n",
4837 if (likely(lfsck->li_di_oit)) {
4838 const struct dt_it_ops *iops =
4839 &lfsck->li_obj_oit->do_index_ops->dio_it;
4841 /* The low layer otable-based iteration position may NOT
4842 * exactly match the namespace-based directory traversal
4843 * cookie. Generally, it is not a serious issue. But the
4844 * caller should NOT make assumption on that. */
4845 pos.lp_oit_cookie = iops->store(env, lfsck->li_di_oit);
4846 if (!lfsck->li_current_oit_processed)
4847 pos.lp_oit_cookie--;
4849 spin_lock(&lfsck->li_lock);
4850 if (lfsck->li_di_dir) {
4851 pos.lp_dir_cookie = lfsck->li_cookie_dir;
4852 if (pos.lp_dir_cookie >= MDS_DIR_END_OFF) {
4853 fid_zero(&pos.lp_dir_parent);
4854 pos.lp_dir_cookie = 0;
4857 *lfsck_dto2fid(lfsck->li_obj_dir);
4860 fid_zero(&pos.lp_dir_parent);
4861 pos.lp_dir_cookie = 0;
4863 spin_unlock(&lfsck->li_lock);
4865 pos = ns->ln_pos_last_checkpoint;
4868 lfsck_pos_dump(m, &pos, "current_position");
4869 } else if (ns->ln_status == LS_SCANNING_PHASE2) {
4870 time64_t duration = ktime_get_seconds() -
4871 com->lc_time_last_checkpoint;
4872 __u64 checked = ns->ln_objs_checked_phase2 +
4873 com->lc_new_checked;
4874 __u64 speed1 = ns->ln_items_checked;
4875 __u64 speed2 = checked;
4876 __u64 speed0 = speed1 + speed2;
4877 __u64 new_checked = com->lc_new_checked;
4878 time64_t rtime = ns->ln_run_time_phase2 + duration;
4879 time64_t time0 = ns->ln_run_time_phase1 + rtime;
4882 new_checked = div64_s64(new_checked, duration);
4884 if (ns->ln_run_time_phase1 != 0)
4885 speed1 = div64_s64(speed1, ns->ln_run_time_phase1);
4886 else if (ns->ln_items_checked != 0)
4890 speed2 = div64_s64(speed2, rtime);
4891 else if (checked != 0)
4895 speed0 = div64_s64(speed0, time0);
4897 lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked,
4899 ns->ln_run_time_phase1, rtime,
4900 bk->lb_param & LPF_DRYRUN);
4901 seq_printf(m, "average_speed_phase1: %llu items/sec\n"
4902 "average_speed_phase2: %llu objs/sec\n"
4903 "average_speed_total: %llu items/sec\n"
4904 "real_time_speed_phase1: N/A\n"
4905 "real_time_speed_phase2: %llu objs/sec\n"
4906 "current_position: "DFID"\n",
4911 PFID(&ns->ln_fid_latest_scanned_phase2));
4913 __u64 speed1 = ns->ln_items_checked;
4914 __u64 speed2 = ns->ln_objs_checked_phase2;
4915 __u64 speed0 = speed1 + speed2;
4916 time64_t time0 = ns->ln_run_time_phase1 + ns->ln_run_time_phase2;
4918 if (ns->ln_run_time_phase1 != 0)
4919 speed1 = div64_s64(speed1, ns->ln_run_time_phase1);
4920 else if (ns->ln_items_checked != 0)
4923 if (ns->ln_run_time_phase2 != 0)
4924 speed2 = div64_s64(speed2, ns->ln_run_time_phase2);
4925 else if (ns->ln_objs_checked_phase2 != 0)
4929 speed0 = div64_s64(speed0, time0);
4931 lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked,
4932 ns->ln_objs_checked_phase2,
4933 ns->ln_run_time_phase1,
4934 ns->ln_run_time_phase2,
4935 bk->lb_param & LPF_DRYRUN);
4936 seq_printf(m, "average_speed_phase1: %llu items/sec\n"
4937 "average_speed_phase2: %llu objs/sec\n"
4938 "average_speed_total: %llu items/sec\n"
4939 "real_time_speed_phase1: N/A\n"
4940 "real_time_speed_phase2: N/A\n"
4941 "current_position: N/A\n",
4947 up_read(&com->lc_sem);
4950 static int lfsck_namespace_double_scan(const struct lu_env *env,
4951 struct lfsck_component *com)
4953 struct lfsck_namespace *ns = com->lc_file_ram;
4954 struct lfsck_assistant_data *lad = com->lc_data;
4955 struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
4956 struct lfsck_tgt_desc *ltd;
4957 struct lfsck_tgt_desc *next;
4960 rc = lfsck_double_scan_generic(env, com, ns->ln_status);
4961 if (thread_is_stopped(&lad->lad_thread)) {
4962 LASSERT(list_empty(&lad->lad_req_list));
4963 LASSERT(list_empty(&lad->lad_mdt_phase1_list));
4965 spin_lock(<ds->ltd_lock);
4966 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4967 ltd_namespace_phase_list) {
4968 list_del_init(<d->ltd_namespace_phase_list);
4970 spin_unlock(<ds->ltd_lock);
4976 static void lfsck_namespace_data_release(const struct lu_env *env,
4977 struct lfsck_component *com)
4979 struct lfsck_assistant_data *lad = com->lc_data;
4980 struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
4981 struct lfsck_tgt_desc *ltd;
4982 struct lfsck_tgt_desc *next;
4984 LASSERT(lad != NULL);
4985 LASSERT(thread_is_init(&lad->lad_thread) ||
4986 thread_is_stopped(&lad->lad_thread));
4987 LASSERT(list_empty(&lad->lad_req_list));
4989 com->lc_data = NULL;
4990 lfsck_namespace_release_lmv(env, com);
4992 spin_lock(<ds->ltd_lock);
4993 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
4994 ltd_namespace_phase_list) {
4995 list_del_init(<d->ltd_namespace_phase_list);
4997 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4998 ltd_namespace_phase_list) {
4999 list_del_init(<d->ltd_namespace_phase_list);
5001 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_list,
5002 ltd_namespace_list) {
5003 list_del_init(<d->ltd_namespace_list);
5005 spin_unlock(<ds->ltd_lock);
5007 if (likely(lad->lad_bitmap != NULL))
5008 CFS_FREE_BITMAP(lad->lad_bitmap);
5013 static void lfsck_namespace_quit(const struct lu_env *env,
5014 struct lfsck_component *com)
5016 struct lfsck_assistant_data *lad = com->lc_data;
5017 struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
5018 struct lfsck_tgt_desc *ltd;
5019 struct lfsck_tgt_desc *next;
5021 LASSERT(lad != NULL);
5023 lfsck_quit_generic(env, com);
5025 LASSERT(thread_is_init(&lad->lad_thread) ||
5026 thread_is_stopped(&lad->lad_thread));
5027 LASSERT(list_empty(&lad->lad_req_list));
5029 lfsck_namespace_release_lmv(env, com);
5031 spin_lock(<ds->ltd_lock);
5032 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
5033 ltd_namespace_phase_list) {
5034 list_del_init(<d->ltd_namespace_phase_list);
5036 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
5037 ltd_namespace_phase_list) {
5038 list_del_init(<d->ltd_namespace_phase_list);
5040 spin_unlock(<ds->ltd_lock);
5043 static int lfsck_namespace_in_notify(const struct lu_env *env,
5044 struct lfsck_component *com,
5045 struct lfsck_request *lr)
5047 struct lfsck_instance *lfsck = com->lc_lfsck;
5048 struct lfsck_namespace *ns = com->lc_file_ram;
5049 struct lfsck_assistant_data *lad = com->lc_data;
5050 struct lfsck_tgt_descs *ltds = &lfsck->li_mdt_descs;
5051 struct lfsck_tgt_desc *ltd;
5056 switch (lr->lr_event) {
5057 case LE_SET_LMV_MASTER: {
5058 struct dt_object *obj;
5060 obj = lfsck_object_find_bottom(env, lfsck, &lr->lr_fid);
5062 RETURN(PTR_ERR(obj));
5064 if (likely(dt_object_exists(obj)))
5065 rc = lfsck_namespace_notify_lmv_master_local(env, com,
5068 lfsck_object_put(env, obj);
5070 RETURN(rc > 0 ? 0 : rc);
5072 case LE_SET_LMV_SLAVE: {
5073 if (!(lr->lr_flags & LEF_RECHECK_NAME_HASH))
5074 ns->ln_striped_shards_repaired++;
5076 rc = lfsck_namespace_trace_update(env, com, &lr->lr_fid,
5077 LNTF_RECHECK_NAME_HASH, true);
5079 RETURN(rc > 0 ? 0 : rc);
5081 case LE_PHASE1_DONE:
5082 case LE_PHASE2_DONE:
5089 CDEBUG(D_LFSCK, "%s: namespace LFSCK handles notify %u from MDT %x, "
5090 "status %d, flags %x\n", lfsck_lfsck2name(lfsck), lr->lr_event,
5091 lr->lr_index, lr->lr_status, lr->lr_flags2);
5093 spin_lock(<ds->ltd_lock);
5094 ltd = lfsck_ltd2tgt(ltds, lr->lr_index);
5096 spin_unlock(<ds->ltd_lock);
5101 list_del_init(<d->ltd_namespace_phase_list);
5102 switch (lr->lr_event) {
5103 case LE_PHASE1_DONE:
5104 if (lr->lr_status <= 0) {
5105 ltd->ltd_namespace_done = 1;
5106 list_del_init(<d->ltd_namespace_list);
5107 CDEBUG(D_LFSCK, "%s: MDT %x failed/stopped at "
5108 "phase1 for namespace LFSCK: rc = %d.\n",
5109 lfsck_lfsck2name(lfsck),
5110 ltd->ltd_index, lr->lr_status);
5111 ns->ln_flags |= LF_INCOMPLETE;
5116 if (lr->lr_flags2 & LF_INCOMPLETE)
5117 ns->ln_flags |= LF_INCOMPLETE;
