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 = dt_trans_create(env, dev);
536 GOTO(log, rc = PTR_ERR(th));
539 rc = dt_declare_delete(env, obj,
540 (const struct dt_key *)key, th);
546 rc = dt_declare_insert(env, obj,
547 (const struct dt_rec *)&new,
548 (const struct dt_key *)key, th);
553 rc = dt_trans_start_local(env, dev, th);
558 rc = dt_delete(env, obj, (const struct dt_key *)key, th);
564 rc = dt_insert(env, obj, (const struct dt_rec *)&new,
565 (const struct dt_key *)key, th);
573 if (th != NULL && !IS_ERR(th))
574 dt_trans_stop(env, dev, th);
576 CDEBUG(D_LFSCK, "%s: namespace LFSCK %s flags for "DFID" in the "
577 "trace file, flags %x, old %x, new %x: rc = %d\n",
578 lfsck_lfsck2name(lfsck), add ? "add" : "del", PFID(fid),
579 (__u32)flags, (__u32)old, (__u32)new, rc);
582 mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
583 lfsck_object_put(env, obj);
588 int lfsck_namespace_check_exist(const struct lu_env *env,
589 struct dt_object *dir,
590 struct dt_object *obj, const char *name)
592 struct lu_fid *fid = &lfsck_env_info(env)->lti_fid;
596 if (unlikely(lfsck_is_dead_obj(obj)))
597 RETURN(LFSCK_NAMEENTRY_DEAD);
599 rc = dt_lookup(env, dir, (struct dt_rec *)fid,
600 (const struct dt_key *)name);
602 RETURN(LFSCK_NAMEENTRY_REMOVED);
607 if (!lu_fid_eq(fid, lfsck_dto2fid(obj)))
608 RETURN(LFSCK_NAMEENTRY_RECREATED);
613 static int lfsck_declare_namespace_exec_dir(const struct lu_env *env,
614 struct dt_object *obj,
615 struct thandle *handle)
619 /* For remote updating LINKEA, there may be further LFSCK action
620 * on remote MDT after the updating, so update the LINKEA ASAP. */
621 if (dt_object_remote(obj))
624 /* For destroying all invalid linkEA entries. */
625 rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, handle);
627 /* For insert new linkEA entry. */
628 rc = dt_declare_xattr_set(env, obj,
629 lfsck_buf_get_const(env, NULL, MAX_LINKEA_SIZE),
630 XATTR_NAME_LINK, 0, handle);
634 int __lfsck_links_read(const struct lu_env *env, struct dt_object *obj,
635 struct linkea_data *ldata, bool with_rec)
639 if (ldata->ld_buf->lb_buf == NULL)
642 if (!dt_object_exists(obj))
645 rc = dt_xattr_get(env, obj, ldata->ld_buf, XATTR_NAME_LINK);
647 /* Buf was too small, figure out what we need. */
648 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LINK);
649 if (unlikely(rc == 0))
655 lu_buf_realloc(ldata->ld_buf, rc);
656 if (ldata->ld_buf->lb_buf == NULL)
659 rc = dt_xattr_get(env, obj, ldata->ld_buf, XATTR_NAME_LINK);
662 if (unlikely(rc == 0))
667 rc = linkea_init_with_rec(ldata);
669 rc = linkea_init(ldata);
676 * Remove linkEA for the given object.
678 * The caller should take the ldlm lock before the calling.
680 * \param[in] env pointer to the thread context
681 * \param[in] com pointer to the lfsck component
682 * \param[in] obj pointer to the dt_object to be handled
684 * \retval 0 for repaired cases
685 * \retval negative error number on failure
687 static int lfsck_namespace_links_remove(const struct lu_env *env,
688 struct lfsck_component *com,
689 struct dt_object *obj)
691 struct lfsck_instance *lfsck = com->lc_lfsck;
692 struct dt_device *dev = lfsck_obj2dev(obj);
693 struct thandle *th = NULL;
697 LASSERT(dt_object_remote(obj) == 0);
699 th = dt_trans_create(env, dev);
701 GOTO(log, rc = PTR_ERR(th));
703 rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, th);
707 rc = dt_trans_start_local(env, dev, th);
711 dt_write_lock(env, obj, 0);
712 if (unlikely(lfsck_is_dead_obj(obj)))
713 GOTO(unlock, rc = -ENOENT);
715 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
716 GOTO(unlock, rc = 0);
718 rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, th);
723 dt_write_unlock(env, obj);
726 dt_trans_stop(env, dev, th);
729 CDEBUG(D_LFSCK, "%s: namespace LFSCK remove invalid linkEA "
730 "for the object "DFID": rc = %d\n",
731 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
734 struct lfsck_namespace *ns = com->lc_file_ram;
736 ns->ln_flags |= LF_INCONSISTENT;
742 static int lfsck_links_write(const struct lu_env *env, struct dt_object *obj,
743 struct linkea_data *ldata, struct thandle *handle)
748 lfsck_buf_init(&buf, ldata->ld_buf->lb_buf, ldata->ld_leh->leh_len);
751 rc = dt_xattr_set(env, obj, &buf, XATTR_NAME_LINK, 0, handle);
752 if (unlikely(rc == -ENOSPC)) {
753 rc = linkea_overflow_shrink(ldata);
754 if (likely(rc > 0)) {
763 static inline bool linkea_reclen_is_valid(const struct linkea_data *ldata)
765 if (ldata->ld_reclen <= 0)
768 if ((char *)ldata->ld_lee + ldata->ld_reclen >
769 (char *)ldata->ld_leh + ldata->ld_leh->leh_len)
775 static inline bool linkea_entry_is_valid(const struct linkea_data *ldata,
776 const struct lu_name *cname,
777 const struct lu_fid *pfid)
779 if (!linkea_reclen_is_valid(ldata))
782 if (cname->ln_namelen <= 0 || cname->ln_namelen > NAME_MAX)
785 if (!fid_is_sane(pfid))
791 static int lfsck_namespace_unpack_linkea_entry(struct linkea_data *ldata,
792 struct lu_name *cname,
794 char *buf, const int buflen)
796 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, cname, pfid);
797 if (unlikely(!linkea_entry_is_valid(ldata, cname, pfid)))
800 /* To guarantee the 'name' is terminated with '0'. */
801 memcpy(buf, cname->ln_name, cname->ln_namelen);
802 buf[cname->ln_namelen] = 0;
803 cname->ln_name = buf;
808 static void lfsck_linkea_del_buf(struct linkea_data *ldata,
809 const struct lu_name *lname)
811 LASSERT(ldata->ld_leh != NULL && ldata->ld_lee != NULL);
813 /* If current record is corrupted, all the subsequent
814 * records will be dropped. */
815 if (unlikely(!linkea_reclen_is_valid(ldata))) {
816 void *ptr = ldata->ld_lee;
818 ldata->ld_leh->leh_len = sizeof(struct link_ea_header);
819 ldata->ld_leh->leh_reccount = 0;
820 linkea_first_entry(ldata);
821 while (ldata->ld_lee != NULL &&
822 (char *)ldata->ld_lee < (char *)ptr) {
823 int reclen = (ldata->ld_lee->lee_reclen[0] << 8) |
824 ldata->ld_lee->lee_reclen[1];
826 ldata->ld_leh->leh_len += reclen;
827 ldata->ld_leh->leh_reccount++;
828 ldata->ld_lee = (struct link_ea_entry *)
829 ((char *)ldata->ld_lee + reclen);
832 ldata->ld_lee = NULL;
834 linkea_del_buf(ldata, lname);
838 static int lfsck_namespace_filter_linkea_entry(struct linkea_data *ldata,
839 struct lu_name *cname,
843 struct link_ea_entry *oldlee;
847 oldlee = ldata->ld_lee;
848 oldlen = ldata->ld_reclen;
849 linkea_next_entry(ldata);
850 while (ldata->ld_lee != NULL) {
851 ldata->ld_reclen = (ldata->ld_lee->lee_reclen[0] << 8) |
852 ldata->ld_lee->lee_reclen[1];
853 if (unlikely(!linkea_reclen_is_valid(ldata))) {
854 lfsck_linkea_del_buf(ldata, NULL);
855 LASSERT(ldata->ld_lee == NULL);
856 } else if (unlikely(ldata->ld_reclen == oldlen &&
857 memcmp(ldata->ld_lee, oldlee, oldlen) == 0)) {
862 lfsck_linkea_del_buf(ldata, cname);
864 linkea_next_entry(ldata);
867 ldata->ld_lee = oldlee;
868 ldata->ld_reclen = oldlen;
874 * Insert orphan into .lustre/lost+found/MDTxxxx/ locally.
876 * Add the specified orphan MDT-object to the .lustre/lost+found/MDTxxxx/
877 * with the given type to generate the name, the detailed rules for name
878 * have been described as following.
880 * The function also generates the linkEA corresponding to the name entry
881 * under the .lustre/lost+found/MDTxxxx/ for the orphan MDT-object.
883 * \param[in] env pointer to the thread context
884 * \param[in] com pointer to the lfsck component
885 * \param[in] orphan pointer to the orphan MDT-object
886 * \param[in] infix additional information for the orphan name, such as
887 * the FID for original
888 * \param[in] type the type for describing why the orphan MDT-object is
889 * created. The rules are as following:
891 * type "D": The MDT-object is a directory, it may knows its parent
892 * but because there is no valid linkEA, the LFSCK cannot
893 * know where to put it back to the namespace.
894 * type "O": The MDT-object has no linkEA, and there is no name
895 * entry that references the MDT-object.
897 * type "S": The orphan MDT-object is a shard of a striped directory
899 * \see lfsck_layout_recreate_parent() for more types.
901 * The orphan name will be like:
902 * ${FID}-${infix}-${type}-${conflict_version}
904 * \param[out] count if some others inserted some linkEA entries by race,
905 * then return the linkEA entries count.
907 * \retval positive number for repaired cases
908 * \retval 0 if needs to repair nothing
909 * \retval negative error number on failure
911 static int lfsck_namespace_insert_orphan(const struct lu_env *env,
912 struct lfsck_component *com,
913 struct dt_object *orphan,
914 const char *infix, const char *type,
917 struct lfsck_thread_info *info = lfsck_env_info(env);
918 struct lu_name *cname = &info->lti_name;
919 struct dt_insert_rec *rec = &info->lti_dt_rec;
920 struct lu_attr *la = &info->lti_la2;
921 const struct lu_fid *cfid = lfsck_dto2fid(orphan);
922 const struct lu_fid *pfid;
924 struct lfsck_instance *lfsck = com->lc_lfsck;
925 struct dt_device *dev = lfsck_obj2dev(orphan);
926 struct dt_object *parent;
927 struct thandle *th = NULL;
928 struct lfsck_lock_handle *pllh = &info->lti_llh;
929 struct lustre_handle clh = { 0 };
930 struct linkea_data ldata2 = { NULL };
931 struct lu_buf linkea_buf;
938 cname->ln_name = NULL;
939 if (unlikely(lfsck->li_lpf_obj == NULL))
940 GOTO(log, rc = -ENXIO);
942 parent = lfsck->li_lpf_obj;
943 pfid = lfsck_dto2fid(parent);
947 namelen = snprintf(info->lti_key, NAME_MAX, DFID"%s-%s-%d",
948 PFID(cfid), infix, type, idx++);
949 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
950 (const struct dt_key *)info->lti_key);
951 if (rc != 0 && rc != -ENOENT)
954 if (unlikely(rc == 0 && lu_fid_eq(cfid, &tfid)))
956 } while (rc == 0 && !exist);
958 rc = lfsck_lock(env, lfsck, parent, info->lti_key, pllh,
959 MDS_INODELOCK_UPDATE, LCK_PW);
963 /* Re-check whether the name conflict with othrs after taken
965 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
966 (const struct dt_key *)info->lti_key);
968 if (!lu_fid_eq(cfid, &tfid)) {
975 } else if (rc != -ENOENT) {
981 cname->ln_name = info->lti_key;
982 cname->ln_namelen = namelen;
983 rc = linkea_links_new(&ldata2, &info->lti_linkea_buf2,
988 rc = lfsck_ibits_lock(env, lfsck, orphan, &clh,
989 MDS_INODELOCK_UPDATE | MDS_INODELOCK_LOOKUP |
990 MDS_INODELOCK_XATTR, LCK_EX);
994 lfsck_buf_init(&linkea_buf, ldata2.ld_buf->lb_buf,
995 ldata2.ld_leh->leh_len);
996 th = dt_trans_create(env, dev);
998 GOTO(log, rc = PTR_ERR(th));
1000 if (S_ISDIR(lfsck_object_type(orphan))) {
1001 rc = dt_declare_delete(env, orphan,
1002 (const struct dt_key *)dotdot, th);
1006 rec->rec_type = S_IFDIR;
1007 rec->rec_fid = pfid;
1008 rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
1009 (const struct dt_key *)dotdot, th);
1014 rc = dt_declare_xattr_set(env, orphan, &linkea_buf,
1015 XATTR_NAME_LINK, 0, th);
1020 rec->rec_type = lfsck_object_type(orphan) & S_IFMT;
1021 rec->rec_fid = cfid;
1022 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1023 (const struct dt_key *)cname->ln_name,
1028 if (S_ISDIR(rec->rec_type)) {
1029 rc = dt_declare_ref_add(env, parent, th);
1035 memset(la, 0, sizeof(*la));
1036 la->la_ctime = ktime_get_real_seconds();
1037 la->la_valid = LA_CTIME;
1038 rc = dt_declare_attr_set(env, orphan, la, th);
1042 rc = dt_trans_start_local(env, dev, th);
1046 dt_write_lock(env, orphan, 0);
1047 rc = lfsck_links_read2_with_rec(env, orphan, &ldata2);
1048 if (likely(rc == -ENODATA || rc == -EINVAL)) {
1049 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1050 GOTO(unlock, rc = 1);
1052 if (S_ISDIR(lfsck_object_type(orphan))) {
1053 rc = dt_delete(env, orphan,
1054 (const struct dt_key *)dotdot, th);
1058 rec->rec_type = S_IFDIR;
1059 rec->rec_fid = pfid;
1060 rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1061 (const struct dt_key *)dotdot, th);
1066 rc = dt_xattr_set(env, orphan, &linkea_buf, XATTR_NAME_LINK, 0,
1069 if (rc == 0 && count != NULL)
1070 *count = ldata2.ld_leh->leh_reccount;
1074 dt_write_unlock(env, orphan);
1076 if (rc == 0 && !exist) {
1077 rec->rec_type = lfsck_object_type(orphan) & S_IFMT;
1078 rec->rec_fid = cfid;
1079 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1080 (const struct dt_key *)cname->ln_name, th);
1081 if (rc == 0 && S_ISDIR(rec->rec_type)) {
1082 dt_write_lock(env, parent, 0);
1083 rc = dt_ref_add(env, parent, th);
1084 dt_write_unlock(env, parent);
1089 rc = dt_attr_set(env, orphan, la, th);
1091 GOTO(stop, rc = (rc == 0 ? 1 : rc));
1094 dt_write_unlock(env, orphan);
1097 dt_trans_stop(env, dev, th);
1100 lfsck_ibits_unlock(&clh, LCK_EX);
1102 CDEBUG(D_LFSCK, "%s: namespace LFSCK insert orphan for the "
1103 "object "DFID", name = %s: rc = %d\n",
1104 lfsck_lfsck2name(lfsck), PFID(cfid),
1105 cname->ln_name != NULL ? cname->ln_name : "<NULL>", rc);
1108 struct lfsck_namespace *ns = com->lc_file_ram;
1110 ns->ln_flags |= LF_INCONSISTENT;
1116 static int lfsck_lmv_set(const struct lu_env *env,
1117 struct lfsck_instance *lfsck,
1118 struct dt_object *obj,
1119 struct lmv_mds_md_v1 *lmv)
1121 struct dt_device *dev = lfsck->li_next;
1122 struct thandle *th = NULL;
1123 struct lu_buf buf = { lmv, sizeof(*lmv) };
1128 th = dt_trans_create(env, dev);
1130 RETURN(PTR_ERR(th));
1132 rc = dt_declare_xattr_set(env, obj, &buf, XATTR_NAME_LMV, 0, th);
1136 rc = dt_trans_start_local(env, dev, th);
1140 rc = dt_xattr_set(env, obj, &buf, XATTR_NAME_LMV, 0, th);
1146 dt_trans_stop(env, dev, th);
1151 static int lfsck_lmv_delete(const struct lu_env *env,
1152 struct lfsck_instance *lfsck,
1153 struct dt_object *obj)
1155 struct dt_device *dev = lfsck->li_next;
1156 struct thandle *th = NULL;
1161 th = dt_trans_create(env, dev);
1163 RETURN(PTR_ERR(th));
1165 rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LMV, th);
1169 rc = dt_trans_start_local(env, dev, th);
1173 rc = dt_xattr_del(env, obj, XATTR_NAME_LMV, th);
1179 dt_trans_stop(env, dev, th);
1184 static inline int lfsck_object_is_shard(const struct lu_env *env,
1185 struct lfsck_instance *lfsck,
1186 struct dt_object *obj,
1187 const struct lu_name *lname)
1189 struct lfsck_thread_info *info = lfsck_env_info(env);
1190 struct lmv_mds_md_v1 *lmv = &info->lti_lmv;
1193 rc = lfsck_shard_name_to_index(env, lname->ln_name, lname->ln_namelen,
1194 lfsck_object_type(obj),
1195 lfsck_dto2fid(obj));
1199 rc = lfsck_read_stripe_lmv(env, lfsck, obj, lmv);
1203 if (!rc && lmv->lmv_magic == LMV_MAGIC_STRIPE)
1210 * Add the specified name entry back to namespace.
1212 * If there is a linkEA entry that back references a name entry under
1213 * some parent directory, but such parent directory does not have the
1214 * claimed name entry. On the other hand, the linkEA entries count is
1215 * not larger than the MDT-object's hard link count. Under such case,
1216 * it is quite possible that the name entry is lost. Then the LFSCK
1217 * should add the name entry back to the namespace.
1219 * If \a child is shard, which means \a parent is a striped directory,
1220 * if \a parent has LMV, we need to delete it before insertion because
1221 * now parent's striping is broken and can't be parsed correctly.
1223 * \param[in] env pointer to the thread context
1224 * \param[in] com pointer to the lfsck component
1225 * \param[in] parent pointer to the directory under which the name entry
1226 * will be inserted into
1227 * \param[in] child pointer to the object referenced by the name entry
1228 * that to be inserted into the parent
1229 * \param[in] lname the name for the child in the parent directory
1231 * \retval positive number for repaired cases
1232 * \retval 0 if nothing to be repaired
1233 * \retval negative error number on failure
1235 static int lfsck_namespace_insert_normal(const struct lu_env *env,
1236 struct lfsck_component *com,
1237 struct dt_object *parent,
1238 struct dt_object *child,
1239 const struct lu_name *lname)
1241 struct lfsck_thread_info *info = lfsck_env_info(env);
1242 struct lu_attr *la = &info->lti_la;
1243 struct dt_insert_rec *rec = &info->lti_dt_rec;
1244 struct lfsck_instance *lfsck = com->lc_lfsck;
1245 /* The child and its name may be on different MDTs. */
1246 const struct lu_fid *pfid = lfsck_dto2fid(parent);
1247 const struct lu_fid *cfid = lfsck_dto2fid(child);
1248 struct dt_device *dev = lfsck->li_next;
1249 struct thandle *th = NULL;
1250 struct lfsck_lock_handle *llh = &info->lti_llh;
1251 struct lmv_mds_md_v1 *lmv = &info->lti_lmv;
1252 struct lu_buf buf = { lmv, sizeof(*lmv) };
1253 /* whether parent's LMV is deleted before insertion */
1254 bool parent_lmv_deleted = false;
1255 /* whether parent's LMV is missing */
1256 bool parent_lmv_lost = false;
1261 /* @parent/@child may be based on lfsck->li_bottom,
1262 * but here we need the object based on the lfsck->li_next. */
1264 parent = lfsck_object_locate(dev, parent);
1266 GOTO(log, rc = PTR_ERR(parent));
1268 child = lfsck_object_locate(dev, child);
1270 GOTO(log, rc = PTR_ERR(child));
1272 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1275 rc = lfsck_lock(env, lfsck, parent, lname->ln_name, llh,
1276 MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
1277 MDS_INODELOCK_XATTR, LCK_EX);
1281 rc = lfsck_object_is_shard(env, lfsck, child, lname);
1286 rc = lfsck_read_stripe_lmv(env, lfsck, parent, lmv);
1289 * To add a shard, we need to convert parent to a
1290 * plain directory by deleting its LMV, and after
1291 * insertion set it back.
1293 rc = lfsck_lmv_delete(env, lfsck, parent);
1296 parent_lmv_deleted = true;
1297 lmv->lmv_layout_version++;
1298 lfsck_lmv_header_cpu_to_le(lmv, lmv);
1299 } else if (rc == -ENODATA) {
1300 struct lu_seq_range *range = &info->lti_range;
1301 struct seq_server_site *ss = lfsck_dev_site(lfsck);
1303 rc = lfsck_read_stripe_lmv(env, lfsck, child, lmv);
1307 fld_range_set_mdt(range);
1308 rc = fld_server_lookup(env, ss->ss_server_fld,
1309 fid_seq(lfsck_dto2fid(parent)), range);
1313 parent_lmv_lost = true;
1314 lmv->lmv_magic = LMV_MAGIC;
1315 lmv->lmv_master_mdt_index = range->lsr_index;
1316 lmv->lmv_layout_version++;
1317 lfsck_lmv_header_cpu_to_le(lmv, lmv);
1323 if (unlikely(!dt_try_as_dir(env, parent)))
1324 GOTO(unlock, rc = -ENOTDIR);
1326 th = dt_trans_create(env, dev);
1328 GOTO(unlock, rc = PTR_ERR(th));
1330 rec->rec_type = lfsck_object_type(child) & S_IFMT;
1331 rec->rec_fid = cfid;
1332 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1333 (const struct dt_key *)lname->ln_name, th);
1337 if (S_ISDIR(rec->rec_type)) {
1338 rc = dt_declare_ref_add(env, parent, th);
1343 if (parent_lmv_lost) {
1344 rc = dt_declare_xattr_set(env, parent, &buf, XATTR_NAME_LMV,
1350 la->la_ctime = ktime_get_real_seconds();
1351 la->la_valid = LA_CTIME;
1352 rc = dt_declare_attr_set(env, parent, la, th);
1356 rc = dt_declare_attr_set(env, child, la, th);
1360 rc = dt_trans_start_local(env, dev, th);
1364 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1365 (const struct dt_key *)lname->ln_name, th);
1369 if (S_ISDIR(rec->rec_type)) {
1370 dt_write_lock(env, parent, 0);
1371 rc = dt_ref_add(env, parent, th);
1372 dt_write_unlock(env, parent);
1377 if (parent_lmv_lost) {
1378 rc = dt_xattr_set(env, parent, &buf, XATTR_NAME_LMV, 0, th);
1383 rc = dt_attr_set(env, parent, la, th);
1387 rc = dt_attr_set(env, child, la, th);
1389 GOTO(stop, rc = (rc == 0 ? 1 : rc));
1392 dt_trans_stop(env, dev, th);
1395 if (parent_lmv_deleted)
1396 lfsck_lmv_set(env, lfsck, parent, lmv);
1401 CDEBUG(D_LFSCK, "%s: namespace LFSCK insert object "DFID" with "
1402 "the name %s and type %o to the parent "DFID": rc = %d\n",
1403 lfsck_lfsck2name(lfsck), PFID(cfid), lname->ln_name,
1404 lfsck_object_type(child) & S_IFMT, PFID(pfid), rc);
1407 struct lfsck_namespace *ns = com->lc_file_ram;
1409 ns->ln_flags |= LF_INCONSISTENT;
1411 ns->ln_lost_dirent_repaired++;
1418 * Create the specified orphan directory.
1420 * For the case that the parent MDT-object stored in some MDT-object's
1421 * linkEA entry is lost, the LFSCK will re-create the parent object as
1422 * an orphan and insert it into .lustre/lost+found/MDTxxxx/ directory
1423 * with the name ${FID}-P-${conflict_version}.
1425 * \param[in] env pointer to the thread context
1426 * \param[in] com pointer to the lfsck component
1427 * \param[in] orphan pointer to the orphan MDT-object to be created
1428 * \param[in] lmv pointer to master LMV EA that will be set to the orphan
1430 * \retval positive number for repaired cases
1431 * \retval negative error number on failure
1433 static int lfsck_namespace_create_orphan_dir(const struct lu_env *env,
1434 struct lfsck_component *com,
1435 struct dt_object *orphan,
1436 struct lmv_mds_md_v1 *lmv)
1438 struct lfsck_thread_info *info = lfsck_env_info(env);
1439 struct lu_attr *la = &info->lti_la;
1440 struct dt_allocation_hint *hint = &info->lti_hint;
1441 struct dt_object_format *dof = &info->lti_dof;
1442 struct lu_name *cname = &info->lti_name2;
1443 struct dt_insert_rec *rec = &info->lti_dt_rec;
1444 struct lmv_mds_md_v1 *lmv2 = &info->lti_lmv2;
1445 const struct lu_fid *cfid = lfsck_dto2fid(orphan);
1447 struct lfsck_instance *lfsck = com->lc_lfsck;
1448 struct lfsck_namespace *ns = com->lc_file_ram;
1449 struct dt_device *dev = lfsck_obj2dev(orphan);
1450 struct dt_object *parent = NULL;
1451 struct thandle *th = NULL;
1452 struct lfsck_lock_handle *llh = &info->lti_llh;
1453 struct linkea_data ldata = { NULL };
1454 struct lu_buf linkea_buf;
1455 struct lu_buf lmv_buf;
1463 LASSERT(!dt_object_exists(orphan));
1465 cname->ln_name = NULL;
1466 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1469 if (dt_object_remote(orphan)) {
1470 LASSERT(lfsck->li_lpf_root_obj != NULL);
1472 idx = lfsck_find_mdt_idx_by_fid(env, lfsck, cfid);
1474 GOTO(log, rc = idx);
1476 snprintf(name, 8, "MDT%04x", idx);
1477 rc = dt_lookup(env, lfsck->li_lpf_root_obj,
1478 (struct dt_rec *)&tfid,
1479 (const struct dt_key *)name);
1481 GOTO(log, rc = (rc == -ENOENT ? -ENXIO : rc));
1483 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
1485 GOTO(log, rc = PTR_ERR(parent));
1487 if (unlikely(!dt_try_as_dir(env, parent)))
1488 GOTO(log, rc = -ENOTDIR);
1490 if (unlikely(lfsck->li_lpf_obj == NULL))
1491 GOTO(log, rc = -ENXIO);
1493 parent = lfsck->li_lpf_obj;
1496 dev = lfsck_find_dev_by_fid(env, lfsck, cfid);
1498 GOTO(log, rc = PTR_ERR(dev));
1504 namelen = snprintf(name, 31, DFID"-P-%d",
1506 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
1507 (const struct dt_key *)name);
1508 if (rc != 0 && rc != -ENOENT)
1512 rc = lfsck_lock(env, lfsck, parent, name, llh,
1513 MDS_INODELOCK_UPDATE, LCK_PW);
1517 /* Re-check whether the name conflict with othrs after taken
1519 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
1520 (const struct dt_key *)name);
1521 if (unlikely(rc == 0)) {
1529 cname->ln_name = name;
1530 cname->ln_namelen = namelen;
1532 memset(la, 0, sizeof(*la));
1533 la->la_mode = S_IFDIR | 0700;
1534 la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
1535 LA_ATIME | LA_MTIME | LA_CTIME;
1537 orphan->do_ops->do_ah_init(env, hint, parent, orphan,
1538 la->la_mode & S_IFMT);
1540 memset(dof, 0, sizeof(*dof));
1541 dof->dof_type = dt_mode_to_dft(S_IFDIR);
1543 rc = linkea_links_new(&ldata, &info->lti_linkea_buf2,
1544 cname, lfsck_dto2fid(parent));
1548 th = dt_trans_create(env, dev);
1550 GOTO(unlock1, rc = PTR_ERR(th));
1552 /* Sync the remote transaction to guarantee that the subsequent
1553 * lock against the @orphan can find the @orphan in time. */
1554 if (dt_object_remote(orphan))
1557 rc = dt_declare_create(env, orphan, la, hint, dof, th);
1561 if (unlikely(!dt_try_as_dir(env, orphan)))
1562 GOTO(stop, rc = -ENOTDIR);
1564 rc = dt_declare_ref_add(env, orphan, th);
1568 rec->rec_type = S_IFDIR;
1569 rec->rec_fid = cfid;
1570 rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
1571 (const struct dt_key *)dot, th);
1575 rec->rec_fid = lfsck_dto2fid(parent);
1576 rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
1577 (const struct dt_key *)dotdot, th);
1582 lmv->lmv_magic = LMV_MAGIC;
1583 lmv->lmv_master_mdt_index = lfsck_dev_idx(lfsck);
1584 lfsck_lmv_header_cpu_to_le(lmv2, lmv);
1585 lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
1586 rc = dt_declare_xattr_set(env, orphan, &lmv_buf, XATTR_NAME_LMV,
1592 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
1593 ldata.ld_leh->leh_len);
1594 rc = dt_declare_xattr_set(env, orphan, &linkea_buf,
1595 XATTR_NAME_LINK, 0, th);
1599 rec->rec_fid = cfid;
1600 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1601 (const struct dt_key *)name, th);
1603 rc = dt_declare_ref_add(env, parent, th);
1608 rc = dt_trans_start_local(env, dev, th);
1612 dt_write_lock(env, orphan, 0);
1613 rc = dt_create(env, orphan, la, hint, dof, th);
1617 rc = dt_ref_add(env, orphan, th);
1621 rec->rec_fid = cfid;
1622 rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1623 (const struct dt_key *)dot, th);
1627 rec->rec_fid = lfsck_dto2fid(parent);
1628 rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1629 (const struct dt_key *)dotdot, th);
1634 rc = dt_xattr_set(env, orphan, &lmv_buf, XATTR_NAME_LMV, 0, th);
1639 rc = dt_xattr_set(env, orphan, &linkea_buf,
1640 XATTR_NAME_LINK, 0, th);
1641 dt_write_unlock(env, orphan);
1645 rec->rec_fid = cfid;
1646 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1647 (const struct dt_key *)name, th);
1649 dt_write_lock(env, parent, 0);
1650 rc = dt_ref_add(env, parent, th);
1651 dt_write_unlock(env, parent);
1654 GOTO(stop, rc = (rc == 0 ? 1 : rc));
1657 dt_write_unlock(env, orphan);
1660 rc1 = dt_trans_stop(env, dev, th);
1661 if (rc1 != 0 && rc > 0)
1668 CDEBUG(D_LFSCK, "%s: namespace LFSCK create orphan dir for "
1669 "the object "DFID", name = %s: rc = %d\n",
1670 lfsck_lfsck2name(lfsck), PFID(cfid),
1671 cname->ln_name != NULL ? cname->ln_name : "<NULL>", rc);
1673 if (parent != NULL && !IS_ERR(parent) && parent != lfsck->li_lpf_obj)
1674 lfsck_object_put(env, parent);
1677 ns->ln_flags |= LF_INCONSISTENT;
1683 * Remove the specified entry from the linkEA.