5119 if (list_empty(<d->ltd_namespace_list))
5120 list_add_tail(<d->ltd_namespace_list,
5121 &lad->lad_mdt_list);
5122 list_add_tail(<d->ltd_namespace_phase_list,
5123 &lad->lad_mdt_phase2_list);
5125 case LE_PHASE2_DONE:
5126 ltd->ltd_namespace_done = 1;
5127 list_del_init(<d->ltd_namespace_list);
5131 ltd->ltd_namespace_done = 1;
5132 list_del_init(<d->ltd_namespace_list);
5133 if (!(lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT)) {
5135 "%s: the peer MDT %x exit namespace LFSCK\n",
5136 lfsck_lfsck2name(lfsck), ltd->ltd_index);
5137 ns->ln_flags |= LF_INCOMPLETE;
5143 spin_unlock(<ds->ltd_lock);
5145 if (fail && lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) {
5146 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
5148 memset(stop, 0, sizeof(*stop));
5149 stop->ls_status = lr->lr_status;
5150 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
5151 lfsck_stop(env, lfsck->li_bottom, stop);
5152 } else if (lfsck_phase2_next_ready(lad)) {
5153 wake_up(&lad->lad_thread.t_ctl_waitq);
5159 static void lfsck_namespace_repaired(struct lfsck_namespace *ns, __u64 *count)
5161 *count += ns->ln_objs_nlink_repaired;
5162 *count += ns->ln_dirent_repaired;
5163 *count += ns->ln_linkea_repaired;
5164 *count += ns->ln_mul_linked_repaired;
5165 *count += ns->ln_unmatched_pairs_repaired;
5166 *count += ns->ln_dangling_repaired;
5167 *count += ns->ln_mul_ref_repaired;
5168 *count += ns->ln_bad_type_repaired;
5169 *count += ns->ln_lost_dirent_repaired;
5170 *count += ns->ln_striped_dirs_disabled;
5171 *count += ns->ln_striped_dirs_repaired;
5172 *count += ns->ln_striped_shards_repaired;
5173 *count += ns->ln_name_hash_repaired;
5174 *count += ns->ln_local_lpf_moved;
5177 static int lfsck_namespace_query_all(const struct lu_env *env,
5178 struct lfsck_component *com,
5179 __u32 *mdts_count, __u64 *repaired)
5181 struct lfsck_namespace *ns = com->lc_file_ram;
5182 struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
5183 struct lfsck_tgt_desc *ltd;
5188 rc = lfsck_query_all(env, com);
5192 down_read(<ds->ltd_rw_sem);
5193 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
5194 ltd = lfsck_ltd2tgt(ltds, idx);
5195 LASSERT(ltd != NULL);
5197 mdts_count[ltd->ltd_namespace_status]++;
5198 *repaired += ltd->ltd_namespace_repaired;
5200 up_read(<ds->ltd_rw_sem);
5202 down_read(&com->lc_sem);
5203 mdts_count[ns->ln_status]++;
5204 lfsck_namespace_repaired(ns, repaired);
5205 up_read(&com->lc_sem);
5210 static int lfsck_namespace_query(const struct lu_env *env,
5211 struct lfsck_component *com,
5212 struct lfsck_request *req,
5213 struct lfsck_reply *rep,
5214 struct lfsck_query *que, int idx)
5216 struct lfsck_namespace *ns = com->lc_file_ram;
5220 LASSERT(com->lc_lfsck->li_master);
5222 rc = lfsck_namespace_query_all(env, com,
5223 que->lu_mdts_count[idx],
5224 &que->lu_repaired[idx]);
5226 down_read(&com->lc_sem);
5227 rep->lr_status = ns->ln_status;
5228 if (req->lr_flags & LEF_QUERY_ALL)
5229 lfsck_namespace_repaired(ns, &rep->lr_repaired);
5230 up_read(&com->lc_sem);
5236 static const struct lfsck_operations lfsck_namespace_ops = {
5237 .lfsck_reset = lfsck_namespace_reset,
5238 .lfsck_fail = lfsck_namespace_fail,
5239 .lfsck_close_dir = lfsck_namespace_close_dir,
5240 .lfsck_open_dir = lfsck_namespace_open_dir,
5241 .lfsck_checkpoint = lfsck_namespace_checkpoint,
5242 .lfsck_prep = lfsck_namespace_prep,
5243 .lfsck_exec_oit = lfsck_namespace_exec_oit,
5244 .lfsck_exec_dir = lfsck_namespace_exec_dir,
5245 .lfsck_post = lfsck_namespace_post,
5246 .lfsck_dump = lfsck_namespace_dump,
5247 .lfsck_double_scan = lfsck_namespace_double_scan,
5248 .lfsck_data_release = lfsck_namespace_data_release,
5249 .lfsck_quit = lfsck_namespace_quit,
5250 .lfsck_in_notify = lfsck_namespace_in_notify,
5251 .lfsck_query = lfsck_namespace_query,
5255 * Repair dangling name entry.
5257 * For the name entry with dangling reference, we need to repare the
5258 * inconsistency according to the LFSCK sponsor's requirement:
5260 * 1) Keep the inconsistency there and report the inconsistency case,
5261 * then give the chance to the application to find related issues,
5262 * and the users can make the decision about how to handle it with
5263 * more human knownledge. (by default)
5265 * 2) Re-create the missing MDT-object with the FID information.
5267 * \param[in] env pointer to the thread context
5268 * \param[in] com pointer to the lfsck component
5269 * \param[in] parent pointer to the dir object that contains the dangling
5271 * \param[in] child pointer to the object corresponding to the dangling
5273 * \param[in] lnr pointer to the namespace request that contains the
5274 * name's name, parent object, parent's LMV, and ect.
5276 * \retval positive number if no need to repair
5277 * \retval zero for repaired successfully
5278 * \retval negative error number on failure
5280 int lfsck_namespace_repair_dangling(const struct lu_env *env,
5281 struct lfsck_component *com,
5282 struct dt_object *parent,
5283 struct dt_object *child,
5284 struct lfsck_namespace_req *lnr)
5286 struct lfsck_thread_info *info = lfsck_env_info(env);
5287 struct lu_attr *la = &info->lti_la;
5288 struct dt_allocation_hint *hint = &info->lti_hint;
5289 struct dt_object_format *dof = &info->lti_dof;
5290 struct dt_insert_rec *rec = &info->lti_dt_rec;
5291 struct lmv_mds_md_v1 *lmv2 = &info->lti_lmv2;
5292 const struct lu_name *cname;
5293 const struct lu_fid *pfid = lfsck_dto2fid(parent);
5294 const struct lu_fid *cfid = lfsck_dto2fid(child);
5295 struct linkea_data ldata = { NULL };
5296 struct lfsck_lock_handle *llh = &info->lti_llh;
5297 struct lustre_handle rlh = { 0 };
5298 struct lustre_handle clh = { 0 };
5299 struct lu_buf linkea_buf;
5300 struct lu_buf lmv_buf;
5301 struct lfsck_instance *lfsck = com->lc_lfsck;
5302 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5303 struct dt_device *dev = lfsck->li_next;
5304 struct thandle *th = NULL;
5306 __u16 type = lnr->lnr_type;
5310 cname = lfsck_name_get_const(env, lnr->lnr_name, lnr->lnr_namelen);
5311 if (bk->lb_param & LPF_CREATE_MDTOBJ)
5316 if (!create || bk->lb_param & LPF_DRYRUN)
5319 /* We may need to create the sub-objects of the @child via LOD,
5320 * so make the modification based on lfsck->li_next. */
5322 parent = lfsck_object_locate(dev, parent);
5324 GOTO(log, rc = PTR_ERR(parent));
5326 if (unlikely(!dt_try_as_dir(env, parent, true)))
5327 GOTO(log, rc = -ENOTDIR);
5329 child = lfsck_object_locate(dev, child);
5331 GOTO(log, rc = PTR_ERR(child));
5333 rc = linkea_links_new(&ldata, &info->lti_linkea_buf2,
5338 rc = lfsck_lock(env, lfsck, parent, lnr->lnr_name, llh,
5339 MDS_INODELOCK_UPDATE, LCK_PW);
5343 rc = lfsck_namespace_check_exist(env, parent, child, lnr->lnr_name);
5347 if (dt_object_remote(child)) {
5348 rc = lfsck_remote_lookup_lock(env, lfsck, parent, child, &rlh,
5354 rc = lfsck_ibits_lock(env, lfsck, child, &clh,
5355 MDS_INODELOCK_UPDATE | MDS_INODELOCK_LOOKUP |
5356 MDS_INODELOCK_XATTR, LCK_EX);
5358 GOTO(unlock_remote_lookup, rc);
5360 /* Set the ctime as zero, then others can know it is created for
5361 * repairing dangling name entry by LFSCK. And if the LFSCK made
5362 * wrong decision and the real MDT-object has been found later,
5363 * then the LFSCK has chance to fix the incosistency properly. */
5364 memset(la, 0, sizeof(*la));
5366 la->la_mode = (type & S_IFMT) | 0700;
5368 la->la_mode = (type & S_IFMT) | 0600;
5369 la->la_valid = LA_TYPE | LA_MODE | LA_CTIME;
5372 * if it's directory, skip do_ah_init() to create a plain directory
5373 * because it may have shards already, which will be inserted back
5374 * later, besides, it may be remote, and creating stripe directory
5375 * remotely is not supported.