1685 * Locate the linkEA entry with the given @cname and @pfid, then
1686 * remove this entry or the other entries those are repeated with
1689 * \param[in] env pointer to the thread context
1690 * \param[in] com pointer to the lfsck component
1691 * \param[in] obj pointer to the dt_object to be handled
1692 * \param[in,out]ldata pointer to the buffer that holds the linkEA
1693 * \param[in] cname the name for the child in the parent directory
1694 * \param[in] pfid the parent directory's FID for the linkEA
1695 * \param[in] next if true, then remove the first found linkEA
1696 * entry, and move the ldata->ld_lee to next entry
1698 * \retval positive number for repaired cases
1699 * \retval 0 if nothing to be repaired
1700 * \retval negative error number on failure
1702 static int lfsck_namespace_shrink_linkea(const struct lu_env *env,
1703 struct lfsck_component *com,
1704 struct dt_object *obj,
1705 struct linkea_data *ldata,
1706 struct lu_name *cname,
1707 struct lu_fid *pfid,
1710 struct lfsck_instance *lfsck = com->lc_lfsck;
1711 struct dt_device *dev = lfsck_obj2dev(obj);
1712 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
1713 struct thandle *th = NULL;
1714 struct lustre_handle lh = { 0 };
1715 struct linkea_data ldata_new = { NULL };
1716 struct lu_buf linkea_buf;
1721 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
1722 MDS_INODELOCK_UPDATE | MDS_INODELOCK_XATTR,
1728 lfsck_linkea_del_buf(ldata, cname);
1730 lfsck_namespace_filter_linkea_entry(ldata, cname, pfid,
1732 if (ldata->ld_leh->leh_reccount > 0 ||
1733 unlikely(ldata->ld_leh->leh_overflow_time)) {
1734 lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
1735 ldata->ld_leh->leh_len);
1736 buflen = linkea_buf.lb_len;
1740 th = dt_trans_create(env, dev);
1742 GOTO(unlock1, rc = PTR_ERR(th));
1745 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
1746 XATTR_NAME_LINK, 0, th);
1748 rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, th);
1752 rc = dt_trans_start_local(env, dev, th);
1756 dt_write_lock(env, obj, 0);
1757 if (unlikely(lfsck_is_dead_obj(obj)))
1758 GOTO(unlock2, rc = -ENOENT);
1760 rc = lfsck_links_read2_with_rec(env, obj, &ldata_new);
1762 GOTO(unlock2, rc = (rc == -ENODATA ? 0 : rc));
1764 /* The specified linkEA entry has been removed by race. */
1765 rc = linkea_links_find(&ldata_new, cname, pfid);
1767 GOTO(unlock2, rc = 0);
1769 if (bk->lb_param & LPF_DRYRUN)
1770 GOTO(unlock2, rc = 1);
1773 lfsck_linkea_del_buf(&ldata_new, cname);
1775 lfsck_namespace_filter_linkea_entry(&ldata_new, cname, pfid,
1779 * linkea may change because it doesn't take lock in the first read, if
1780 * it becomes larger, restart from beginning.
1782 if ((ldata_new.ld_leh->leh_reccount > 0 ||
1783 unlikely(ldata_new.ld_leh->leh_overflow_time)) &&
1784 buflen < ldata_new.ld_leh->leh_len) {
1785 dt_write_unlock(env, obj);
1786 dt_trans_stop(env, dev, th);
1787 lfsck_buf_init(&linkea_buf, ldata_new.ld_buf->lb_buf,
1788 ldata_new.ld_leh->leh_len);
1789 buflen = linkea_buf.lb_len;
1794 rc = lfsck_links_write(env, obj, &ldata_new, th);
1796 rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, th);
1798 GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
1801 dt_write_unlock(env, obj);
1804 dt_trans_stop(env, dev, th);
1807 lfsck_ibits_unlock(&lh, LCK_EX);
1810 CDEBUG(D_LFSCK, "%s: namespace LFSCK remove %s linkEA entry "
1811 "for the object: "DFID", parent "DFID", name %.*s\n",
1812 lfsck_lfsck2name(lfsck), next ? "invalid" : "redundant",
1813 PFID(lfsck_dto2fid(obj)), PFID(pfid), cname->ln_namelen,
1817 struct lfsck_namespace *ns = com->lc_file_ram;
1819 ns->ln_flags |= LF_INCONSISTENT;
1826 * Conditionally remove the specified entry from the linkEA.
1828 * Take the parent lock firstly, then check whether the specified
1829 * name entry exists or not: if yes, do nothing; otherwise, call
1830 * lfsck_namespace_shrink_linkea() to remove the linkea entry.
1832 * \param[in] env pointer to the thread context
1833 * \param[in] com pointer to the lfsck component
1834 * \param[in] parent pointer to the parent directory
1835 * \param[in] child pointer to the child object that holds the linkEA
1836 * \param[in,out]ldata pointer to the buffer that holds the linkEA
1837 * \param[in] cname the name for the child in the parent directory
1838 * \param[in] pfid the parent directory's FID for the linkEA
1840 * \retval positive number for repaired cases
1841 * \retval 0 if nothing to be repaired
1842 * \retval negative error number on failure
1844 static int lfsck_namespace_shrink_linkea_cond(const struct lu_env *env,
1845 struct lfsck_component *com,
1846 struct dt_object *parent,
1847 struct dt_object *child,
1848 struct linkea_data *ldata,
1849 struct lu_name *cname,
1850 struct lu_fid *pfid)
1852 struct lfsck_thread_info *info = lfsck_env_info(env);
1853 struct lu_fid *cfid = &info->lti_fid3;
1854 struct lfsck_lock_handle *llh = &info->lti_llh;
1858 rc = lfsck_lock(env, com->lc_lfsck, parent, cname->ln_name, llh,
1859 MDS_INODELOCK_UPDATE, LCK_PR);
1863 dt_read_lock(env, parent, 0);
1864 if (unlikely(lfsck_is_dead_obj(parent))) {
1865 dt_read_unlock(env, parent);
1867 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
1873 rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
1874 (const struct dt_key *)cname->ln_name);
1875 dt_read_unlock(env, parent);
1877 /* It is safe to release the ldlm lock, because when the logic come
1878 * here, we have got all the needed information above whether the
1879 * linkEA entry is valid or not. It is not important that others
1880 * may add new linkEA entry after the ldlm lock released. If other
1881 * has removed the specified linkEA entry by race, then it is OK,
1882 * because the subsequent lfsck_namespace_shrink_linkea() can handle
1885 if (rc == -ENOENT) {
1886 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
1895 /* The LFSCK just found some internal status of cross-MDTs
1896 * create operation. That is normal. */
1897 if (lu_fid_eq(cfid, lfsck_dto2fid(child))) {
1898 linkea_next_entry(ldata);
1903 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata, cname,
1910 * Conditionally replace name entry in the parent.
1912 * As required, the LFSCK may re-create the lost MDT-object for dangling
1913 * name entry, but such repairing may be wrong because of bad FID in the
1914 * name entry. As the LFSCK processing, the real MDT-object may be found,
1915 * then the LFSCK should check whether the former re-created MDT-object
1916 * has been modified or not, if not, then destroy it and update the name
1917 * entry in the parent to reference the real MDT-object.
1919 * \param[in] env pointer to the thread context
1920 * \param[in] com pointer to the lfsck component
1921 * \param[in] parent pointer to the parent directory
1922 * \param[in] child pointer to the MDT-object that may be the real
1923 * MDT-object corresponding to the name entry in parent
1924 * \param[in] cfid the current FID in the name entry
1925 * \param[in] cname contains the name of the child in the parent directory
1927 * \retval positive number for repaired cases
1928 * \retval 0 if nothing to be repaired
1929 * \retval negative error number on failure
1931 static int lfsck_namespace_replace_cond(const struct lu_env *env,
1932 struct lfsck_component *com,
1933 struct dt_object *parent,
1934 struct dt_object *child,
1935 const struct lu_fid *cfid,
1936 const struct lu_name *cname)
1938 struct lfsck_thread_info *info = lfsck_env_info(env);
1939 struct lu_attr *la = &info->lti_la;
1940 struct dt_insert_rec *rec = &info->lti_dt_rec;
1942 struct lfsck_instance *lfsck = com->lc_lfsck;
1943 /* The child and its name may be on different MDTs. */
1944 struct dt_device *dev = lfsck->li_next;
1945 const char *name = cname->ln_name;
1946 const struct lu_fid *pfid = lfsck_dto2fid(parent);
1947 struct dt_object *cobj = NULL;
1948 struct lfsck_lock_handle *pllh = &info->lti_llh;
1949 struct lustre_handle clh = { 0 };
1950 struct linkea_data ldata = { NULL };
1951 struct thandle *th = NULL;
1956 /* @parent/@child may be based on lfsck->li_bottom,
1957 * but here we need the object based on the lfsck->li_next. */
1959 parent = lfsck_object_locate(dev, parent);
1961 GOTO(log, rc = PTR_ERR(parent));
1963 if (unlikely(!dt_try_as_dir(env, parent)))
1964 GOTO(log, rc = -ENOTDIR);
1966 rc = lfsck_lock(env, lfsck, parent, name, pllh,
1967 MDS_INODELOCK_UPDATE, LCK_PW);
1971 if (!fid_is_sane(cfid)) {
1976 cobj = lfsck_object_find_by_dev(env, dev, cfid);
1979 if (rc == -ENOENT) {
1987 if (!dt_object_exists(cobj)) {
1992 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
1993 (const struct dt_key *)name);
1994 if (rc == -ENOENT) {
2002 /* Someone changed the name entry, cannot replace it. */
2003 if (!lu_fid_eq(cfid, &tfid))
2006 /* lock the object to be destroyed. */
2007 rc = lfsck_ibits_lock(env, lfsck, cobj, &clh,
2008 MDS_INODELOCK_UPDATE |
2009 MDS_INODELOCK_UPDATE | MDS_INODELOCK_XATTR,
2014 if (unlikely(lfsck_is_dead_obj(cobj))) {
2019 rc = dt_attr_get(env, cobj, la);
2023 /* The object has been modified by other(s), or it is not created by
2024 * LFSCK, the two cases are indistinguishable. So cannot replace it. */
2025 if (la->la_ctime != 0)
2028 if (S_ISREG(la->la_mode)) {
2029 rc = dt_xattr_get(env, cobj, &LU_BUF_NULL, XATTR_NAME_LOV);
2030 /* If someone has created related OST-object(s),
2032 if ((rc > 0) || (rc < 0 && rc != -ENODATA))
2033 GOTO(log, rc = (rc > 0 ? 0 : rc));
2037 dt_read_lock(env, child, 0);
2038 rc = lfsck_links_read2_with_rec(env, child, &ldata);
2039 dt_read_unlock(env, child);
2041 /* Someone changed the child, no need to replace. */
2048 rc = linkea_links_find(&ldata, cname, pfid);
2049 /* Someone moved the child, no need to replace. */
2053 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2056 th = dt_trans_create(env, dev);
2058 GOTO(log, rc = PTR_ERR(th));
2061 rc = dt_declare_destroy(env, cobj, th);
2066 rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
2070 rec->rec_type = S_IFDIR;
2071 rec->rec_fid = lfsck_dto2fid(child);
2072 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
2073 (const struct dt_key *)name, th);
2077 rc = dt_trans_start_local(env, dev, th);
2082 rc = dt_destroy(env, cobj, th);
2087 /* The old name entry maybe not exist. */
2088 rc = dt_delete(env, parent, (const struct dt_key *)name, th);
2089 if (rc != 0 && rc != -ENOENT)
2092 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
2093 (const struct dt_key *)name, th);
2095 GOTO(stop, rc = (rc == 0 ? 1 : rc));
2098 dt_trans_stop(env, dev, th);
2101 lfsck_ibits_unlock(&clh, LCK_EX);
2104 if (cobj != NULL && !IS_ERR(cobj))
2105 lfsck_object_put(env, cobj);
2107 CDEBUG(D_LFSCK, "%s: namespace LFSCK conditionally destroy the "
2108 "object "DFID" because of conflict with the object "DFID
2109 " under the parent "DFID" with name %s: rc = %d\n",
2110 lfsck_lfsck2name(lfsck), PFID(cfid),
2111 PFID(lfsck_dto2fid(child)), PFID(pfid), name, rc);
2117 * Overwrite the linkEA for the object with the given ldata.
2119 * The caller should take the ldlm lock before the calling.
2121 * \param[in] env pointer to the thread context
2122 * \param[in] com pointer to the lfsck component
2123 * \param[in] obj pointer to the dt_object to be handled
2124 * \param[in] ldata pointer to the new linkEA data
2126 * \retval positive number for repaired cases
2127 * \retval 0 if nothing to be repaired
2128 * \retval negative error number on failure
2130 int lfsck_namespace_rebuild_linkea(const struct lu_env *env,
2131 struct lfsck_component *com,
2132 struct dt_object *obj,
2133 struct linkea_data *ldata)
2135 struct lfsck_instance *lfsck = com->lc_lfsck;
2136 struct dt_device *dev = lfsck_obj2dev(obj);
2137 struct thandle *th = NULL;
2138 struct lu_buf linkea_buf;
2142 th = dt_trans_create(env, dev);
2144 GOTO(log, rc = PTR_ERR(th));
2146 lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
2147 ldata->ld_leh->leh_len);
2148 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
2149 XATTR_NAME_LINK, 0, th);
2153 rc = dt_trans_start_local(env, dev, th);
2157 dt_write_lock(env, obj, 0);
2158 if (unlikely(lfsck_is_dead_obj(obj)))
2159 GOTO(unlock, rc = 0);
2161 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2162 GOTO(unlock, rc = 1);
2164 rc = dt_xattr_set(env, obj, &linkea_buf,
2165 XATTR_NAME_LINK, 0, th);
2167 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2170 dt_write_unlock(env, obj);
2173 dt_trans_stop(env, dev, th);
2176 CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild linkEA for the "
2177 "object "DFID": rc = %d\n",
2178 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
2181 struct lfsck_namespace *ns = com->lc_file_ram;
2183 ns->ln_flags |= LF_INCONSISTENT;
2190 * Repair invalid name entry.
2192 * If the name entry contains invalid information, such as bad file type
2193 * or (and) corrupted object FID, then either remove the name entry or
2194 * udpate the name entry with the given (right) information.