5378 child->do_ops->do_ah_init(env, hint, parent, child, type);
5379 else if (S_ISDIR(type))
5380 child->do_ops->do_ah_init(env, hint, NULL, child, type);
5382 memset(dof, 0, sizeof(*dof));
5383 dof->dof_type = dt_mode_to_dft(type);
5384 /* If the target is a regular file, then the LFSCK will only create
5385 * the MDT-object without stripes (dof->dof_reg.striped = 0). related
5386 * OST-objects will be created when write open. */
5388 th = lfsck_trans_create(env, dev, lfsck);
5390 GOTO(unlock_child, rc = PTR_ERR(th));
5392 /* 1a. create child. */
5393 rc = dt_declare_create(env, child, la, hint, dof, th);
5397 if (S_ISDIR(type)) {
5398 if (unlikely(!dt_try_as_dir(env, child, false)))
5399 GOTO(stop, rc = -ENOTDIR);
5401 /* 2a. increase child nlink */
5402 rc = dt_declare_ref_add(env, child, th);
5406 /* 3a. insert dot into child dir */
5407 rec->rec_type = S_IFDIR;
5408 rec->rec_fid = cfid;
5409 rc = dt_declare_insert(env, child,
5410 (const struct dt_rec *)rec,
5411 (const struct dt_key *)dot, th);
5415 /* 4a. insert dotdot into child dir */
5416 rec->rec_fid = pfid;
5417 rc = dt_declare_insert(env, child,
5418 (const struct dt_rec *)rec,
5419 (const struct dt_key *)dotdot, th);
5423 /* 5a. generate slave LMV EA. */
5424 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
5427 idx = lfsck_shard_name_to_index(env,
5428 lnr->lnr_name, lnr->lnr_namelen,
5430 if (unlikely(idx < 0))
5431 GOTO(stop, rc = idx);
5433 *lmv2 = lnr->lnr_lmv->ll_lmv;
5434 lmv2->lmv_magic = LMV_MAGIC_STRIPE;
5435 lmv2->lmv_master_mdt_index = idx;
5437 lfsck_lmv_header_cpu_to_le(lmv2, lmv2);
5438 lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
5439 rc = dt_declare_xattr_set(env, child, &lmv_buf,
5440 XATTR_NAME_LMV, 0, th);
5446 /* 6a. insert linkEA for child */
5447 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
5448 ldata.ld_leh->leh_len);
5449 rc = dt_declare_xattr_set(env, child, &linkea_buf,
5450 XATTR_NAME_LINK, 0, th);
5454 /* 7a. if child is remote, delete and insert to generate local agent */
5455 if (dt_object_remote(child)) {
5456 rc = dt_declare_delete(env, parent,
5457 (const struct dt_key *)lnr->lnr_name,
5462 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
5463 (const struct dt_key *)lnr->lnr_name,
5469 rc = dt_trans_start_local(env, dev, th);
5471 GOTO(stop, rc = (rc == -EEXIST ? 1 : rc));
5473 dt_write_lock(env, child, 0);
5474 /* 1b. create child */
5475 rc = dt_create(env, child, la, hint, dof, th);
5477 GOTO(unlock, rc = (rc == -EEXIST ? 1 : rc));
5479 if (S_ISDIR(type)) {
5480 /* 2b. increase child nlink */
5481 rc = dt_ref_add(env, child, th);
5485 /* 3b. insert dot into child dir */
5486 rec->rec_type = S_IFDIR;
5487 rec->rec_fid = cfid;
5488 rc = dt_insert(env, child, (const struct dt_rec *)rec,
5489 (const struct dt_key *)dot, th);
5493 /* 4b. insert dotdot into child dir */
5494 rec->rec_fid = pfid;
5495 rc = dt_insert(env, child, (const struct dt_rec *)rec,
5496 (const struct dt_key *)dotdot, th);
5500 /* 5b. generate slave LMV EA. */
5501 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
5502 rc = dt_xattr_set(env, child, &lmv_buf, XATTR_NAME_LMV,
5509 /* 6b. insert linkEA for child. */
5510 rc = dt_xattr_set(env, child, &linkea_buf,
5511 XATTR_NAME_LINK, 0, th);
5515 /* 7b. if child is remote, delete and insert to generate local agent */
5516 if (dt_object_remote(child)) {
5517 rc = dt_delete(env, parent,
5518 (const struct dt_key *)lnr->lnr_name, th);
5522 rec->rec_type = type;
5523 rec->rec_fid = cfid;
5524 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
5525 (const struct dt_key *)lnr->lnr_name, th);
5533 dt_write_unlock(env, child);
5536 dt_trans_stop(env, dev, th);
5539 lfsck_ibits_unlock(&clh, LCK_EX);
5540 unlock_remote_lookup:
5541 if (dt_object_remote(child))
5542 lfsck_ibits_unlock(&rlh, LCK_EX);
5545 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found dangling "
5546 "reference for: parent "DFID", child "DFID", type %u, "
5547 "name %s. %s: rc = %d\n", lfsck_lfsck2name(lfsck),
5548 PFID(pfid), PFID(cfid), type, cname->ln_name,
5549 create ? "Create the lost MDT-object as required" :
5550 "Keep the MDT-object there by default", rc);
5553 struct lfsck_namespace *ns = com->lc_file_ram;
5555 ns->ln_flags |= LF_INCONSISTENT;
5561 static int lfsck_namespace_assistant_handler_p1(const struct lu_env *env,
5562 struct lfsck_component *com,
5563 struct lfsck_assistant_req *lar)
5565 struct lfsck_thread_info *info = lfsck_env_info(env);
5566 struct lu_attr *la = &info->lti_la;
5567 struct lfsck_instance *lfsck = com->lc_lfsck;
5568 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5569 struct lfsck_namespace *ns = com->lc_file_ram;
5570 struct lfsck_assistant_data *lad = com->lc_data;
5571 struct linkea_data ldata = { NULL };
5572 const struct lu_name *cname;
5573 struct thandle *handle = NULL;
5574 struct lfsck_namespace_req *lnr =
5575 container_of(lar, struct lfsck_namespace_req, lnr_lar);
5576 struct dt_object *dir = NULL;
5577 struct dt_object *obj = NULL;
5578 struct lfsck_assistant_object *lso = lar->lar_parent;
5579 const struct lu_fid *pfid = &lso->lso_fid;
5580 struct dt_device *dev = NULL;
5581 struct lustre_handle lh = { 0 };
5582 bool repaired = false;
5583 bool dtlocked = false;
5584 bool remove = false;
5585 bool newdata = false;
5587 bool bad_hash = false;
5588 bool bad_linkea = false;
5592 enum lfsck_namespace_inconsistency_type type = LNIT_NONE;
5599 if (lnr->lnr_attr & LUDA_UPGRADE) {
5600 ns->ln_flags |= LF_UPGRADE;
5601 ns->ln_dirent_repaired++;
5603 } else if (lnr->lnr_attr & LUDA_REPAIR) {
5604 ns->ln_flags |= LF_INCONSISTENT;
5605 ns->ln_dirent_repaired++;
5609 if (unlikely(fid_is_zero(&lnr->lnr_fid) &&
5610 strcmp(lnr->lnr_name, dotdot) == 0)) {
5611 rc = lfsck_namespace_trace_update(env, com, pfid,
5612 LNTF_CHECK_PARENT, true);
5617 if (unlikely(!fid_is_sane(&lnr->lnr_fid))) {
5618 CDEBUG(D_LFSCK, "%s: dir scan find invalid FID "DFID
5619 " for the name entry %.*s under "DFID"\n",
5620 lfsck_lfsck2name(lfsck), PFID(&lnr->lnr_fid),
5621 lnr->lnr_namelen, lnr->lnr_name, PFID(pfid));
5623 if (strcmp(lnr->lnr_name, dotdot) != 0)
5624 /* invalid FID means bad name entry, remove it. */
5625 type = LNIT_BAD_DIRENT;
5627 /* If the parent FID is invalid, we cannot remove
5628 * the ".." entry directly. */
5629 rc = lfsck_namespace_trace_update(env, com, pfid,
5630 LNTF_CHECK_PARENT, true);
5635 if (unlikely(lnr->lnr_dir_cookie == MDS_DIR_END_OFF)) {
5636 rc = lfsck_namespace_striped_dir_rescan(env, com, lnr);
5641 if (fid_seq_is_dot(fid_seq(&lnr->lnr_fid)))
5644 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
5645 rc = lfsck_namespace_handle_striped_master(env, com, lnr);
5650 idx = lfsck_find_mdt_idx_by_fid(env, lfsck, &lnr->lnr_fid);
5652 GOTO(out, rc = idx);
5654 if (idx == lfsck_dev_idx(lfsck)) {
5655 if (unlikely(strcmp(lnr->lnr_name, dotdot) == 0))
5658 dev = lfsck->li_bottom;
5660 struct lfsck_tgt_desc *ltd;
5662 /* Usually, some local filesystem consistency verification
5663 * tools can guarantee the local namespace tree consistenct.
5664 * So the LFSCK will only verify the remote directory. */
5665 if (unlikely(strcmp(lnr->lnr_name, dotdot) == 0)) {
5666 rc = lfsck_namespace_trace_update(env, com, pfid,
5667 LNTF_CHECK_PARENT, true);
5672 ltd = lfsck_ltd2tgt(&lfsck->li_mdt_descs, idx);
5673 if (unlikely(ltd == NULL)) {
5674 CDEBUG(D_LFSCK, "%s: cannot talk with MDT %x which "
5675 "did not join the namespace LFSCK\n",
5676 lfsck_lfsck2name(lfsck), idx);
5677 lfsck_lad_set_bitmap(env, com, idx);
5679 GOTO(out, rc = -ENODEV);
5685 obj = lfsck_object_find_by_dev(env, dev, &lnr->lnr_fid);
5687 GOTO(out, rc = PTR_ERR(obj));
5689 cname = lfsck_name_get_const(env, lnr->lnr_name, lnr->lnr_namelen);
5690 if (dt_object_exists(obj) == 0) {
5694 dir = lfsck_assistant_object_load(env, lfsck, lso);
5698 GOTO(trace, rc == -ENOENT ? 0 : rc);
5702 rc = lfsck_namespace_check_exist(env, dir, obj, lnr->lnr_name);
5704 if (!lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5705 lnr->lnr_name, lnr->lnr_namelen)) {
5706 type = LNIT_BAD_DIRENT;
5711 type = LNIT_DANGLING;
5712 rc = lfsck_namespace_repair_dangling(env, com, dir,
5721 if (!(bk->lb_param & LPF_DRYRUN) && lad->lad_advance_lock) {
5724 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
5725 MDS_INODELOCK_UPDATE |
5726 MDS_INODELOCK_XATTR, LCK_EX);
5730 handle = lfsck_trans_create(env, dev, lfsck);
5732 GOTO(out, rc = PTR_ERR(handle));
5734 rc = lfsck_declare_namespace_exec_dir(env, obj, handle);
5738 rc = dt_trans_start_local(env, dev, handle);
5742 dt_write_lock(env, obj, 0);
5746 rc = lfsck_links_read(env, obj, &ldata);
5747 if (unlikely(rc == -ENOENT)) {
5748 if (handle != NULL) {
5749 dt_write_unlock(env, obj);
5752 dt_trans_stop(env, dev, handle);
5755 lfsck_ibits_unlock(&lh, LCK_EX);
5758 /* It may happen when the remote object has been removed,
5759 * but the local MDT is not aware of that. */
5761 } else if (rc == 0) {
5762 count = ldata.ld_leh->leh_reccount;
5763 rc = linkea_links_find(&ldata, cname, pfid);
5765 (count == 1 || !S_ISDIR(lfsck_object_type(obj)))) {
5766 if ((lfsck_object_type(obj) & S_IFMT) !=
5768 ns->ln_flags |= LF_INCONSISTENT;
5769 type = LNIT_BAD_TYPE;
5775 /* If the name entry hash does not match the slave striped
5776 * directory, and the name entry does not match also, then
5777 * it is quite possible that name entry is corrupted. */
5778 if (!lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5779 lnr->lnr_name, lnr->lnr_namelen)) {
5780 ns->ln_flags |= LF_INCONSISTENT;
5781 type = LNIT_BAD_DIRENT;
5786 /* If the file type stored in the name entry does not match
5787 * the file type claimed by the object, and the object does
5788 * not recognize the name entry, then it is quite possible
5789 * that the name entry is corrupted. */
5790 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type) {
5791 ns->ln_flags |= LF_INCONSISTENT;
5792 type = LNIT_BAD_DIRENT;
5797 /* For sub-dir object, we cannot make sure whether the sub-dir
5798 * back references the parent via ".." name entry correctly or
5799 * not in the LFSCK first-stage scanning. It may be that the
5800 * (remote) sub-dir ".." name entry has no parent FID after
5801 * file-level backup/restore and its linkEA may be wrong.
5802 * So under such case, we should replace the linkEA according
5803 * to current name entry. But this needs to be done during the
5804 * LFSCK second-stage scanning. The LFSCK will record the name
5805 * entry for further possible using. */
5809 } else if (unlikely(rc == -EINVAL)) {
5810 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type)
5811 type = LNIT_BAD_TYPE;
5814 /* The magic crashed, we are not sure whether there are more
5815 * corrupt data in the linkea, so remove all linkea entries. */
5819 } else if (rc == -ENODATA) {
5820 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type)
5821 type = LNIT_BAD_TYPE;
5828 if (rc == -ENOENT &&
5829 linkea_will_overflow(&ldata, cname)) {
5830 CDEBUG(D_INODE, "No enough space to hold linkea entry '"
5831 DFID": %.*s' at %u\n", PFID(pfid),
5832 cname->ln_namelen, cname->ln_name,
5833 ldata.ld_leh->leh_overflow_time);
5839 if (bk->lb_param & LPF_DRYRUN) {
5841 ns->ln_flags |= LF_UPGRADE;
5843 ns->ln_flags |= LF_INCONSISTENT;
5844 ns->ln_linkea_repaired++;
5850 if (!lustre_handle_is_used(&lh)) {
5858 LASSERT(handle != NULL);
5861 dir = lfsck_assistant_object_load(env, lfsck, lso);
5865 GOTO(stop, rc == -ENOENT ? 0 : rc);
5869 rc = lfsck_namespace_check_exist(env, dir, obj, lnr->lnr_name);
5874 if (!remove && newdata)
5875 ns->ln_flags |= LF_UPGRADE;
5876 else if (remove || !(ns->ln_flags & LF_UPGRADE))
5877 ns->ln_flags |= LF_INCONSISTENT;
5882 rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, handle);
5883 if (rc != 0 && rc != -ENOENT && rc != -ENODATA)
5888 rc = linkea_data_new(&ldata,
5889 &lfsck_env_info(env)->lti_linkea_buf);
5894 rc = linkea_add_buf(&ldata, cname, pfid, false);
5896 rc = lfsck_links_write(env, obj, &ldata, handle);
5900 count = ldata.ld_leh->leh_reccount;
5901 if (!S_ISDIR(lfsck_object_type(obj)) ||
5902 !dt_object_remote(obj)) {
5903 ns->ln_linkea_repaired++;
5913 dt_write_unlock(env, obj);
5915 if (handle != NULL && !IS_ERR(handle))
5916 dt_trans_stop(env, dev, handle);
5919 lfsck_ibits_unlock(&lh, LCK_EX);
5921 if (!name_is_dot_or_dotdot(lnr->lnr_name, lnr->lnr_namelen) &&
5922 !lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5923 lnr->lnr_name, lnr->lnr_namelen) &&
5924 type != LNIT_BAD_DIRENT) {
5925 ns->ln_flags |= LF_INCONSISTENT;
5929 dir = lfsck_assistant_object_load(env, lfsck, lso);
5933 GOTO(trace, rc == -ENOENT ? 0 : rc);
5937 rc = lfsck_namespace_repair_bad_name_hash(env, com, dir,
5938 lnr->lnr_lmv, lnr->lnr_name);
5944 if (type != LNIT_NONE && dir == NULL) {
5945 dir = lfsck_assistant_object_load(env, lfsck, lso);
5949 GOTO(trace, rc == -ENOENT ? 0 : rc);
5956 rc = lfsck_namespace_repair_dirent(env, com, dir,
5957 obj, lnr->lnr_name, lnr->lnr_name,
5958 lnr->lnr_type, true, false);
5962 case LNIT_BAD_DIRENT:
5964 /* XXX: This is a bad dirent, we do not know whether
5965 * the original name entry reference a regular
5966 * file or a directory, then keep the parent's
5967 * nlink count unchanged here. */
5968 rc = lfsck_namespace_repair_dirent(env, com, dir,
5969 obj, lnr->lnr_name, lnr->lnr_name,
5970 lnr->lnr_type, false, false);
5978 if (obj != NULL && count == 1 &&
5979 S_ISREG(lfsck_object_type(obj)))
5980 dt_attr_get(env, obj, la);
5982 /* if new linkea entry is added, the old entry may be stale,
5983 * check it in phase 2. Sigh, linkea check can only be done
5986 if (bad_linkea && !remove && !newdata &&
5987 !dt_object_remote(obj) && count > 1)
5988 rc = lfsck_namespace_trace_update(env, com,
5995 down_write(&com->lc_sem);
5997 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail to handle "
5998 "the entry: "DFID", parent "DFID", name %.*s: rc = %d\n",
5999 lfsck_lfsck2name(lfsck), PFID(&lnr->lnr_fid), PFID(pfid),
6000 lnr->lnr_namelen, lnr->lnr_name, rc);
6002 lfsck_namespace_record_failure(env, lfsck, ns);
6003 if ((rc == -ENOTCONN || rc == -ESHUTDOWN || rc == -EREMCHG ||
6004 rc == -ETIMEDOUT || rc == -EHOSTDOWN ||
6005 rc == -EHOSTUNREACH || rc == -EINPROGRESS) &&
6006 dev != NULL && dev != lfsck->li_bottom)
6007 lfsck_lad_set_bitmap(env, com, idx);
6009 if (!(bk->lb_param & LPF_FAILOUT))
6013 ns->ln_items_repaired++;
6015 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant "
6016 "repaired the entry: "DFID", parent "DFID
6017 ", name %.*s, type %d\n",
6018 lfsck_lfsck2name(lfsck),
6019 PFID(&lnr->lnr_fid), PFID(pfid),
6020 lnr->lnr_namelen, lnr->lnr_name, type);
6024 ns->ln_dangling_repaired++;
6027 ns->ln_bad_type_repaired++;
6029 case LNIT_BAD_DIRENT:
6030 ns->ln_dirent_repaired++;
6036 if (bk->lb_param & LPF_DRYRUN &&
6037 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent))
6038 lfsck_pos_fill(env, lfsck,
6039 &ns->ln_pos_first_inconsistent,
6044 ns->ln_name_hash_repaired++;
6046 /* Not count repeatedly. */
6048 ns->ln_items_repaired++;
6050 CDEBUG(D_LFSCK, "%s: namespace LFSCK "
6051 "assistant repaired the entry: "
6054 lfsck_lfsck2name(lfsck),
6055 PFID(&lnr->lnr_fid), PFID(pfid),
6056 lnr->lnr_namelen, lnr->lnr_name);
6059 if (bk->lb_param & LPF_DRYRUN &&
6060 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent))
6061 lfsck_pos_fill(env, lfsck,
6062 &ns->ln_pos_first_inconsistent,
6069 if (count > 1 || la->la_nlink > 1)
6070 ns->ln_mul_linked_checked++;
6072 up_write(&com->lc_sem);
6074 if (obj != NULL && !IS_ERR(obj))
6075 lfsck_object_put(env, obj);
6077 if (dir != NULL && !IS_ERR(dir))
6078 lfsck_object_put(env, dir);
6080 lad->lad_advance_lock = bad_linkea;
6086 * Handle one orphan under the backend /lost+found directory
6088 * Insert the orphan FID into the namespace LFSCK trace file for further
6089 * processing (via the subsequent namespace LFSCK second-stage scanning).