2196 * \param[in] env pointer to the thread context
2197 * \param[in] com pointer to the lfsck component
2198 * \param[in] parent pointer to the parent directory
2199 * \param[in] child pointer to the object referenced by the name entry
2200 * \param[in] name the old name of the child under the parent directory
2201 * \param[in] name2 the new name of the child under the parent directory
2202 * \param[in] type the type claimed by the name entry
2203 * \param[in] update update the name entry if true; otherwise, remove it
2204 * \param[in] dec decrease the parent nlink count if true
2206 * \retval positive number for repaired successfully
2207 * \retval 0 if nothing to be repaired
2208 * \retval negative error number on failure
2210 int lfsck_namespace_repair_dirent(const struct lu_env *env,
2211 struct lfsck_component *com,
2212 struct dt_object *parent,
2213 struct dt_object *child,
2214 const char *name, const char *name2,
2215 __u16 type, bool update, bool dec)
2217 struct lfsck_thread_info *info = lfsck_env_info(env);
2218 struct dt_insert_rec *rec = &info->lti_dt_rec;
2219 const struct lu_fid *pfid = lfsck_dto2fid(parent);
2220 struct lu_fid cfid = {0};
2222 struct lfsck_instance *lfsck = com->lc_lfsck;
2223 struct dt_device *dev = lfsck->li_next;
2224 struct thandle *th = NULL;
2225 struct lfsck_lock_handle *llh = &info->lti_llh;
2226 struct lustre_handle lh = { 0 };
2231 cfid = *lfsck_dto2fid(child);
2232 parent = lfsck_object_locate(dev, parent);
2234 GOTO(log, rc = PTR_ERR(parent));
2236 if (unlikely(!dt_try_as_dir(env, parent)))
2237 GOTO(log, rc = -ENOTDIR);
2239 if (!update || strcmp(name, name2) == 0)
2240 rc = lfsck_lock(env, lfsck, parent, name, llh,
2241 MDS_INODELOCK_UPDATE, LCK_PW);
2243 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
2244 MDS_INODELOCK_UPDATE, LCK_PW);
2248 th = dt_trans_create(env, dev);
2250 GOTO(unlock1, rc = PTR_ERR(th));
2252 rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
2257 rec->rec_type = lfsck_object_type(child) & S_IFMT;
2258 LASSERT(!fid_is_zero(&cfid));
2259 rec->rec_fid = &cfid;
2260 rc = dt_declare_insert(env, parent,
2261 (const struct dt_rec *)rec,
2262 (const struct dt_key *)name2, th);
2267 if (dec && S_ISDIR(type)) {
2268 rc = dt_declare_ref_del(env, parent, th);
2273 rc = dt_trans_start_local(env, dev, th);
2278 dt_write_lock(env, parent, 0);
2279 rc = dt_lookup(env, dt_object_child(parent), (struct dt_rec *)&tfid,
2280 (const struct dt_key *)name);
2281 /* Someone has removed the bad name entry by race. */
2283 GOTO(unlock2, rc = 0);
2288 /* Someone has removed the bad name entry and reused it for other
2289 * object by race. */
2290 if (!lu_fid_eq(&tfid, &cfid))
2291 GOTO(unlock2, rc = 0);
2293 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2294 GOTO(unlock2, rc = 1);
2296 rc = dt_delete(env, parent, (const struct dt_key *)name, th);
2301 rc = dt_insert(env, parent,
2302 (const struct dt_rec *)rec,
2303 (const struct dt_key *)name2, th);
2308 if (dec && S_ISDIR(type)) {
2309 rc = dt_ref_del(env, parent, th);
2314 GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
2317 dt_write_unlock(env, parent);
2320 dt_trans_stop(env, dev, th);
2322 /* We are not sure whether the child will become orphan or not.
2323 * Record it in the LFSCK trace file for further checking in
2324 * the second-stage scanning. */
2325 if (!update && !dec && child && rc == 0)
2326 lfsck_namespace_trace_update(env, com, &cfid,
2327 LNTF_CHECK_LINKEA, true);
2330 /* It is harmless even if unlock the unused lock_handle */
2331 lfsck_ibits_unlock(&lh, LCK_PW);
2335 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found bad name "
2336 "entry for: parent "DFID", child "DFID", name %s, type "
2337 "in name entry %o, type claimed by child %o. repair it "
2338 "by %s with new name2 %s: rc = %d\n",
2339 lfsck_lfsck2name(lfsck), PFID(pfid), PFID(&cfid),
2340 name, type, update ? lfsck_object_type(child) : 0,
2341 update ? "updating" : "removing", name2, rc);
2344 struct lfsck_namespace *ns = com->lc_file_ram;
2346 ns->ln_flags |= LF_INCONSISTENT;
2353 * Update the ".." name entry for the given object.
2355 * The object's ".." is corrupted, this function will update the ".." name
2356 * entry with the given pfid, and the linkEA with the given ldata.
2358 * The caller should take the ldlm lock before the calling.
2360 * \param[in] env pointer to the thread context
2361 * \param[in] com pointer to the lfsck component
2362 * \param[in] obj pointer to the dt_object to be handled
2363 * \param[in] pfid the new fid for the object's ".." name entry
2364 * \param[in] cname the name for the @obj in the parent directory
2366 * \retval positive number for repaired cases
2367 * \retval 0 if nothing to be repaired
2368 * \retval negative error number on failure
2370 static int lfsck_namespace_repair_unmatched_pairs(const struct lu_env *env,
2371 struct lfsck_component *com,
2372 struct dt_object *obj,
2373 const struct lu_fid *pfid,
2374 struct lu_name *cname)
2376 struct lfsck_thread_info *info = lfsck_env_info(env);
2377 struct dt_insert_rec *rec = &info->lti_dt_rec;
2378 struct lfsck_instance *lfsck = com->lc_lfsck;
2379 struct dt_device *dev = lfsck_obj2dev(obj);
2380 struct thandle *th = NULL;
2381 struct linkea_data ldata = { NULL };
2382 struct lu_buf linkea_buf;
2386 LASSERT(!dt_object_remote(obj));
2387 LASSERT(S_ISDIR(lfsck_object_type(obj)));
2389 rc = linkea_links_new(&ldata, &info->lti_big_buf, cname, pfid);
2393 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
2394 ldata.ld_leh->leh_len);
2396 th = dt_trans_create(env, dev);
2398 GOTO(log, rc = PTR_ERR(th));
2400 rc = dt_declare_delete(env, obj, (const struct dt_key *)dotdot, th);
2404 rec->rec_type = S_IFDIR;
2405 rec->rec_fid = pfid;
2406 rc = dt_declare_insert(env, obj, (const struct dt_rec *)rec,
2407 (const struct dt_key *)dotdot, th);
2411 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
2412 XATTR_NAME_LINK, 0, th);
2416 rc = dt_trans_start_local(env, dev, th);
2420 dt_write_lock(env, obj, 0);
2421 if (unlikely(lfsck_is_dead_obj(obj)))
2422 GOTO(unlock, rc = 0);
2424 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2425 GOTO(unlock, rc = 1);
2427 /* The old ".." name entry maybe not exist. */
2428 dt_delete(env, obj, (const struct dt_key *)dotdot, th);
2430 rc = dt_insert(env, obj, (const struct dt_rec *)rec,
2431 (const struct dt_key *)dotdot, th);
2435 rc = lfsck_links_write(env, obj, &ldata, th);
2437 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2440 dt_write_unlock(env, obj);
2443 dt_trans_stop(env, dev, th);
2446 CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild dotdot name entry for "
2447 "the object "DFID", new parent "DFID": rc = %d\n",
2448 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)),
2452 struct lfsck_namespace *ns = com->lc_file_ram;
2454 ns->ln_flags |= LF_INCONSISTENT;
2461 * Handle orphan @obj during Double Scan Directory.
2463 * Remove the @obj's current (invalid) linkEA entries, and insert
2464 * it in the directory .lustre/lost+found/MDTxxxx/ with the name:
2465 * ${FID}-${PFID}-D-${conflict_version}
2467 * The caller should take the ldlm lock before the calling.
2469 * \param[in] env pointer to the thread context
2470 * \param[in] com pointer to the lfsck component
2471 * \param[in] obj pointer to the orphan object to be handled
2472 * \param[in] pfid the new fid for the object's ".." name entry
2473 * \param[in,out] lh ldlm lock handler for the given @obj
2474 * \param[out] type to tell the caller what the inconsistency is
2476 * \retval positive number for repaired cases
2477 * \retval 0 if nothing to be repaired
2478 * \retval negative error number on failure
2481 lfsck_namespace_dsd_orphan(const struct lu_env *env,
2482 struct lfsck_component *com,
2483 struct dt_object *obj,
2484 const struct lu_fid *pfid,
2485 struct lustre_handle *lh,
2486 enum lfsck_namespace_inconsistency_type *type)
2488 struct lfsck_thread_info *info = lfsck_env_info(env);
2489 struct lfsck_namespace *ns = com->lc_file_ram;
2493 /* Remove the unrecognized linkEA. */
2494 rc = lfsck_namespace_links_remove(env, com, obj);
2495 lfsck_ibits_unlock(lh, LCK_EX);
2496 if (rc < 0 && rc != -ENODATA)
2499 *type = LNIT_MUL_REF;
2501 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
2502 * ever tried to verify some remote MDT-object that resides on this
2503 * MDT, but this MDT failed to respond such request. So means there
2504 * may be some remote name entry on other MDT that references this
2505 * object with another name, so we cannot know whether this linkEA
2506 * is valid or not. So keep it there and maybe resolved when next
2508 if (ns->ln_flags & LF_INCOMPLETE)
2511 /* The unique linkEA is invalid, even if the ".." name entry may be
2512 * valid, we still cannot know via which name entry this directory
2513 * will be referenced. Then handle it as pure orphan. */
2514 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2515 "-"DFID, PFID(pfid));
2516 rc = lfsck_namespace_insert_orphan(env, com, obj,
2517 info->lti_tmpbuf, "D", NULL);
2523 * Double Scan Directory object for single linkEA entry case.
2525 * The given @child has unique linkEA entry. If the linkEA entry is valid,
2526 * then check whether the name is in the namespace or not, if not, add the
2527 * missing name entry back to namespace. If the linkEA entry is invalid,
2528 * then remove it and insert the @child in the .lustre/lost+found/MDTxxxx/
2531 * \param[in] env pointer to the thread context
2532 * \param[in] com pointer to the lfsck component
2533 * \param[in] child pointer to the directory to be double scanned
2534 * \param[in] pfid the FID corresponding to the ".." entry
2535 * \param[in] ldata pointer to the linkEA data for the given @child
2536 * \param[in,out] lh ldlm lock handler for the given @child
2537 * \param[out] type to tell the caller what the inconsistency is
2538 * \param[in] retry if found inconsistency, but the caller does not hold
2539 * ldlm lock on the @child, then set @retry as true
2540 * \param[in] unknown set if does not know how to repair the inconsistency
2542 * \retval positive number for repaired cases
2543 * \retval 0 if nothing to be repaired
2544 * \retval negative error number on failure
2547 lfsck_namespace_dsd_single(const struct lu_env *env,
2548 struct lfsck_component *com,
2549 struct dt_object *child,
2550 const struct lu_fid *pfid,
2551 struct linkea_data *ldata,
2552 struct lustre_handle *lh,
2553 enum lfsck_namespace_inconsistency_type *type,
2554 bool *retry, bool *unknown)
2556 struct lfsck_thread_info *info = lfsck_env_info(env);
2557 struct lu_name *cname = &info->lti_name;
2558 const struct lu_fid *cfid = lfsck_dto2fid(child);
2560 struct lfsck_namespace *ns = com->lc_file_ram;
2561 struct lfsck_instance *lfsck = com->lc_lfsck;
2562 struct dt_object *parent = NULL;
2563 struct lmv_mds_md_v1 *lmv;
2567 rc = lfsck_namespace_unpack_linkea_entry(ldata, cname, &tfid,
2569 sizeof(info->lti_key));
2570 /* The unique linkEA entry with bad parent will be handled as orphan. */
2572 if (!lustre_handle_is_used(lh) && retry != NULL)
2575 rc = lfsck_namespace_dsd_orphan(env, com, child,
2581 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
2583 GOTO(out, rc = PTR_ERR(parent));
2585 /* We trust the unique linkEA entry in spite of whether it matches the
2586 * ".." name entry or not. Because even if the linkEA entry is wrong
2587 * and the ".." name entry is right, we still cannot know via which
2588 * name entry the child will be referenced, since all known entries
2589 * have been verified during the first-stage scanning. */
2590 if (!dt_object_exists(parent)) {
2591 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2592 * has ever tried to verify some remote MDT-object that resides
2593 * on this MDT, but this MDT failed to respond such request. So
2594 * means there may be some remote name entry on other MDT that
2595 * references this object with another name, so we cannot know
2596 * whether this linkEA is valid or not. So keep it there and
2597 * maybe resolved when next LFSCK run. */
2598 if (ns->ln_flags & LF_INCOMPLETE)
2601 if (!lustre_handle_is_used(lh) && retry != NULL) {
2607 lfsck_ibits_unlock(lh, LCK_EX);
2610 lmv = &info->lti_lmv;
2611 rc = lfsck_read_stripe_lmv(env, lfsck, child, lmv);
2612 if (rc != 0 && rc != -ENODATA)
2615 if (rc == -ENODATA || lmv->lmv_magic != LMV_MAGIC_STRIPE) {
2617 } else if (lfsck_shard_name_to_index(env,
2618 cname->ln_name, cname->ln_namelen,
2619 S_IFDIR, cfid) < 0) {
2620 /* It is an invalid name entry, we
2621 * cannot trust the parent also. */
2622 rc = lfsck_namespace_shrink_linkea(env, com, child,
2623 ldata, cname, &tfid, true);
2627 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2628 "-"DFID, PFID(pfid));
2629 rc = lfsck_namespace_insert_orphan(env, com, child,
2630 info->lti_tmpbuf, "S", NULL);
2635 /* Create the lost parent as an orphan. */
2636 rc = lfsck_namespace_create_orphan_dir(env, com, parent, lmv);
2638 /* Add the missing name entry to the parent. */
2639 rc = lfsck_namespace_insert_normal(env, com, parent,
2641 if (unlikely(rc == -EEXIST)) {
2642 /* Unfortunately, someone reused the name
2643 * under the parent by race. So we have
2644 * to remove the linkEA entry from
2645 * current child object. It means that the
2646 * LFSCK cannot recover the system
2647 * totally back to its original status,
2648 * but it is necessary to make the
2649 * current system to be consistent. */
2650 rc = lfsck_namespace_shrink_linkea(env,
2652 cname, &tfid, true);
2654 snprintf(info->lti_tmpbuf,
2655 sizeof(info->lti_tmpbuf),
2656 "-"DFID, PFID(pfid));
2657 rc = lfsck_namespace_insert_orphan(env,
2658 com, child, info->lti_tmpbuf,
2665 } /* !dt_object_exists(parent) */
2667 /* The unique linkEA entry with bad parent will be handled as orphan. */
2668 if (unlikely(!dt_try_as_dir(env, parent))) {
2669 if (!lustre_handle_is_used(lh) && retry != NULL)
2672 rc = lfsck_namespace_dsd_orphan(env, com, child,
2678 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
2679 (const struct dt_key *)cname->ln_name);
2680 if (rc == -ENOENT) {
2681 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2682 * has ever tried to verify some remote MDT-object that resides
2683 * on this MDT, but this MDT failed to respond such request. So
2684 * means there may be some remote name entry on other MDT that
2685 * references this object with another name, so we cannot know
2686 * whether this linkEA is valid or not. So keep it there and
2687 * maybe resolved when next LFSCK run. */
2688 if (ns->ln_flags & LF_INCOMPLETE)
2691 if (!lustre_handle_is_used(lh) && retry != NULL) {
2697 lfsck_ibits_unlock(lh, LCK_EX);
2698 rc = lfsck_namespace_check_name(env, lfsck, parent, child,
2706 /* It is an invalid name entry, drop it. */
2707 if (unlikely(rc > 0)) {
2708 rc = lfsck_namespace_shrink_linkea(env, com, child,
2709 ldata, cname, &tfid, true);
2711 snprintf(info->lti_tmpbuf,
2712 sizeof(info->lti_tmpbuf),
2713 "-"DFID, PFID(pfid));
2714 rc = lfsck_namespace_insert_orphan(env, com,
2715 child, info->lti_tmpbuf, "D", NULL);
2721 /* Add the missing name entry back to the namespace. */
2722 rc = lfsck_namespace_insert_normal(env, com, parent, child,
2724 if (unlikely(rc == -ESTALE))
2725 /* It may happen when the remote object has been
2726 * removed, but the local MDT is not aware of that. */
2729 if (unlikely(rc == -EEXIST)) {
2730 /* Unfortunately, someone reused the name under the
2731 * parent by race. So we have to remove the linkEA
2732 * entry from current child object. It means that the
2733 * LFSCK cannot recover the system totally back to
2734 * its original status, but it is necessary to make
2735 * the current system to be consistent.
2737 * It also may be because of the LFSCK found some
2738 * internal status of create operation. Under such
2739 * case, nothing to be done. */
2740 rc = lfsck_namespace_shrink_linkea_cond(env, com,
2741 parent, child, ldata, cname, &tfid);
2743 snprintf(info->lti_tmpbuf,
2744 sizeof(info->lti_tmpbuf),
2745 "-"DFID, PFID(pfid));
2746 rc = lfsck_namespace_insert_orphan(env, com,
2747 child, info->lti_tmpbuf, "D", NULL);
2752 } /* rc == -ENOENT */
2757 if (!lu_fid_eq(&tfid, cfid)) {
2758 if (!lustre_handle_is_used(lh) && retry != NULL) {
2764 lfsck_ibits_unlock(lh, LCK_EX);
2765 /* The name entry references another MDT-object that
2766 * may be created by the LFSCK for repairing dangling
2767 * name entry. Try to replace it. */
2768 rc = lfsck_namespace_replace_cond(env, com, parent, child,
2771 rc = lfsck_namespace_dsd_orphan(env, com, child,
2777 /* Zero FID may because the remote directroy object has invalid linkEA,
2778 * or lost linkEA. Under such case, the LFSCK on this MDT does not know
2779 * how to repair the inconsistency, but the namespace LFSCK on the MDT
2780 * where its name entry resides may has more information (name, FID) to
2781 * repair such inconsistency. So here, keep the inconsistency to avoid
2782 * some imporper repairing. */
2783 if (fid_is_zero(pfid)) {
2790 /* The ".." name entry is wrong, update it. */
2791 if (!lu_fid_eq(pfid, lfsck_dto2fid(parent))) {
2792 if (!lustre_handle_is_used(lh) && retry != NULL) {
2798 *type = LNIT_UNMATCHED_PAIRS;
2799 rc = lfsck_namespace_repair_unmatched_pairs(env, com, child,
2800 lfsck_dto2fid(parent), cname);
2806 if (parent != NULL && !IS_ERR(parent))
2807 lfsck_object_put(env, parent);
2813 * Double Scan Directory object for multiple linkEA entries case.
2815 * The given @child has multiple linkEA entries. There is at most one linkEA
2816 * entry will be valid, all the others will be removed. Firstly, the function
2817 * will try to find out the linkEA entry for which the name entry exists under
2818 * the given parent (@pfid). If there is no linkEA entry that matches the given
2819 * ".." name entry, then tries to find out the first linkEA entry that both the
2820 * parent and the name entry exist to rebuild a new ".." name entry.
2822 * \param[in] env pointer to the thread context
2823 * \param[in] com pointer to the lfsck component
2824 * \param[in] child pointer to the directory to be double scanned
2825 * \param[in] pfid the FID corresponding to the ".." entry
2826 * \param[in] ldata pointer to the linkEA data for the given @child
2827 * \param[in,out] lh ldlm lock handler for the given @child
2828 * \param[out] type to tell the caller what the inconsistency is
2829 * \param[in] lpf true if the ".." entry is under lost+found/MDTxxxx/
2830 * \param[in] unknown set if does not know how to repair the inconsistency
2832 * \retval positive number for repaired cases
2833 * \retval 0 if nothing to be repaired
2834 * \retval negative error number on failure
2837 lfsck_namespace_dsd_multiple(const struct lu_env *env,
2838 struct lfsck_component *com,
2839 struct dt_object *child,
2840 const struct lu_fid *pfid,
2841 struct linkea_data *ldata,
2842 struct lustre_handle *lh,
2843 enum lfsck_namespace_inconsistency_type *type,
2844 bool lpf, bool *unknown)
2846 struct lfsck_thread_info *info = lfsck_env_info(env);
2847 struct lu_name *cname = &info->lti_name;
2848 const struct lu_fid *cfid = lfsck_dto2fid(child);
2849 struct lu_fid *pfid2 = &info->lti_fid3;
2851 struct lfsck_namespace *ns = com->lc_file_ram;
2852 struct lfsck_instance *lfsck = com->lc_lfsck;
2853 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
2854 struct dt_object *parent = NULL;
2855 struct linkea_data ldata_new = { NULL };
2856 int dirent_count = 0;
2862 while (ldata->ld_lee != NULL) {
2863 rc = lfsck_namespace_unpack_linkea_entry(ldata, cname, &tfid,
2865 sizeof(info->lti_key));
2866 /* Drop invalid linkEA entry. */
2868 lfsck_linkea_del_buf(ldata, cname);
2872 /* Drop repeated linkEA entries. */
2873 lfsck_namespace_filter_linkea_entry(ldata, cname, &tfid, true);
2875 /* If current dotdot is the .lustre/lost+found/MDTxxxx/,
2876 * then it is possible that: the directry object has ever
2877 * been lost, but its name entry was there. In the former
2878 * LFSCK run, during the first-stage scanning, the LFSCK
2879 * found the dangling name entry, but it did not recreate
2880 * the lost object, and when moved to the second-stage
2881 * scanning, some children objects of the lost directory
2882 * object were found, then the LFSCK recreated such lost
2883 * directory object as an orphan.
2885 * When the LFSCK runs again, if the dangling name is still
2886 * there, the LFSCK should move the orphan directory object
2887 * back to the normal namespace. */
2888 if (!lpf && !fid_is_zero(pfid) &&
2889 !lu_fid_eq(pfid, &tfid) && once) {
2890 linkea_next_entry(ldata);
2894 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
2896 RETURN(PTR_ERR(parent));
2898 if (!dt_object_exists(parent)) {
2899 lfsck_object_put(env, parent);
2900 if (ldata->ld_leh->leh_reccount > 1) {
2901 /* If it is NOT the last linkEA entry, then
2902 * there is still other chance to make the
2903 * child to be visible via other parent, then
2904 * remove this linkEA entry. */
2905 lfsck_linkea_del_buf(ldata, cname);
2912 /* The linkEA entry with bad parent will be removed. */
2913 if (unlikely(!dt_try_as_dir(env, parent))) {
2914 lfsck_object_put(env, parent);
2915 lfsck_linkea_del_buf(ldata, cname);
2919 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
2920 (const struct dt_key *)cname->ln_name);
2921 *pfid2 = *lfsck_dto2fid(parent);
2922 if (rc == -ENOENT) {
2923 lfsck_object_put(env, parent);
2924 linkea_next_entry(ldata);
2929 lfsck_object_put(env, parent);
2934 if (lu_fid_eq(&tfid, cfid)) {
2935 lfsck_object_put(env, parent);
2936 /* If the parent (that is declared via linkEA entry)
2937 * directory contains the specified child, but such
2938 * parent does not match the dotdot name entry, then
2939 * trust the linkEA. */
2940 if (!fid_is_zero(pfid) && !lu_fid_eq(pfid, pfid2)) {
2941 *type = LNIT_UNMATCHED_PAIRS;
2942 rc = lfsck_namespace_repair_unmatched_pairs(env,
2943 com, child, pfid2, cname);
2949 /* It is the most common case that we find the
2950 * name entry corresponding to the linkEA entry
2951 * that matches the ".." name entry. */
2952 rc = linkea_links_new(&ldata_new, &info->lti_big_buf,
2957 rc = lfsck_namespace_rebuild_linkea(env, com, child,
2962 lfsck_linkea_del_buf(ldata, cname);
2963 linkea_first_entry(ldata);
2964 /* There may be some invalid dangling name entries under
2965 * other parent directories, remove all of them. */
2966 while (ldata->ld_lee != NULL) {
2967 rc = lfsck_namespace_unpack_linkea_entry(ldata,
2968 cname, &tfid, info->lti_key,
2969 sizeof(info->lti_key));
2973 parent = lfsck_object_find_bottom(env, lfsck,
2975 if (IS_ERR(parent)) {
2976 rc = PTR_ERR(parent);
2977 if (rc != -ENOENT &&
2978 bk->lb_param & LPF_FAILOUT)
2984 if (!dt_object_exists(parent)) {
2985 lfsck_object_put(env, parent);
2989 rc = lfsck_namespace_repair_dirent(env, com,
2990 parent, child, cname->ln_name,
2991 cname->ln_name, S_IFDIR, false, true);
2992 lfsck_object_put(env, parent);
2994 if (bk->lb_param & LPF_FAILOUT)
3003 lfsck_linkea_del_buf(ldata, cname);
3006 ns->ln_dirent_repaired += dirent_count;
3009 } /* lu_fid_eq(&tfid, lfsck_dto2fid(child)) */
3011 lfsck_ibits_unlock(lh, LCK_EX);
3012 /* The name entry references another MDT-object that may be
3013 * created by the LFSCK for repairing dangling name entry.
3014 * Try to replace it. */
3015 rc = lfsck_namespace_replace_cond(env, com, parent, child,
3017 lfsck_object_put(env, parent);
3024 lfsck_linkea_del_buf(ldata, cname);
3025 } /* while (ldata->ld_lee != NULL) */
3027 /* If there is still linkEA overflow, return. */
3028 if (unlikely(ldata->ld_leh->leh_overflow_time))
3031 linkea_first_entry(ldata);
3032 if (ldata->ld_leh->leh_reccount == 1) {
3033 rc = lfsck_namespace_dsd_single(env, com, child, pfid, ldata,
3034 lh, type, NULL, unknown);
3039 /* All linkEA entries are invalid and removed, then handle the @child
3041 if (ldata->ld_leh->leh_reccount == 0) {
3042 rc = lfsck_namespace_dsd_orphan(env, com, child, pfid, lh,
3048 /* If the dangling name entry for the orphan directory object has
3049 * been remvoed, then just check whether the directory object is
3050 * still under the .lustre/lost+found/MDTxxxx/ or not. */
3056 /* There is no linkEA entry that matches the ".." name entry. Find
3057 * the first linkEA entry that both parent and name entry exist to
3058 * rebuild a new ".." name entry. */
3068 * Repair the object's nlink attribute.
3070 * If all the known name entries have been verified, then the object's hard
3071 * link attribute should match the object's linkEA entries count unless the
3072 * object's has too many hard link to be recorded in the linkEA. Such cases
3073 * should have been marked in the LFSCK trace file. Otherwise, trust the
3074 * linkEA to update the object's nlink attribute.
3076 * \param[in] env pointer to the thread context
3077 * \param[in] com pointer to the lfsck component
3078 * \param[in] obj pointer to the dt_object to be handled
3079 * \param[in,out] la pointer to buffer to object's attribute before
3080 * and after the repairing
3082 * \retval positive number for repaired cases
3083 * \retval 0 if nothing to be repaired
3084 * \retval negative error number on failure
3086 static int lfsck_namespace_repair_nlink(const struct lu_env *env,
3087 struct lfsck_component *com,
3088 struct dt_object *obj,
3091 struct lfsck_namespace *ns = com->lc_file_ram;
3092 struct lfsck_instance *lfsck = com->lc_lfsck;
3093 struct dt_device *dev = lfsck_obj2dev(obj);
3094 const struct lu_fid *cfid = lfsck_dto2fid(obj);
3095 struct thandle *th = NULL;
3096 struct linkea_data ldata = { NULL };
3097 struct lustre_handle lh = { 0 };
3098 __u32 old = la->la_nlink;
3102 LASSERT(!dt_object_remote(obj));
3104 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
3105 MDS_INODELOCK_UPDATE, LCK_PW);
3109 th = dt_trans_create(env, dev);
3111 GOTO(log, rc = PTR_ERR(th));
3113 la->la_valid = LA_NLINK;
3114 rc = dt_declare_attr_set(env, obj, la, th);
3118 rc = dt_trans_start_local(env, dev, th);
3122 dt_write_lock(env, obj, 0);
3123 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
3124 * ever tried to verify some remote MDT-object that resides on this
3125 * MDT, but this MDT failed to respond such request. So means there
3126 * may be some remote name entry on other MDT that references this
3127 * object with another name, so we cannot know whether this linkEA
3128 * is valid or not. So keep it there and maybe resolved when next
3130 if (ns->ln_flags & LF_INCOMPLETE)
3131 GOTO(unlock, rc = 0);
3133 rc = dt_attr_get(env, obj, la);
3135 GOTO(unlock, rc = (rc == -ENOENT ? 0 : rc));
3137 rc = lfsck_links_read2_with_rec(env, obj, &ldata);
3139 GOTO(unlock, rc = (rc == -ENODATA ? 0 : rc));
3141 /* XXX: Currently, we only update the nlink attribute if the known
3142 * linkEA entries is larger than the nlink attribute. That is
3144 if (la->la_nlink >= ldata.ld_leh->leh_reccount ||
3145 unlikely(la->la_nlink == 0 ||
3146 ldata.ld_leh->leh_overflow_time))
3147 GOTO(unlock, rc = 0);
3149 la->la_nlink = ldata.ld_leh->leh_reccount;
3150 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
3151 GOTO(unlock, rc = 1);
3153 rc = dt_attr_set(env, obj, la, th);
3155 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
3158 dt_write_unlock(env, obj);
3161 dt_trans_stop(env, dev, th);
3164 lfsck_ibits_unlock(&lh, LCK_PW);
3165 CDEBUG(D_LFSCK, "%s: namespace LFSCK repaired the object "DFID"'s "
3166 "nlink count from %u to %u: rc = %d\n",
3167 lfsck_lfsck2name(lfsck), PFID(cfid), old, la->la_nlink, rc);
3170 ns->ln_flags |= LF_INCONSISTENT;
3176 * Double scan the directory object for namespace LFSCK.
3178 * This function will verify the <parent, child> pairs in the namespace tree:
3179 * the parent references the child via some name entry that should be in the
3180 * child's linkEA entry, the child should back references the parent via its
3183 * The LFSCK will scan every linkEA entry in turn until find out the first
3184 * matched pairs. If found, then all other linkEA entries will be dropped.