6090 * At the same time, remove the orphan name entry from backend /lost+found
6091 * directory. There is an interval between the orphan name entry removed
6092 * from the backend /lost+found directory and the orphan FID in the LFSCK
6093 * trace file handled. In such interval, the LFSCK can be reset, then
6094 * all the FIDs recorded in the namespace LFSCK trace file will be dropped.
6095 * To guarantee that the orphans can be found when LFSCK run next time
6096 * without e2fsck again, when remove the orphan name entry, the LFSCK
6097 * will set the orphan's ctime attribute as 1. Since normal applications
6098 * cannot change the object's ctime attribute as 1. Then when LFSCK run
6099 * next time, it can record the object (that ctime is 1) in the namespace
6100 * LFSCK trace file during the first-stage scanning.
6102 * \param[in] env pointer to the thread context
6103 * \param[in] com pointer to the lfsck component
6104 * \param[in] parent pointer to the object for the backend /lost+found
6105 * \param[in] ent pointer to the name entry for the target under the
6106 * backend /lost+found
6108 * \retval positive for repaired
6109 * \retval 0 if needs to repair nothing
6110 * \retval negative error number on failure
6112 static int lfsck_namespace_scan_local_lpf_one(const struct lu_env *env,
6113 struct lfsck_component *com,
6114 struct dt_object *parent,
6115 struct lu_dirent *ent)
6117 struct lfsck_thread_info *info = lfsck_env_info(env);
6118 struct lu_fid *key = &info->lti_fid;
6119 struct lu_attr *la = &info->lti_la;
6120 struct lfsck_instance *lfsck = com->lc_lfsck;
6121 struct dt_object *obj;
6122 struct dt_device *dev = lfsck->li_bottom;
6123 struct dt_object *child = NULL;
6124 struct thandle *th = NULL;
6132 child = lfsck_object_find_by_dev(env, dev, &ent->lde_fid);
6134 RETURN(PTR_ERR(child));
6136 if (!dt_object_exists(child)) {
6137 CDEBUG(D_LFSCK, "%s: lost+found/%s doesn't exist\n",
6138 lfsck_lfsck2name(lfsck), ent->lde_name);
6139 GOTO(out, rc = -ENOENT);
6142 LASSERT(!dt_object_remote(child));
6144 idx = lfsck_sub_trace_file_fid2idx(&ent->lde_fid);
6145 obj = com->lc_sub_trace_objs[idx].lsto_obj;
6146 fid_cpu_to_be(key, &ent->lde_fid);
6147 rc = dt_lookup(env, obj, (struct dt_rec *)&flags,
6148 (const struct dt_key *)key);
6151 flags |= LNTF_CHECK_ORPHAN;
6152 } else if (rc == -ENOENT) {
6153 flags = LNTF_CHECK_ORPHAN;
6158 th = lfsck_trans_create(env, dev, lfsck);
6160 GOTO(out, rc = PTR_ERR(th));
6162 /* a1. remove name entry from backend /lost+found */
6163 rc = dt_declare_delete(env, parent,
6164 (const struct dt_key *)ent->lde_name, th);
6168 if (S_ISDIR(lfsck_object_type(child))) {
6169 /* a2. decrease parent's nlink */
6170 rc = dt_declare_ref_del(env, parent, th);
6176 /* a3. remove child's FID from the LFSCK trace file. */
6177 rc = dt_declare_delete(env, obj,
6178 (const struct dt_key *)key, th);
6182 /* a4. set child's ctime as 1 */
6183 memset(la, 0, sizeof(*la));
6185 la->la_valid = LA_CTIME;
6186 rc = dt_declare_attr_set(env, child, la, th);
6191 /* a5. insert child's FID into the LFSCK trace file. */
6192 rc = dt_declare_insert(env, obj, (const struct dt_rec *)&flags,
6193 (const struct dt_key *)key, th);
6197 rc = dt_trans_start_local(env, dev, th);
6201 /* b1. remove name entry from backend /lost+found */
6202 rc = dt_delete(env, parent, (const struct dt_key *)ent->lde_name, th);
6206 if (S_ISDIR(lfsck_object_type(child))) {
6207 /* b2. decrease parent's nlink */
6208 dt_write_lock(env, parent, 0);
6209 rc = dt_ref_del(env, parent, th);
6210 dt_write_unlock(env, parent);
6216 /* a3. remove child's FID from the LFSCK trace file. */
6217 rc = dt_delete(env, obj, (const struct dt_key *)key, th);
6221 /* b4. set child's ctime as 1 */
6222 rc = dt_attr_set(env, child, la, th);
6227 /* b5. insert child's FID into the LFSCK trace file. */
6228 rc = dt_insert(env, obj, (const struct dt_rec *)&flags,
6229 (const struct dt_key *)key, th);
6231 GOTO(stop, rc = (rc == 0 ? 1 : rc));
6234 dt_trans_stop(env, dev, th);
6237 lfsck_object_put(env, child);
6243 * Handle orphans under the backend /lost+found directory
6245 * Some backend checker, such as e2fsck for ldiskfs may find some orphans
6246 * and put them under the backend /lost+found directory that is invisible
6247 * to client. The LFSCK will scan such directory, for the original client
6248 * visible orphans, add their fids into the namespace LFSCK trace file,
6249 * then the subsenquent namespace LFSCK second-stage scanning can handle
6250 * them as other objects to be double scanned: either move back to normal
6251 * namespace, or to the global visible orphan directory:
6252 * /ROOT/.lustre/lost+found/MDTxxxx/
6254 * \param[in] env pointer to the thread context
6255 * \param[in] com pointer to the lfsck component
6257 static void lfsck_namespace_scan_local_lpf(const struct lu_env *env,
6258 struct lfsck_component *com)
6260 struct lfsck_thread_info *info = lfsck_env_info(env);
6261 struct lu_dirent *ent =
6262 (struct lu_dirent *)info->lti_key;
6263 struct lu_seq_range *range = &info->lti_range;
6264 struct lfsck_instance *lfsck = com->lc_lfsck;
6265 struct ptlrpc_thread *thread = &lfsck->li_thread;
6266 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
6267 struct lfsck_namespace *ns = com->lc_file_ram;
6268 struct dt_object *parent;
6269 const struct dt_it_ops *iops;
6271 struct seq_server_site *ss = lfsck_dev_site(lfsck);
6273 __u32 idx = lfsck_dev_idx(lfsck);
6278 parent = lfsck_object_find_by_dev(env, lfsck->li_bottom,
6279 &LU_BACKEND_LPF_FID);
6280 if (IS_ERR(parent)) {
6281 CERROR("%s: fail to find backend /lost+found: rc = %ld\n",
6282 lfsck_lfsck2name(lfsck), PTR_ERR(parent));
6286 /* It is normal that the /lost+found does not exist for ZFS backend. */
6287 if (!dt_object_exists(parent))
6290 if (unlikely(!dt_try_as_dir(env, parent, true)))
6291 GOTO(out, rc = -ENOTDIR);
6293 CDEBUG(D_LFSCK, "%s: start to scan backend /lost+found\n",
6294 lfsck_lfsck2name(lfsck));
6296 com->lc_new_scanned = 0;
6297 iops = &parent->do_index_ops->dio_it;
6298 di = iops->init(env, parent, LUDA_64BITHASH | LUDA_TYPE);
6300 GOTO(out, rc = PTR_ERR(di));
6302 rc = iops->load(env, di, 0);
6304 rc = iops->next(env, di);
6309 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
6310 unlikely(!thread_is_running(thread)))
6313 rc = iops->rec(env, di, (struct dt_rec *)ent,
6314 LUDA_64BITHASH | LUDA_TYPE);
6316 rc = lfsck_unpack_ent(ent, &cookie, &type);
6318 if (unlikely(rc != 0)) {
6319 CDEBUG(D_LFSCK, "%s: fail to iterate backend "
6320 "/lost+found: rc = %d\n",
6321 lfsck_lfsck2name(lfsck), rc);
6326 /* skip dot and dotdot entries */
6327 if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen))
6330 if (!fid_seq_in_fldb(fid_seq(&ent->lde_fid)))
6333 if (fid_is_norm(&ent->lde_fid)) {
6334 fld_range_set_mdt(range);
6335 rc = fld_local_lookup(env, ss->ss_server_fld,
6336 fid_seq(&ent->lde_fid), range);
6339 } else if (idx != 0) {
6340 /* If the returned FID is IGIF, then there are three
6343 * 1) The object is upgraded from old Lustre-1.8 with
6344 * IGIF assigned to such object.