3185 * If all the linkEA entries cannot match the ".." name entry, then there
3186 * are serveral possible cases:
3188 * 1) If there is only one linkEA entry, then trust it as long as the PFID
3189 * in the linkEA entry is valid.
3191 * 2) If there are multiple linkEA entries, then try to find the linkEA
3192 * that matches the ".." name entry. If found, then all other entries
3193 * are invalid; otherwise, it is quite possible that the ".." name entry
3194 * is corrupted. Under such case, the LFSCK will rebuild the ".." name
3195 * entry according to the first valid linkEA entry (both the parent and
3196 * the name entry should exist).
3198 * 3) If the directory object has no (valid) linkEA entry, then the
3199 * directory object will be handled as pure orphan and inserted
3200 * in the .lustre/lost+found/MDTxxxx/ with the name:
3201 * ${self_FID}-${PFID}-D-${conflict_version}
3203 * \param[in] env pointer to the thread context
3204 * \param[in] com pointer to the lfsck component
3205 * \param[in] child pointer to the directory object to be handled
3206 * \param[in] flags to indicate the specical checking on the @child
3208 * \retval positive number for repaired cases
3209 * \retval 0 if nothing to be repaired
3210 * \retval negative error number on failure
3212 static int lfsck_namespace_double_scan_dir(const struct lu_env *env,
3213 struct lfsck_component *com,
3214 struct dt_object *child, __u8 flags)
3216 struct lfsck_thread_info *info = lfsck_env_info(env);
3217 const struct lu_fid *cfid = lfsck_dto2fid(child);
3218 struct lu_fid *pfid = &info->lti_fid2;
3219 struct lfsck_namespace *ns = com->lc_file_ram;
3220 struct lfsck_instance *lfsck = com->lc_lfsck;
3221 struct lustre_handle lh = { 0 };
3222 struct linkea_data ldata = { NULL };
3223 bool unknown = false;
3226 enum lfsck_namespace_inconsistency_type type = LNIT_BAD_LINKEA;
3230 LASSERT(!dt_object_remote(child));
3232 if (flags & LNTF_UNCERTAIN_LMV) {
3233 if (flags & LNTF_RECHECK_NAME_HASH) {
3234 rc = lfsck_namespace_scan_shard(env, com, child);
3238 ns->ln_striped_shards_scanned++;
3240 ns->ln_striped_shards_skipped++;
3244 flags &= ~(LNTF_RECHECK_NAME_HASH | LNTF_UNCERTAIN_LMV);
3248 if (flags & (LNTF_CHECK_LINKEA | LNTF_CHECK_PARENT) &&
3249 !(lfsck->li_bookmark_ram.lb_param & LPF_ALL_TGT)) {
3251 "%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",
3252 lfsck_lfsck2name(lfsck), PFID(cfid));
3257 if (unlikely(!dt_try_as_dir(env, child)))
3258 GOTO(out, rc = -ENOTDIR);
3260 /* We only take ldlm lock on the @child when required. When the
3261 * logic comes here for the first time, it is always false. */
3265 rc = lfsck_ibits_lock(env, lfsck, child, &lh,
3266 MDS_INODELOCK_UPDATE |
3267 MDS_INODELOCK_XATTR, LCK_EX);
3272 dt_read_lock(env, child, 0);
3273 if (unlikely(lfsck_is_dead_obj(child))) {
3274 dt_read_unlock(env, child);
3279 rc = dt_lookup(env, child, (struct dt_rec *)pfid,
3280 (const struct dt_key *)dotdot);
3282 if (rc != -ENOENT && rc != -ENODATA && rc != -EINVAL) {
3283 dt_read_unlock(env, child);
3288 if (!lustre_handle_is_used(&lh)) {
3289 dt_read_unlock(env, child);
3294 } else if (lfsck->li_lpf_obj != NULL &&
3295 lu_fid_eq(pfid, lfsck_dto2fid(lfsck->li_lpf_obj))) {
3297 } else if (unlikely(!fid_is_sane(pfid))) {
3301 rc = lfsck_links_read(env, child, &ldata);
3302 dt_read_unlock(env, child);
3304 if (rc != -ENODATA && rc != -EINVAL)
3307 if (!lustre_handle_is_used(&lh))
3310 if (rc == -EINVAL && !fid_is_zero(pfid)) {
3311 /* Remove the corrupted linkEA. */
3312 rc = lfsck_namespace_links_remove(env, com, child);
3314 /* Here, because of the crashed linkEA, we
3315 * cannot know whether there is some parent
3316 * that references the child directory via
3317 * some name entry or not. So keep it there,
3318 * when the LFSCK run next time, if there is
3319 * some parent that references this object,
3320 * then the LFSCK can rebuild the linkEA;
3321 * otherwise, this object will be handled
3322 * as orphan as above. */
3325 /* 1. If we have neither ".." nor linkEA,
3326 * then it is an orphan.
3328 * 2. If we only have the ".." name entry,
3329 * but no parent references this child
3330 * directory, then handle it as orphan. */
3331 lfsck_ibits_unlock(&lh, LCK_EX);
3332 type = LNIT_MUL_REF;
3334 /* If the LFSCK is marked as LF_INCOMPLETE,
3335 * then means some MDT has ever tried to
3336 * verify some remote MDT-object that resides
3337 * on this MDT, but this MDT failed to respond
3338 * such request. So means there may be some
3339 * remote name entry on other MDT that
3340 * references this object with another name,
3341 * so we cannot know whether this linkEA is
3342 * valid or not. So keep it there and maybe
3343 * resolved when next LFSCK run. */
3344 if (ns->ln_flags & LF_INCOMPLETE)
3347 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
3348 "-"DFID, PFID(pfid));
3349 rc = lfsck_namespace_insert_orphan(env, com, child,
3350 info->lti_tmpbuf, "D", NULL);
3356 linkea_first_entry(&ldata);
3357 /* This is the most common case: the object has unique linkEA entry. */
3358 if (ldata.ld_leh->leh_reccount == 1) {
3359 rc = lfsck_namespace_dsd_single(env, com, child, pfid, &ldata,
3360 &lh, &type, &retry, &unknown);
3362 LASSERT(!lustre_handle_is_used(&lh));
3371 if (!lustre_handle_is_used(&lh))
3374 if (unlikely(ldata.ld_leh->leh_reccount == 0)) {
3375 rc = lfsck_namespace_dsd_orphan(env, com, child, pfid, &lh,
3381 /* When we come here, the cases usually like that:
3382 * 1) The directory object has a corrupted linkEA entry. During the
3383 * first-stage scanning, the LFSCK cannot know such corruption,
3384 * then it appends the right linkEA entry according to the found
3385 * name entry after the bad one.
3387 * 2) The directory object has a right linkEA entry. During the
3388 * first-stage scanning, the LFSCK finds some bad name entry,
3389 * but the LFSCK cannot aware that at that time, then it adds
3390 * the bad linkEA entry for further processing. */
3391 rc = lfsck_namespace_dsd_multiple(env, com, child, pfid, &ldata,
3392 &lh, &type, lpf, &unknown);
3397 lfsck_ibits_unlock(&lh, LCK_EX);
3400 case LNIT_BAD_LINKEA:
3401 ns->ln_linkea_repaired++;
3403 case LNIT_UNMATCHED_PAIRS:
3404 ns->ln_unmatched_pairs_repaired++;
3407 ns->ln_mul_ref_repaired++;
3415 ns->ln_unknown_inconsistency++;
3421 lfsck_namespace_linkea_stale_overflow(struct linkea_data *ldata,
3422 struct lfsck_namespace *ns)
3424 /* Both the leh_overflow_time and ln_time_latest_reset are
3425 * local time based, so need NOT to care about clock drift
3426 * among the servers. */
3427 return ldata->ld_leh->leh_overflow_time &&
3428 ldata->ld_leh->leh_overflow_time < ns->ln_time_latest_reset;
3432 * Clear the object's linkEA overflow timestamp.
3434 * If the MDT-object has too many hard links as to the linkEA cannot hold
3435 * all of them, then overflow timestamp will be set in the linkEA header.
3436 * If some hard links are removed after that, then it is possible to hold
3437 * other missed linkEA entries. If the namespace LFSCK have added all the
3438 * related linkEA entries, then it will remove the overflow timestamp.
3440 * \param[in] env pointer to the thread context
3441 * \param[in] com pointer to the lfsck component
3442 * \param[in] ldata pointer to the linkEA data for the given @obj
3443 * \param[in] obj pointer to the dt_object to be handled
3445 * \retval positive number for repaired cases
3446 * \retval 0 if nothing to be repaired
3447 * \retval negative error number on failure
3449 static int lfsck_namespace_linkea_clear_overflow(const struct lu_env *env,
3450 struct lfsck_component *com,
3451 struct linkea_data *ldata,
3452 struct dt_object *obj)
3454 struct lfsck_namespace *ns = com->lc_file_ram;
3455 struct lfsck_instance *lfsck = com->lc_lfsck;
3456 struct dt_device *dev = lfsck_obj2dev(obj);
3457 struct thandle *th = NULL;
3458 struct lustre_handle lh = { 0 };
3459 struct lu_buf linkea_buf;
3463 LASSERT(!dt_object_remote(obj));
3465 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
3466 MDS_INODELOCK_UPDATE, LCK_PW);
3470 th = dt_trans_create(env, dev);
3472 GOTO(log, rc = PTR_ERR(th));
3474 rc = dt_declare_xattr_set(env, obj,
3475 lfsck_buf_get_const(env, NULL, MAX_LINKEA_SIZE),
3476 XATTR_NAME_LINK, 0, th);
3480 rc = dt_trans_start_local(env, dev, th);
3484 dt_write_lock(env, obj, 0);
3485 rc = lfsck_links_read(env, obj, ldata);
3489 if (unlikely(!lfsck_namespace_linkea_stale_overflow(ldata, ns)))
3490 GOTO(unlock, rc = 0);
3492 ldata->ld_leh->leh_overflow_time = 0;
3493 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
3494 GOTO(unlock, rc = 1);
3496 /* If all known entries are in the linkEA, then the 'leh_reccount'
3497 * should NOT be zero. */
3498 LASSERT(ldata->ld_leh->leh_reccount > 0);
3500 lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
3501 ldata->ld_leh->leh_len);
3502 rc = dt_xattr_set(env, obj, &linkea_buf, XATTR_NAME_LINK, 0, th);
3503 if (unlikely(rc == -ENOSPC))
3511 dt_write_unlock(env, obj);
3514 dt_trans_stop(env, dev, th);
3517 lfsck_ibits_unlock(&lh, LCK_PW);
3518 CDEBUG(D_LFSCK, "%s: clear linkea overflow timestamp for the object "
3520 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
3526 * Verify the object's agent entry.
3528 * If the object claims to have agent entry but the linkEA does not contain
3529 * remote parent, then remove the agent entry. Otherwise, if the object has
3530 * no agent entry but its linkEA contains remote parent, then will generate
3531 * agent entry for it.
3533 * \param[in] env pointer to the thread context
3534 * \param[in] com pointer to the lfsck component
3535 * \param[in] obj pointer to the dt_object to be handled
3537 * \retval positive number for repaired cases
3538 * \retval 0 if nothing to be repaired
3539 * \retval negative error number on failure
3541 static int lfsck_namespace_check_agent_entry(const struct lu_env *env,
3542 struct lfsck_component *com,
3543 struct dt_object *obj)
3545 struct linkea_data ldata = { NULL };
3546 struct lfsck_thread_info *info = lfsck_env_info(env);
3547 struct lfsck_namespace *ns = com->lc_file_ram;
3548 struct lfsck_instance *lfsck = com->lc_lfsck;
3549 struct lu_fid *pfid = &info->lti_fid2;
3550 struct lu_name *cname = &info->lti_name;
3551 struct lu_seq_range *range = &info->lti_range;
3552 struct seq_server_site *ss = lfsck_dev_site(lfsck);
3553 __u32 idx = lfsck_dev_idx(lfsck);
3555 bool remote = false;
3558 if (!(lfsck->li_bookmark_ram.lb_param & LPF_ALL_TGT))
3561 rc = lfsck_links_read_with_rec(env, obj, &ldata);
3562 if (rc == -ENOENT || rc == -ENODATA)
3565 if (rc && rc != -EINVAL)
3568 /* We check the agent entry again after verifying the linkEA
3569 * successfully. So invalid linkEA should be dryrun mode. */
3570 if (rc == -EINVAL || unlikely(!ldata.ld_leh->leh_reccount))
3573 linkea_first_entry(&ldata);
3574 while (ldata.ld_lee != NULL && !remote) {
3575 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
3577 if (!linkea_entry_is_valid(&ldata, cname, pfid))
3580 fld_range_set_mdt(range);
3581 rc = fld_server_lookup(env, ss->ss_server_fld,
3582 fid_seq(pfid), range);
3584 GOTO(out, rc = (rc == -ENOENT ? 0 : rc));
3586 if (range->lsr_index != idx)
3589 linkea_next_entry(&ldata);
3592 if ((lu_object_has_agent_entry(&obj->do_lu) && !remote) ||
3593 (!lu_object_has_agent_entry(&obj->do_lu) && remote)) {
3594 struct dt_device *dev = lfsck_obj2dev(obj);
3595 struct linkea_data ldata2 = { NULL };
3596 struct lustre_handle lh = { 0 };
3597 struct lu_buf linkea_buf;
3598 struct thandle *handle;
3600 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
3603 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
3604 MDS_INODELOCK_UPDATE |
3605 MDS_INODELOCK_XATTR, LCK_EX);
3609 handle = dt_trans_create(env, dev);
3611 GOTO(unlock, rc = PTR_ERR(handle));
3613 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
3614 ldata.ld_leh->leh_len);
3615 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
3616 XATTR_NAME_LINK, LU_XATTR_REPLACE, handle);
3620 rc = dt_trans_start_local(env, dev, handle);
3624 dt_write_lock(env, obj, 0);
3625 rc = lfsck_links_read2_with_rec(env, obj, &ldata2);
3627 if (rc == -ENOENT || rc == -ENODATA)
3632 /* If someone changed linkEA by race, then the agent
3633 * entry will be updated by lower layer automatically. */
3634 if (ldata.ld_leh->leh_len != ldata2.ld_leh->leh_len ||
3635 memcmp(ldata.ld_buf->lb_buf, ldata2.ld_buf->lb_buf,
3636 ldata.ld_leh->leh_len) != 0)
3637 GOTO(unlock2, rc = 0);
3639 rc = dt_xattr_set(env, obj, &linkea_buf, XATTR_NAME_LINK,
3640 LU_XATTR_REPLACE, handle);
3647 dt_write_unlock(env, obj);
3649 dt_trans_stop(env, dev, handle);
3651 lfsck_ibits_unlock(&lh, LCK_EX);
3658 ns->ln_agent_entries_repaired++;
3660 CDEBUG(D_LFSCK, "%s: repair agent entry for "DFID": rc = %d\n",
3661 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
3666 * Double scan the MDT-object for namespace LFSCK.
3668 * If the MDT-object contains invalid or repeated linkEA entries, then drop
3669 * those entries from the linkEA; if the linkEA becomes empty or the object
3670 * has no linkEA, then it is an orphan and will be added into the directory
3671 * .lustre/lost+found/MDTxxxx/; if the remote parent is lost, then recreate
3672 * the remote parent; if the name entry corresponding to some linkEA entry
3673 * is lost, then add the name entry back to the namespace.