6345 * 2) The object is a backend local object and is
6346 * invisible to client.
6347 * 3) The object lost its LMV EA, and since there is
6348 * no FID-in-dirent for the orphan in the backend
6349 * /lost+found directory, then the low layer will
6350 * return IGIF for such object.
6352 * For MDTx (x != 0), it is either case 2) or case 3),
6353 * but from the LFSCK view, they are indistinguishable.
6354 * To be safe, the LFSCK will keep it there and report
6355 * some message, then the adminstrator can handle that
6358 * For MDT0, it is more possible the case 1). The LFSCK
6359 * will handle the orphan as an upgraded object. */
6360 CDEBUG(D_LFSCK, "%s: the orphan %.*s with IGIF "DFID
6361 "in the backend /lost+found on the MDT %04x, "
6362 "to be safe, skip it.\n",
6363 lfsck_lfsck2name(lfsck), ent->lde_namelen,
6364 ent->lde_name, PFID(&ent->lde_fid), idx);
6368 rc = lfsck_namespace_scan_local_lpf_one(env, com, parent, ent);
6371 down_write(&com->lc_sem);
6372 com->lc_new_scanned++;
6373 ns->ln_local_lpf_scanned++;
6375 ns->ln_local_lpf_moved++;
6377 ns->ln_local_lpf_skipped++;
6379 ns->ln_local_lpf_failed++;
6380 up_write(&com->lc_sem);
6382 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
6386 lfsck_control_speed_by_self(com);
6387 if (unlikely(!thread_is_running(thread))) {
6392 rc = iops->next(env, di);
6396 iops->fini(env, di);
6401 CDEBUG(D_LFSCK, "%s: stop to scan backend /lost+found: rc = %d\n",
6402 lfsck_lfsck2name(lfsck), rc);
6404 lfsck_object_put(env, parent);
6408 * Rescan the striped directory after the master LMV EA reset.
6410 * Sometimes, the master LMV EA of the striped directory maybe lost, so when
6411 * the namespace LFSCK engine scan the striped directory for the first time,
6412 * it will be regarded as a normal directory. As the LFSCK processing, some
6413 * other LFSCK instance on other MDT will find the shard of this striped dir,
6414 * and find that the master MDT-object of the striped directory lost its LMV
6415 * EA, then such remote LFSCK instance will regenerate the master LMV EA and
6416 * notify the LFSCK instance on this MDT to rescan the striped directory.
6418 * \param[in] env pointer to the thread context
6419 * \param[in] com pointer to the lfsck component
6420 * \param[in] llu the lfsck_lmv_unit that contains the striped directory
6423 * \retval positive number for success
6424 * \retval 0 for LFSCK stopped/paused
6425 * \retval negative error number on failure
6427 static int lfsck_namespace_rescan_striped_dir(const struct lu_env *env,
6428 struct lfsck_component *com,
6429 struct lfsck_lmv_unit *llu)
6431 struct lfsck_thread_info *info = lfsck_env_info(env);
6432 struct lfsck_instance *lfsck = com->lc_lfsck;
6433 struct lfsck_assistant_data *lad = com->lc_data;
6434 struct dt_object *dir;
6435 const struct dt_it_ops *iops;
6437 struct lu_dirent *ent =
6438 (struct lu_dirent *)info->lti_key;
6439 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
6440 struct ptlrpc_thread *thread = &lfsck->li_thread;
6441 struct lfsck_assistant_object *lso = NULL;
6442 struct lfsck_namespace_req *lnr;
6443 struct lfsck_assistant_req *lar;
6448 LASSERT(list_empty(&lad->lad_req_list));
6450 lfsck->li_lmv = &llu->llu_lmv;
6451 lfsck->li_obj_dir = lfsck_object_get(llu->llu_obj);
6452 rc = lfsck_open_dir(env, lfsck, 0);
6456 dir = lfsck->li_obj_dir;
6457 di = lfsck->li_di_dir;
6458 iops = &dir->do_index_ops->dio_it;
6460 rc = iops->rec(env, di, (struct dt_rec *)ent,
6461 lfsck->li_args_dir);
6463 rc = lfsck_unpack_ent(ent, &lfsck->li_cookie_dir,
6467 if (bk->lb_param & LPF_FAILOUT)
6473 if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen))
6477 lso = lfsck_assistant_object_init(env,
6478 lfsck_dto2fid(dir), NULL,
6479 lfsck->li_pos_current.lp_oit_cookie, true);
6481 if (bk->lb_param & LPF_FAILOUT)
6482 GOTO(out, rc = PTR_ERR(lso));
6489 lnr = lfsck_namespace_assistant_req_init(lfsck, lso, ent, type);
6491 if (bk->lb_param & LPF_FAILOUT)
6492 GOTO(out, rc = PTR_ERR(lnr));
6497 lar = &lnr->lnr_lar;
6498 rc = lfsck_namespace_assistant_handler_p1(env, com, lar);
6499 lfsck_namespace_assistant_req_fini(env, lar);
6500 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
6503 if (unlikely(!thread_is_running(thread)))
6507 rc = iops->next(env, di);
6511 if (lso != NULL && !IS_ERR(lso))
6512 lfsck_assistant_object_put(env, lso);
6514 lfsck_close_dir(env, lfsck, rc);
6518 /* The close_dir() may insert a dummy lnr in the lad->lad_req_list. */
6519 if (list_empty(&lad->lad_req_list))
6522 spin_lock(&lad->lad_lock);
6523 lar = list_entry(lad->lad_req_list.next, struct lfsck_assistant_req,
6525 list_del_init(&lar->lar_list);
6526 spin_unlock(&lad->lad_lock);
6528 rc = lfsck_namespace_assistant_handler_p1(env, com, lar);
6529 lfsck_namespace_assistant_req_fini(env, lar);
6531 RETURN(rc == 0 ? 1 : rc);
6535 lfsck_namespace_double_scan_one_trace_file(const struct lu_env *env,
6536 struct lfsck_component *com,
6537 struct dt_object *obj, bool first)
6539 struct lfsck_instance *lfsck = com->lc_lfsck;
6540 struct ptlrpc_thread *thread = &lfsck->li_thread;
6541 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
6542 struct lfsck_namespace *ns = com->lc_file_ram;
6543 const struct dt_it_ops *iops = &obj->do_index_ops->dio_it;
6544 struct dt_object *target;
6552 di = iops->init(env, obj, 0);
6554 RETURN(PTR_ERR(di));
6557 fid_cpu_to_be(&fid, &ns->ln_fid_latest_scanned_phase2);
6560 rc = iops->get(env, di, (const struct dt_key *)&fid);
6565 /* The start one either has been processed or does not exist,
6567 rc = iops->next(env, di);
6573 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
6574 unlikely(!thread_is_running(thread)))
6577 key = iops->key(env, di);
6586 fid_be_to_cpu(&fid, (const struct lu_fid *)key);
6587 if (!fid_is_sane(&fid)) {
6592 target = lfsck_object_find_bottom(env, lfsck, &fid);
6593 if (IS_ERR(target)) {
6594 rc = PTR_ERR(target);
6598 if (dt_object_exists(target)) {
6599 rc = iops->rec(env, di, (struct dt_rec *)&flags, 0);
6601 rc = lfsck_namespace_double_scan_one(env, com,
6608 lfsck_object_put(env, target);
6611 down_write(&com->lc_sem);
6612 com->lc_new_checked++;
6613 com->lc_new_scanned++;
6615 ns->ln_fid_latest_scanned_phase2 = fid;
6618 ns->ln_objs_repaired_phase2++;
6620 ns->ln_objs_failed_phase2++;
6621 up_write(&com->lc_sem);
6623 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
6626 if (unlikely(com->lc_time_next_checkpoint <=
6627 ktime_get_seconds()) &&
6628 com->lc_new_checked != 0) {
6629 down_write(&com->lc_sem);
6630 ns->ln_run_time_phase2 += ktime_get_seconds() -
6631 com->lc_time_last_checkpoint;
6632 ns->ln_time_last_checkpoint = ktime_get_real_seconds();
6633 ns->ln_objs_checked_phase2 += com->lc_new_checked;
6634 com->lc_new_checked = 0;
6635 lfsck_namespace_store(env, com);