3675 * \param[in] env pointer to the thread context
3676 * \param[in] com pointer to the lfsck component
3677 * \param[in] child pointer to the dt_object to be handled
3678 * \param[in] flags some hints to indicate how the @child should be handled
3680 * \retval positive number for repaired cases
3681 * \retval 0 if nothing to be repaired
3682 * \retval negative error number on failure
3684 static int lfsck_namespace_double_scan_one(const struct lu_env *env,
3685 struct lfsck_component *com,
3686 struct dt_object *child, __u8 flags)
3688 struct lfsck_thread_info *info = lfsck_env_info(env);
3689 struct lu_attr *la = &info->lti_la;
3690 struct lu_name *cname = &info->lti_name;
3691 struct lu_fid *pfid = &info->lti_fid;
3692 struct lu_fid *cfid = &info->lti_fid2;
3693 struct lfsck_instance *lfsck = com->lc_lfsck;
3694 struct lfsck_namespace *ns = com->lc_file_ram;
3695 struct dt_object *parent = NULL;
3696 struct linkea_data ldata = { NULL };
3697 bool repaired = false;
3702 dt_read_lock(env, child, 0);
3703 if (unlikely(lfsck_is_dead_obj(child))) {
3704 dt_read_unlock(env, child);
3709 if (S_ISDIR(lfsck_object_type(child))) {
3710 dt_read_unlock(env, child);
3711 rc = lfsck_namespace_double_scan_dir(env, com, child, flags);
3712 if (!rc && flags & LNTF_CHECK_AGENT_ENTRY)
3713 rc = lfsck_namespace_check_agent_entry(env, com, child);
3718 rc = lfsck_links_read(env, child, &ldata);
3719 dt_read_unlock(env, child);
3721 if (rc == -EINVAL) {
3722 struct lustre_handle lh = { 0 };
3724 rc = lfsck_ibits_lock(env, com->lc_lfsck, child, &lh,
3725 MDS_INODELOCK_UPDATE |
3726 MDS_INODELOCK_XATTR, LCK_EX);
3728 rc = lfsck_namespace_links_remove(env, com, child);
3729 lfsck_ibits_unlock(&lh, LCK_EX);
3738 if (!(ns->ln_flags & LF_INCOMPLETE) &&
3739 unlikely(lfsck_namespace_linkea_stale_overflow(&ldata, ns))) {
3740 rc = lfsck_namespace_linkea_clear_overflow(env, com, &ldata,
3746 ns->ln_linkea_overflow_cleared++;
3749 linkea_first_entry(&ldata);
3750 while (ldata.ld_lee != NULL) {
3751 rc = lfsck_namespace_unpack_linkea_entry(&ldata, cname, pfid,
3753 sizeof(info->lti_key));
3754 /* Invalid PFID in the linkEA entry. */
3756 rc = lfsck_namespace_shrink_linkea(env, com, child,
3757 &ldata, cname, pfid, true);
3767 rc = lfsck_namespace_filter_linkea_entry(&ldata, cname, pfid,
3769 /* Found repeated linkEA entries */
3771 rc = lfsck_namespace_shrink_linkea(env, com, child,
3772 &ldata, cname, pfid, false);
3784 parent = lfsck_object_find_bottom(env, lfsck, pfid);
3786 GOTO(out, rc = PTR_ERR(parent));
3788 if (!dt_object_exists(parent)) {
3791 if (ldata.ld_leh->leh_reccount > 1) {
3792 /* If it is NOT the last linkEA entry, then
3793 * there is still other chance to make the
3794 * child to be visible via other parent, then
3795 * remove this linkEA entry. */
3796 rc = lfsck_namespace_shrink_linkea(env, com,
3797 child, &ldata, cname, pfid, true);
3799 /* If the LFSCK is marked as LF_INCOMPLETE,
3800 * then means some MDT has ever tried to
3801 * verify some remote MDT-object that resides
3802 * on this MDT, but this MDT failed to respond
3803 * such request. So means there may be some
3804 * remote name entry on other MDT that
3805 * references this object with another name,
3806 * so we cannot know whether this linkEA is
3807 * valid or not. So keep it there and maybe
3808 * resolved when next LFSCK run. */
3809 if (ns->ln_flags & LF_INCOMPLETE) {
3810 lfsck_object_put(env, parent);
3815 /* Create the lost parent as an orphan. */
3816 rc = lfsck_namespace_create_orphan_dir(env, com,
3819 lfsck_object_put(env, parent);
3827 /* Add the missing name entry to the parent. */
3828 rc = lfsck_namespace_insert_normal(env, com,
3829 parent, child, cname);
3830 if (unlikely(rc == -EEXIST))
3831 /* Unfortunately, someone reused the
3832 * name under the parent by race. So we
3833 * have to remove the linkEA entry from
3834 * current child object. It means that
3835 * the LFSCK cannot recover the system
3836 * totally back to its original status,
3837 * but it is necessary to make the
3838 * current system to be consistent. */
3839 rc = lfsck_namespace_shrink_linkea(env,
3843 linkea_next_entry(&ldata);
3846 lfsck_object_put(env, parent);
3854 } /* !dt_object_exists(parent) */
3856 /* The linkEA entry with bad parent will be removed. */
3857 if (unlikely(!dt_try_as_dir(env, parent))) {
3858 lfsck_object_put(env, parent);
3859 rc = lfsck_namespace_shrink_linkea(env, com, child,
3860 &ldata, cname, pfid, true);
3870 rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
3871 (const struct dt_key *)cname->ln_name);
3872 if (rc != 0 && rc != -ENOENT) {
3873 lfsck_object_put(env, parent);
3879 if (lu_fid_eq(cfid, lfsck_dto2fid(child))) {
3880 /* It is the most common case that we
3881 * find the name entry corresponding
3882 * to the linkEA entry. */
3883 lfsck_object_put(env, parent);
3884 linkea_next_entry(&ldata);
3886 /* The name entry references another
3887 * MDT-object that may be created by
3888 * the LFSCK for repairing dangling
3889 * name entry. Try to replace it. */
3890 rc = lfsck_namespace_replace_cond(env, com,
3891 parent, child, cfid, cname);
3892 lfsck_object_put(env, parent);
3898 linkea_next_entry(&ldata);
3900 rc = lfsck_namespace_shrink_linkea(env,
3914 /* The following handles -ENOENT case */
3916 rc = dt_attr_get(env, child, la);
3920 /* If there is no name entry in the parent dir and the object
3921 * link count is fewer than the linkea entries count, then the
3922 * linkea entry should be removed. */
3923 if (ldata.ld_leh->leh_reccount > la->la_nlink) {
3924 rc = lfsck_namespace_shrink_linkea_cond(env, com,
3925 parent, child, &ldata, cname, pfid);
3926 lfsck_object_put(env, parent);
3936 /* If the LFSCK is marked as LF_INCOMPLETE, then means some
3937 * MDT has ever tried to verify some remote MDT-object that
3938 * resides on this MDT, but this MDT failed to respond such
3939 * request. So means there may be some remote name entry on
3940 * other MDT that references this object with another name,
3941 * so we cannot know whether this linkEA is valid or not.
3942 * So keep it there and maybe resolved when next LFSCK run. */
3943 if (ns->ln_flags & LF_INCOMPLETE) {
3944 lfsck_object_put(env, parent);
3949 rc = lfsck_namespace_check_name(env, lfsck, parent, child,
3955 lfsck_object_put(env, parent);
3960 /* It is an invalid name entry, drop it. */
3961 if (unlikely(rc > 0)) {
3962 lfsck_object_put(env, parent);
3963 rc = lfsck_namespace_shrink_linkea(env, com, child,
3964 &ldata, cname, pfid, true);
3974 /* Add the missing name entry back to the namespace. */
3975 rc = lfsck_namespace_insert_normal(env, com, parent, child,
3977 if (unlikely(rc == -ESTALE))
3978 /* It may happen when the remote object has been
3979 * removed, but the local MDT is not aware of that. */
3982 if (unlikely(rc == -EEXIST))
3983 /* Unfortunately, someone reused the name under the
3984 * parent by race. So we have to remove the linkEA
3985 * entry from current child object. It means that the
3986 * LFSCK cannot recover the system totally back to
3987 * its original status, but it is necessary to make
3988 * the current system to be consistent.
3990 * It also may be because of the LFSCK found some
3991 * internal status of create operation. Under such
3992 * case, nothing to be done. */
3993 rc = lfsck_namespace_shrink_linkea_cond(env, com,
3994 parent, child, &ldata, cname, pfid);
3996 linkea_next_entry(&ldata);
3998 lfsck_object_put(env, parent);
4009 if (rc < 0 && rc != -ENODATA)
4012 if (rc == 0 && ldata.ld_leh != NULL)
4013 count = ldata.ld_leh->leh_reccount;
4016 /* If the LFSCK is marked as LF_INCOMPLETE, then means some
4017 * MDT has ever tried to verify some remote MDT-object that
4018 * resides on this MDT, but this MDT failed to respond such
4019 * request. So means there may be some remote name entry on
4020 * other MDT that references this object with another name,
4021 * so we cannot know whether this linkEA is valid or not.
4022 * So keep it there and maybe resolved when next LFSCK run. */
4023 if (!(ns->ln_flags & LF_INCOMPLETE) &&
4024 (ldata.ld_leh == NULL ||
4025 !ldata.ld_leh->leh_overflow_time)) {
4026 /* If the child becomes orphan, then insert it into
4027 * the global .lustre/lost+found/MDTxxxx directory. */
4028 rc = lfsck_namespace_insert_orphan(env, com, child,
4034 ns->ln_mul_ref_repaired++;
4039 rc = dt_attr_get(env, child, la);
4043 if (la->la_nlink != 0 && la->la_nlink != count) {
4044 if (unlikely(!S_ISREG(lfsck_object_type(child)) &&
4045 !S_ISLNK(lfsck_object_type(child)))) {
4046 CDEBUG(D_LFSCK, "%s: namespace LFSCK finds "
4047 "the object "DFID"'s nlink count %d "
4048 "does not match linkEA count %d, "
4049 "type %o, skip it.\n",
4050 lfsck_lfsck2name(lfsck),
4051 PFID(lfsck_dto2fid(child)),
4052 la->la_nlink, count,
4053 lfsck_object_type(child));
4054 } else if (la->la_nlink < count &&
4055 likely(!ldata.ld_leh->leh_overflow_time)) {
4056 rc = lfsck_namespace_repair_nlink(env, com,
4059 ns->ln_objs_nlink_repaired++;
4067 if (la->la_nlink > 1)
4068 ns->ln_mul_linked_repaired++;
4074 if (!rc && flags & LNTF_CHECK_AGENT_ENTRY)
4075 rc = lfsck_namespace_check_agent_entry(env, com, child);
4080 static void lfsck_namespace_dump_statistics(struct seq_file *m,
4081 struct lfsck_namespace *ns,
4082 __u64 checked_phase1,
4083 __u64 checked_phase2,
4084 time64_t time_phase1,
4085 time64_t time_phase2, bool dryrun)
4087 const char *postfix = dryrun ? "inconsistent" : "repaired";
4089 seq_printf(m, "checked_phase1: %llu\n"
4090 "checked_phase2: %llu\n"
4093 "failed_phase1: %llu\n"
4094 "failed_phase2: %llu\n"
4095 "directories: %llu\n"
4099 "multiple_linked_checked: %llu\n"
4100 "multiple_linked_%s: %llu\n"
4101 "unknown_inconsistency: %llu\n"
4102 "unmatched_pairs_%s: %llu\n"
4103 "dangling_%s: %llu\n"
4104 "multiple_referenced_%s: %llu\n"
4105 "bad_file_type_%s: %llu\n"
4106 "lost_dirent_%s: %llu\n"
4107 "local_lost_found_scanned: %llu\n"
4108 "local_lost_found_moved: %llu\n"
4109 "local_lost_found_skipped: %llu\n"
4110 "local_lost_found_failed: %llu\n"
4111 "striped_dirs_scanned: %llu\n"
4112 "striped_dirs_%s: %llu\n"
4113 "striped_dirs_failed: %llu\n"
4114 "striped_dirs_disabled: %llu\n"
4115 "striped_dirs_skipped: %llu\n"
4116 "striped_shards_scanned: %llu\n"
4117 "striped_shards_%s: %llu\n"
4118 "striped_shards_failed: %llu\n"
4119 "striped_shards_skipped: %llu\n"
4120 "name_hash_%s: %llu\n"
4121 "linkea_overflow_%s: %llu\n"
4122 "agent_entries_%s: %llu\n"
4123 "success_count: %u\n"
4124 "run_time_phase1: %lld seconds\n"
4125 "run_time_phase2: %lld seconds\n",
4128 dryrun ? "inconsistent" : "updated",
4129 ns->ln_items_repaired,
4130 dryrun ? "inconsistent" : "updated",
4131 ns->ln_objs_repaired_phase2,
4132 ns->ln_items_failed,
4133 ns->ln_objs_failed_phase2,
4134 ns->ln_dirs_checked,
4135 postfix, ns->ln_dirent_repaired,
4136 postfix, ns->ln_linkea_repaired,
4137 postfix, ns->ln_objs_nlink_repaired,
4138 ns->ln_mul_linked_checked,
4139 postfix, ns->ln_mul_linked_repaired,
4140 ns->ln_unknown_inconsistency,
4141 postfix, ns->ln_unmatched_pairs_repaired,
4142 postfix, ns->ln_dangling_repaired,
4143 postfix, ns->ln_mul_ref_repaired,
4144 postfix, ns->ln_bad_type_repaired,
4145 postfix, ns->ln_lost_dirent_repaired,
4146 ns->ln_local_lpf_scanned,
4147 ns->ln_local_lpf_moved,
4148 ns->ln_local_lpf_skipped,
4149 ns->ln_local_lpf_failed,
4150 ns->ln_striped_dirs_scanned,
4151 postfix, ns->ln_striped_dirs_repaired,
4152 ns->ln_striped_dirs_failed,
4153 ns->ln_striped_dirs_disabled,
4154 ns->ln_striped_dirs_skipped,
4155 ns->ln_striped_shards_scanned,
4156 postfix, ns->ln_striped_shards_repaired,
4157 ns->ln_striped_shards_failed,
4158 ns->ln_striped_shards_skipped,
4159 postfix, ns->ln_name_hash_repaired,
4160 dryrun ? "inconsistent" : "cleared",
4161 ns->ln_linkea_overflow_cleared,
4162 postfix, ns->ln_agent_entries_repaired,
4163 ns->ln_success_count,
4168 static void lfsck_namespace_release_lmv(const struct lu_env *env,
4169 struct lfsck_component *com)
4171 struct lfsck_instance *lfsck = com->lc_lfsck;
4172 struct lfsck_namespace *ns = com->lc_file_ram;
4174 while (!list_empty(&lfsck->li_list_lmv)) {
4175 struct lfsck_lmv_unit *llu;
4176 struct lfsck_lmv *llmv;
4178 llu = list_entry(lfsck->li_list_lmv.next,
4179 struct lfsck_lmv_unit, llu_link);
4180 llmv = &llu->llu_lmv;
4182 LASSERTF(atomic_read(&llmv->ll_ref) == 1,
4183 "still in using: %u\n",
4184 atomic_read(&llmv->ll_ref));
4186 ns->ln_striped_dirs_skipped++;
4187 lfsck_lmv_put(env, llmv);
4191 static int lfsck_namespace_check_for_double_scan(const struct lu_env *env,
4192 struct lfsck_component *com,
4193 struct dt_object *obj)
4195 struct lu_attr *la = &lfsck_env_info(env)->lti_la;
4198 rc = dt_attr_get(env, obj, la);
4202 /* zero-linkEA object may be orphan, but it also maybe because
4203 * of upgrading. Currently, we cannot record it for double scan.