6636 up_write(&com->lc_sem);
6638 com->lc_time_last_checkpoint = ktime_get_seconds();
6639 com->lc_time_next_checkpoint =
6640 com->lc_time_last_checkpoint +
6641 LFSCK_CHECKPOINT_INTERVAL;
6644 lfsck_control_speed_by_self(com);
6645 if (unlikely(!thread_is_running(thread)))
6648 rc = iops->next(env, di);
6657 iops->fini(env, di);
6662 static int lfsck_namespace_assistant_handler_p2(const struct lu_env *env,
6663 struct lfsck_component *com)
6665 struct lfsck_instance *lfsck = com->lc_lfsck;
6666 struct lfsck_namespace *ns = com->lc_file_ram;
6671 while (!list_empty(&lfsck->li_list_lmv)) {
6672 struct lfsck_lmv_unit *llu;
6674 spin_lock(&lfsck->li_lock);
6675 llu = list_entry(lfsck->li_list_lmv.next,
6676 struct lfsck_lmv_unit, llu_link);
6677 list_del_init(&llu->llu_link);
6678 spin_unlock(&lfsck->li_lock);
6680 rc = lfsck_namespace_rescan_striped_dir(env, com, llu);
6685 CDEBUG(D_LFSCK, "%s: namespace LFSCK phase2 scan start\n",
6686 lfsck_lfsck2name(lfsck));
6688 lfsck_namespace_scan_local_lpf(env, com);
6690 com->lc_new_checked = 0;
6691 com->lc_new_scanned = 0;
6692 com->lc_time_last_checkpoint = ktime_get_seconds();
6693 com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
6694 LFSCK_CHECKPOINT_INTERVAL;
6696 i = lfsck_sub_trace_file_fid2idx(&ns->ln_fid_latest_scanned_phase2);
6697 rc = lfsck_namespace_double_scan_one_trace_file(env, com,
6698 com->lc_sub_trace_objs[i].lsto_obj, true);
6699 while (rc > 0 && ++i < LFSCK_STF_COUNT)
6700 rc = lfsck_namespace_double_scan_one_trace_file(env, com,
6701 com->lc_sub_trace_objs[i].lsto_obj, false);
6703 CDEBUG(D_LFSCK, "%s: namespace LFSCK phase2 scan stop at the No. %d "
6704 "trace file: rc = %d\n", lfsck_lfsck2name(lfsck), i, rc);
6709 static void lfsck_namespace_assistant_fill_pos(const struct lu_env *env,
6710 struct lfsck_component *com,
6711 struct lfsck_position *pos)
6713 struct lfsck_assistant_data *lad = com->lc_data;
6714 struct lfsck_namespace_req *lnr;
6716 if (((struct lfsck_namespace *)(com->lc_file_ram))->ln_status !=
6720 if (list_empty(&lad->lad_req_list))
6723 lnr = list_entry(lad->lad_req_list.next,
6724 struct lfsck_namespace_req,
6726 pos->lp_oit_cookie = lnr->lnr_lar.lar_parent->lso_oit_cookie;
6727 pos->lp_dir_cookie = lnr->lnr_dir_cookie - 1;
6728 pos->lp_dir_parent = lnr->lnr_lar.lar_parent->lso_fid;
6731 static int lfsck_namespace_double_scan_result(const struct lu_env *env,
6732 struct lfsck_component *com,
6735 struct lfsck_instance *lfsck = com->lc_lfsck;
6736 struct lfsck_namespace *ns = com->lc_file_ram;
6738 down_write(&com->lc_sem);
6739 ns->ln_run_time_phase2 += ktime_get_seconds() -
6740 com->lc_time_last_checkpoint;
6741 ns->ln_time_last_checkpoint = ktime_get_real_seconds();
6742 ns->ln_objs_checked_phase2 += com->lc_new_checked;
6743 com->lc_new_checked = 0;
6746 if (ns->ln_flags & LF_INCOMPLETE)
6747 ns->ln_status = LS_PARTIAL;
6749 ns->ln_status = LS_COMPLETED;
6750 ns->ln_flags &= ~LF_SCANNED_ONCE;
6751 if (!(lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN))
6752 ns->ln_flags &= ~LF_INCONSISTENT;
6753 ns->ln_time_last_complete = ns->ln_time_last_checkpoint;
6754 ns->ln_success_count++;
6755 } else if (rc == 0) {
6756 if (lfsck->li_status != 0)
6757 ns->ln_status = lfsck->li_status;
6759 ns->ln_status = LS_STOPPED;
6761 ns->ln_status = LS_FAILED;
6764 rc = lfsck_namespace_store(env, com);
6765 up_write(&com->lc_sem);
6771 lfsck_namespace_assistant_sync_failures_interpret(const struct lu_env *env,
6772 struct ptlrpc_request *req,
6776 struct lfsck_async_interpret_args *laia = args;
6777 struct lfsck_tgt_desc *ltd = laia->laia_ltd;
6779 ltd->ltd_synced_failures = 1;
6786 * Notify remote LFSCK instances about former failures.
6788 * The local LFSCK instance has recorded which MDTs have ever failed to respond
6789 * some LFSCK verification requests (maybe because of network issues or the MDT
6790 * itself trouble). During the respond gap the MDT may missed some name entries
6791 * verification, then the MDT cannot know whether related MDT-objects have been
6792 * referenced by related name entries or not, then in the second-stage scanning,
6793 * these MDT-objects will be regarded as orphan, if the MDT-object contains bad
6794 * linkEA for back reference, then it will misguide the LFSCK to generate wrong
6795 * name entry for repairing the orphan.
6797 * To avoid above trouble, when layout LFSCK finishes the first-stage scanning,
6798 * it will scan the bitmap for the ever failed MDTs, and notify them that they
6799 * have ever missed some name entries verification and should skip the handling
6800 * for orphan MDT-objects.
6802 * \param[in] env pointer to the thread context
6803 * \param[in] com pointer to the lfsck component
6804 * \param[in] lr pointer to the lfsck request
6806 static void lfsck_namespace_assistant_sync_failures(const struct lu_env *env,
6807 struct lfsck_component *com,
6808 struct lfsck_request *lr)
6810 struct lfsck_async_interpret_args *laia =
6811 &lfsck_env_info(env)->lti_laia2;
6812 struct lfsck_assistant_data *lad = com->lc_data;
6813 struct lfsck_namespace *ns = com->lc_file_ram;
6814 struct lfsck_instance *lfsck = com->lc_lfsck;
6815 struct lfsck_tgt_descs *ltds = &lfsck->li_mdt_descs;
6816 struct lfsck_tgt_desc *ltd;
6817 struct ptlrpc_request_set *set;
6822 if (!test_bit(LAD_INCOMPLETE, &lad->lad_flags))
6825 set = ptlrpc_prep_set();
6827 GOTO(out, rc = -ENOMEM);
6829 lr->lr_flags2 = ns->ln_flags | LF_INCOMPLETE;
6830 memset(laia, 0, sizeof(*laia));
6831 lad->lad_touch_gen++;
6833 down_read(<ds->ltd_rw_sem);
6834 cfs_foreach_bit(lad->lad_bitmap, idx) {
6835 ltd = lfsck_ltd2tgt(ltds, idx);
6839 laia->laia_ltd = ltd;
6840 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
6841 lfsck_namespace_assistant_sync_failures_interpret,
6842 laia, LFSCK_NOTIFY);
6844 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail "
6845 "to sync failure with MDT %x: rc = %d\n",
6846 lfsck_lfsck2name(lfsck), ltd->ltd_index, rc);
6848 up_read(<ds->ltd_rw_sem);
6850 rc = ptlrpc_set_wait(env, set);
6851 ptlrpc_set_destroy(set);
6857 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail "
6858 "to sync failure with MDTs, and related MDTs "
6859 "may handle orphan improperly: rc = %d\n",
6860 lfsck_lfsck2name(lfsck), rc);
6865 const struct lfsck_assistant_operations lfsck_namespace_assistant_ops = {
6866 .la_handler_p1 = lfsck_namespace_assistant_handler_p1,
6867 .la_handler_p2 = lfsck_namespace_assistant_handler_p2,
6868 .la_fill_pos = lfsck_namespace_assistant_fill_pos,
6869 .la_double_scan_result = lfsck_namespace_double_scan_result,
6870 .la_req_fini = lfsck_namespace_assistant_req_fini,
6871 .la_sync_failures = lfsck_namespace_assistant_sync_failures,
6875 * Verify the specified linkEA entry for the given directory object.
6876 * If the object has no such linkEA entry or it has more other linkEA
6877 * entries, then re-generate the linkEA with the given information.
6879 * \param[in] env pointer to the thread context
6880 * \param[in] obj pointer to the dt_object to be handled
6881 * \param[in] cname the name for the child in the parent directory
6882 * \param[in] pfid the parent directory's FID for the linkEA
6884 * \retval 0 for success
6885 * \retval negative error number on failure
6887 int lfsck_verify_linkea(const struct lu_env *env, struct lfsck_instance *lfsck,
6888 struct dt_object *obj, const struct lu_name *cname,
6889 const struct lu_fid *pfid)
6891 struct dt_device *dev = lfsck_obj2dev(obj);
6892 struct linkea_data ldata = { NULL };
6893 struct lu_buf linkea_buf;
6896 int fl = LU_XATTR_CREATE;
6901 if (!dt_object_exists(obj))
6904 if (!S_ISDIR(lfsck_object_type(obj)))
6907 rc = lfsck_links_read_with_rec(env, obj, &ldata);
6908 if (rc == -ENODATA) {
6910 } else if (rc == 0) {
6911 fl = LU_XATTR_REPLACE;
6912 if (ldata.ld_leh->leh_reccount != 1) {
6915 rc = linkea_links_find(&ldata, cname, pfid);
6924 rc = linkea_links_new(&ldata, &lfsck_env_info(env)->lti_linkea_buf,
6929 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
6930 ldata.ld_leh->leh_len);
6931 th = lfsck_trans_create(env, dev, lfsck);
6933 RETURN(PTR_ERR(th));
6935 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
6936 XATTR_NAME_LINK, fl, th);
6940 rc = dt_trans_start_local(env, dev, th);
6944 dt_write_lock(env, obj, 0);
6945 rc = dt_xattr_set(env, obj, &linkea_buf,
6946 XATTR_NAME_LINK, fl, th);
6947 dt_write_unlock(env, obj);
6952 dt_trans_stop(env, dev, th);
6957 * Get the name and parent directory's FID from the first linkEA entry.
6959 * \param[in] env pointer to the thread context
6960 * \param[in] obj pointer to the object which get linkEA from
6961 * \param[out] name pointer to the buffer to hold the name
6962 * in the first linkEA entry
6963 * \param[out] pfid pointer to the buffer to hold the parent
6964 * directory's FID in the first linkEA entry
6966 * \retval 0 for success
6967 * \retval negative error number on failure
6969 int lfsck_links_get_first(const struct lu_env *env, struct dt_object *obj,
6970 char *name, struct lu_fid *pfid)
6972 struct lu_name *cname = &lfsck_env_info(env)->lti_name;
6973 struct linkea_data ldata = { NULL };
6976 rc = lfsck_links_read_with_rec(env, obj, &ldata);
6980 linkea_first_entry(&ldata);
6981 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, cname, pfid);
6982 if (!linkea_entry_is_valid(&ldata, cname, pfid))
6985 /* To guarantee the 'name' is terminated with '0'. */
6986 memcpy(name, cname->ln_name, cname->ln_namelen);
6987 name[cname->ln_namelen] = 0;
6993 * Update the object's name entry with the given FID.
6995 * \param[in] env pointer to the thread context
6996 * \param[in] lfsck pointer to the lfsck instance
6997 * \param[in] dir pointer to the directory that holds
6999 * \param[in] name the name for the entry to be updated
7000 * \param[in] fid the new FID for the name entry referenced
7001 * \param[in] type the type for the name entry to be updated
7003 * \retval 0 for success
7004 * \retval negative error number on failure
7006 int lfsck_update_name_entry(const struct lu_env *env,
7007 struct lfsck_instance *lfsck,
7008 struct dt_object *dir, const char *name,
7009 const struct lu_fid *fid, __u32 type)
7011 struct lfsck_thread_info *info = lfsck_env_info(env);
7012 struct dt_insert_rec *rec = &info->lti_dt_rec;
7013 struct lfsck_lock_handle *llh = &info->lti_llh;
7014 struct dt_device *dev = lfsck_obj2dev(dir);
7020 rc = lfsck_lock(env, lfsck, dir, name, llh,
7021 MDS_INODELOCK_UPDATE, LCK_PW);
7025 th = lfsck_trans_create(env, dev, lfsck);
7027 GOTO(unlock, rc = PTR_ERR(th));
7029 rc = dt_declare_delete(env, dir, (const struct dt_key *)name, th);
7033 rec->rec_type = type;
7035 rc = dt_declare_insert(env, dir, (const struct dt_rec *)rec,
7036 (const struct dt_key *)name, th);
7040 rc = dt_declare_ref_add(env, dir, th);
7044 rc = dt_trans_start_local(env, dev, th);
7048 rc = dt_delete(env, dir, (const struct dt_key *)name, th);
7049 if (rc == -ENOENT) {
7057 rc = dt_insert(env, dir, (const struct dt_rec *)rec,
7058 (const struct dt_key *)name, th);
7059 if (rc == 0 && S_ISDIR(type) && !exists) {
7060 dt_write_lock(env, dir, 0);
7061 rc = dt_ref_add(env, dir, th);
7062 dt_write_unlock(env, dir);
7068 dt_trans_stop(env, dev, th);
7072 CDEBUG(D_LFSCK, "%s: update name entry "DFID"/%s with the FID "DFID
7073 " and the type %o: rc = %d\n", lfsck_lfsck2name(lfsck),
7074 PFID(lfsck_dto2fid(dir)), name, PFID(fid), type, rc);
7079 int lfsck_namespace_setup(const struct lu_env *env,
7080 struct lfsck_instance *lfsck)
7082 struct lfsck_component *com;
7083 struct lfsck_namespace *ns;
7084 struct dt_object *root = NULL;
7085 struct dt_object *obj;
7090 LASSERT(lfsck->li_master);
7096 INIT_LIST_HEAD(&com->lc_link);
7097 INIT_LIST_HEAD(&com->lc_link_dir);
7098 init_rwsem(&com->lc_sem);
7099 atomic_set(&com->lc_ref, 1);
7100 com->lc_lfsck = lfsck;
7101 com->lc_type = LFSCK_TYPE_NAMESPACE;
7102 com->lc_ops = &lfsck_namespace_ops;
7103 com->lc_data = lfsck_assistant_data_init(
7104 &lfsck_namespace_assistant_ops,
7106 if (com->lc_data == NULL)
7107 GOTO(out, rc = -ENOMEM);
7109 com->lc_file_size = sizeof(struct lfsck_namespace);
7110 OBD_ALLOC(com->lc_file_ram, com->lc_file_size);
7111 if (com->lc_file_ram == NULL)
7112 GOTO(out, rc = -ENOMEM);
7114 OBD_ALLOC(com->lc_file_disk, com->lc_file_size);
7115 if (com->lc_file_disk == NULL)
7116 GOTO(out, rc = -ENOMEM);
7118 for (i = 0; i < LFSCK_STF_COUNT; i++)
7119 mutex_init(&com->lc_sub_trace_objs[i].lsto_mutex);
7121 root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
7123 GOTO(out, rc = PTR_ERR(root));
7125 if (unlikely(!dt_try_as_dir(env, root, true)))
7126 GOTO(out, rc = -ENOTDIR);
7128 obj = local_index_find_or_create(env, lfsck->li_los, root,
7130 S_IFREG | S_IRUGO | S_IWUSR,
7131 &dt_lfsck_namespace_features);
7133 GOTO(out, rc = PTR_ERR(obj));
7136 rc = lfsck_namespace_load(env, com);
7137 if (rc == -ENODATA) {
7138 rc = lfsck_namespace_init(env, com);
7139 } else if (rc < 0) {
7140 rc = lfsck_namespace_reset(env, com, true);
7142 rc = lfsck_load_sub_trace_files(env, com,
7143 &dt_lfsck_namespace_features, LFSCK_NAMESPACE, false);
7145 rc = lfsck_namespace_reset(env, com, true);
7150 ns = com->lc_file_ram;
7151 switch (ns->ln_status) {
7156 spin_lock(&lfsck->li_lock);
7157 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
7158 spin_unlock(&lfsck->li_lock);
7161 CERROR("%s: unknown lfsck_namespace status %d\n",
7162 lfsck_lfsck2name(lfsck), ns->ln_status);
7164 case LS_SCANNING_PHASE1:
7165 case LS_SCANNING_PHASE2:
7166 /* No need to store the status to disk right now.
7167 * If the system crashed before the status stored,
7168 * it will be loaded back when next time. */
7169 ns->ln_status = LS_CRASHED;
7173 spin_lock(&lfsck->li_lock);
7174 list_add_tail(&com->lc_link, &lfsck->li_list_scan);
7175 list_add_tail(&com->lc_link_dir, &lfsck->li_list_dir);
7176 spin_unlock(&lfsck->li_lock);
7183 if (root != NULL && !IS_ERR(root))
7184 lfsck_object_put(env, root);
7186 lfsck_component_cleanup(env, com);
7187 CERROR("%s: fail to init namespace LFSCK component: rc = %d\n",
7188 lfsck_lfsck2name(lfsck), rc);