4204 * Because it may cause the LFSCK trace file to be too large. */
4206 /* "la_ctime" == 1 means that it has ever been removed from
4207 * backend /lost+found directory but not been added back to
4208 * the normal namespace yet. */
4210 if ((S_ISREG(lfsck_object_type(obj)) && la->la_nlink > 1) ||
4211 unlikely(la->la_ctime == 1))
4212 rc = lfsck_namespace_trace_update(env, com, lfsck_dto2fid(obj),
4213 LNTF_CHECK_LINKEA, true);
4218 /* namespace APIs */
4220 static int lfsck_namespace_reset(const struct lu_env *env,
4221 struct lfsck_component *com, bool init)
4223 struct lfsck_instance *lfsck = com->lc_lfsck;
4224 struct lfsck_namespace *ns = com->lc_file_ram;
4225 struct lfsck_assistant_data *lad = com->lc_data;
4226 struct dt_object *root;
4230 root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
4232 GOTO(log, rc = PTR_ERR(root));
4234 if (unlikely(!dt_try_as_dir(env, root)))
4235 GOTO(put, rc = -ENOTDIR);
4237 down_write(&com->lc_sem);
4239 memset(ns, 0, sizeof(*ns));
4241 __u32 count = ns->ln_success_count;
4242 time64_t last_time = ns->ln_time_last_complete;
4244 memset(ns, 0, sizeof(*ns));
4245 ns->ln_success_count = count;
4246 ns->ln_time_last_complete = last_time;
4248 ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
4249 ns->ln_status = LS_INIT;
4250 ns->ln_time_latest_reset = ktime_get_real_seconds();
4252 rc = lfsck_load_one_trace_file(env, com, root, &com->lc_obj,
4253 &dt_lfsck_namespace_features,
4254 LFSCK_NAMESPACE, true);
4258 rc = lfsck_load_sub_trace_files(env, com, &dt_lfsck_namespace_features,
4259 LFSCK_NAMESPACE, true);
4263 clear_bit(LAD_INCOMPLETE, &lad->lad_flags);
4264 CFS_RESET_BITMAP(lad->lad_bitmap);
4266 rc = lfsck_namespace_store(env, com);
4271 up_write(&com->lc_sem);
4274 lfsck_object_put(env, root);
4276 CDEBUG(D_LFSCK, "%s: namespace LFSCK reset: rc = %d\n",
4277 lfsck_lfsck2name(lfsck), rc);
4282 lfsck_namespace_fail(const struct lu_env *env, struct lfsck_component *com,
4285 struct lfsck_namespace *ns = com->lc_file_ram;
4287 down_write(&com->lc_sem);
4289 com->lc_new_checked++;
4290 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
4291 up_write(&com->lc_sem);
4294 static void lfsck_namespace_close_dir(const struct lu_env *env,
4295 struct lfsck_component *com)
4297 struct lfsck_namespace *ns = com->lc_file_ram;
4298 struct lfsck_assistant_data *lad = com->lc_data;
4299 struct lfsck_assistant_object *lso = NULL;
4300 struct lfsck_instance *lfsck = com->lc_lfsck;
4301 struct lfsck_lmv *llmv = lfsck->li_lmv;
4302 struct lfsck_namespace_req *lnr;
4303 struct lu_attr *la = &lfsck_env_info(env)->lti_la2;
4304 __u32 size = sizeof(*lnr) + LFSCK_TMPBUF_LEN;
4306 bool wakeup = false;
4312 rc = dt_attr_get(env, lfsck->li_obj_dir, la);
4316 OBD_ALLOC(lnr, size);
4318 ns->ln_striped_dirs_skipped++;
4323 lso = lfsck_assistant_object_init(env, lfsck_dto2fid(lfsck->li_obj_dir),
4324 la, lfsck->li_pos_current.lp_oit_cookie, true);
4326 OBD_FREE(lnr, size);
4327 ns->ln_striped_dirs_skipped++;
4332 /* Generate a dummy request to indicate that all shards' name entry
4333 * in this striped directory has been scanned for the first time. */
4334 INIT_LIST_HEAD(&lnr->lnr_lar.lar_list);
4335 lnr->lnr_lar.lar_parent = lso;
4336 lnr->lnr_lmv = lfsck_lmv_get(llmv);
4337 lnr->lnr_fid = *lfsck_dto2fid(lfsck->li_obj_dir);
4338 lnr->lnr_dir_cookie = MDS_DIR_END_OFF;
4339 lnr->lnr_size = size;
4340 lnr->lnr_type = lso->lso_attr.la_mode;
4342 spin_lock(&lad->lad_lock);
4343 if (lad->lad_assistant_status < 0 ||
4344 unlikely(!thread_is_running(&lfsck->li_thread) ||
4345 !thread_is_running(&lad->lad_thread))) {
4346 spin_unlock(&lad->lad_lock);
4347 lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar);
4348 ns->ln_striped_dirs_skipped++;
4353 list_add_tail(&lnr->lnr_lar.lar_list, &lad->lad_req_list);
4354 if (lad->lad_prefetched == 0)
4357 lad->lad_prefetched++;
4358 spin_unlock(&lad->lad_lock);
4360 wake_up_all(&lad->lad_thread.t_ctl_waitq);
4365 static int lfsck_namespace_open_dir(const struct lu_env *env,
4366 struct lfsck_component *com)
4368 struct lfsck_instance *lfsck = com->lc_lfsck;
4369 struct lfsck_namespace *ns = com->lc_file_ram;
4370 struct lfsck_lmv *llmv = lfsck->li_lmv;
4377 if (llmv->ll_lmv_master) {
4378 struct lmv_mds_md_v1 *lmv = &llmv->ll_lmv;
4380 if (lmv->lmv_master_mdt_index != lfsck_dev_idx(lfsck)) {
4381 lmv->lmv_master_mdt_index =
4382 lfsck_dev_idx(lfsck);
4383 ns->ln_flags |= LF_INCONSISTENT;
4384 llmv->ll_lmv_updated = 1;
4387 rc = lfsck_namespace_verify_stripe_slave(env, com,
4388 lfsck->li_obj_dir, llmv);
4391 RETURN(rc > 0 ? 0 : rc);
4394 static int lfsck_namespace_checkpoint(const struct lu_env *env,
4395 struct lfsck_component *com, bool init)
4397 struct lfsck_instance *lfsck = com->lc_lfsck;
4398 struct lfsck_namespace *ns = com->lc_file_ram;
4402 rc = lfsck_checkpoint_generic(env, com);
4407 down_write(&com->lc_sem);
4409 ns->ln_pos_latest_start = lfsck->li_pos_checkpoint;
4411 ns->ln_pos_last_checkpoint = lfsck->li_pos_checkpoint;
4412 ns->ln_run_time_phase1 += ktime_get_seconds() -
4413 lfsck->li_time_last_checkpoint;
4414 ns->ln_time_last_checkpoint = ktime_get_real_seconds();
4415 ns->ln_items_checked += com->lc_new_checked;
4416 com->lc_new_checked = 0;
4419 rc = lfsck_namespace_store(env, com);
4420 up_write(&com->lc_sem);
4423 CDEBUG(D_LFSCK, "%s: namespace LFSCK checkpoint at the pos [%llu"
4424 ", "DFID", %#llx], status = %d: rc = %d\n",
4425 lfsck_lfsck2name(lfsck), lfsck->li_pos_current.lp_oit_cookie,
4426 PFID(&lfsck->li_pos_current.lp_dir_parent),
4427 lfsck->li_pos_current.lp_dir_cookie, ns->ln_status, rc);
4429 return rc > 0 ? 0 : rc;
4432 static int lfsck_namespace_prep(const struct lu_env *env,
4433 struct lfsck_component *com,
4434 struct lfsck_start_param *lsp)
4436 struct lfsck_instance *lfsck = com->lc_lfsck;
4437 struct lfsck_namespace *ns = com->lc_file_ram;
4438 struct lfsck_position *pos = &com->lc_pos_start;
4441 rc = lfsck_namespace_load_bitmap(env, com);
4442 if (rc != 0 || ns->ln_status == LS_COMPLETED) {
4443 rc = lfsck_namespace_reset(env, com, false);
4445 rc = lfsck_set_param(env, lfsck, lsp->lsp_start, true);
4448 CDEBUG(D_LFSCK, "%s: namespace LFSCK prep failed: "
4449 "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
4455 down_write(&com->lc_sem);
4456 ns->ln_time_latest_start = ktime_get_real_seconds();
4457 spin_lock(&lfsck->li_lock);
4459 if (ns->ln_flags & LF_SCANNED_ONCE) {
4460 if (!lfsck->li_drop_dryrun ||
4461 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent)) {
4462 ns->ln_status = LS_SCANNING_PHASE2;
4463 list_move_tail(&com->lc_link,
4464 &lfsck->li_list_double_scan);
4465 if (!list_empty(&com->lc_link_dir))
4466 list_del_init(&com->lc_link_dir);
4467 lfsck_pos_set_zero(pos);
4469 ns->ln_status = LS_SCANNING_PHASE1;
4470 ns->ln_run_time_phase1 = 0;
4471 ns->ln_run_time_phase2 = 0;
4472 ns->ln_items_checked = 0;
4473 ns->ln_items_repaired = 0;
4474 ns->ln_items_failed = 0;
4475 ns->ln_dirs_checked = 0;
4476 ns->ln_objs_checked_phase2 = 0;
4477 ns->ln_objs_repaired_phase2 = 0;
4478 ns->ln_objs_failed_phase2 = 0;
4479 ns->ln_objs_nlink_repaired = 0;
4480 ns->ln_dirent_repaired = 0;
4481 ns->ln_linkea_repaired = 0;
4482 ns->ln_mul_linked_checked = 0;
4483 ns->ln_mul_linked_repaired = 0;
4484 ns->ln_unknown_inconsistency = 0;
4485 ns->ln_unmatched_pairs_repaired = 0;
4486 ns->ln_dangling_repaired = 0;
4487 ns->ln_mul_ref_repaired = 0;
4488 ns->ln_bad_type_repaired = 0;
4489 ns->ln_lost_dirent_repaired = 0;
4490 ns->ln_striped_dirs_scanned = 0;
4491 ns->ln_striped_dirs_repaired = 0;
4492 ns->ln_striped_dirs_failed = 0;
4493 ns->ln_striped_dirs_disabled = 0;
4494 ns->ln_striped_dirs_skipped = 0;
4495 ns->ln_striped_shards_scanned = 0;
4496 ns->ln_striped_shards_repaired = 0;
4497 ns->ln_striped_shards_failed = 0;
4498 ns->ln_striped_shards_skipped = 0;
4499 ns->ln_name_hash_repaired = 0;
4500 fid_zero(&ns->ln_fid_latest_scanned_phase2);
4501 if (list_empty(&com->lc_link_dir))
4502 list_add_tail(&com->lc_link_dir,
4503 &lfsck->li_list_dir);
4504 *pos = ns->ln_pos_first_inconsistent;
4507 ns->ln_status = LS_SCANNING_PHASE1;
4508 if (list_empty(&com->lc_link_dir))
4509 list_add_tail(&com->lc_link_dir,
4510 &lfsck->li_list_dir);
4511 if (!lfsck->li_drop_dryrun ||
4512 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent)) {
4513 *pos = ns->ln_pos_last_checkpoint;
4514 pos->lp_oit_cookie++;
4516 *pos = ns->ln_pos_first_inconsistent;
4520 spin_unlock(&lfsck->li_lock);
4521 up_write(&com->lc_sem);
4523 rc = lfsck_start_assistant(env, com, lsp);
4525 CDEBUG(D_LFSCK, "%s: namespace LFSCK prep done, start pos [%llu, "
4526 DFID", %#llx]: rc = %d\n",
4527 lfsck_lfsck2name(lfsck), pos->lp_oit_cookie,
4528 PFID(&pos->lp_dir_parent), pos->lp_dir_cookie, rc);
4533 static int lfsck_namespace_exec_oit(const struct lu_env *env,
4534 struct lfsck_component *com,
4535 struct dt_object *obj)
4537 struct lfsck_thread_info *info = lfsck_env_info(env);
4538 struct lfsck_namespace *ns = com->lc_file_ram;
4539 struct lfsck_instance *lfsck = com->lc_lfsck;
4540 const struct lu_fid *fid = lfsck_dto2fid(obj);
4541 struct lu_fid *pfid = &info->lti_fid2;
4542 struct lu_name *cname = &info->lti_name;
4543 struct lu_seq_range *range = &info->lti_range;
4544 struct seq_server_site *ss = lfsck_dev_site(lfsck);
4545 struct linkea_data ldata = { NULL };
4546 __u32 idx = lfsck_dev_idx(lfsck);
4547 struct lu_attr la = { .la_valid = 0 };
4548 bool remote = false;
4552 rc = dt_attr_get(env, obj, &la);
4553 if (unlikely(rc || (la.la_valid & LA_FLAGS &&
4554 la.la_flags & LUSTRE_ORPHAN_FL))) {
4556 "%s: skip orphan "DFID", %llx/%x: rc = %d\n",
4557 lfsck_lfsck2name(lfsck), PFID(fid),
4558 la.la_valid, la.la_flags, rc);
4563 rc = lfsck_links_read(env, obj, &ldata);
4567 /* -EINVAL means crashed linkEA, should be verified. */
4568 if (rc == -EINVAL) {
4569 rc = lfsck_namespace_trace_update(env, com, fid,
4570 LNTF_CHECK_LINKEA, true);
4572 struct lustre_handle lh = { 0 };
4574 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
4575 MDS_INODELOCK_UPDATE |
4576 MDS_INODELOCK_XATTR, LCK_EX);
4578 rc = lfsck_namespace_links_remove(env, com,
4580 lfsck_ibits_unlock(&lh, LCK_EX);
4584 GOTO(out, rc = (rc == -ENOENT ? 0 : rc));
4587 if (rc && rc != -ENODATA)
4590 if (rc == -ENODATA || unlikely(!ldata.ld_leh->leh_reccount)) {
4591 rc = lfsck_namespace_check_for_double_scan(env, com, obj);
4596 linkea_first_entry(&ldata);
4597 while (ldata.ld_lee != NULL) {
4598 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
4600 if (!fid_is_sane(pfid)) {
4601 rc = lfsck_namespace_trace_update(env, com, fid,
4602 LNTF_CHECK_PARENT, true);
4603 } else if (!linkea_entry_is_valid(&ldata, cname, pfid)) {
4606 fld_range_set_mdt(range);
4607 rc = fld_server_lookup(env, ss->ss_server_fld,
4608 fid_seq(pfid), range);
4609 if ((rc == -ENOENT) ||
4610 (!rc && range->lsr_index != idx)) {
4618 linkea_next_entry(&ldata);
4621 if ((lu_object_has_agent_entry(&obj->do_lu) && !remote) ||
4622 (!lu_object_has_agent_entry(&obj->do_lu) && remote)) {
4623 rc = lfsck_namespace_trace_update(env, com, fid,
4624 LNTF_CHECK_AGENT_ENTRY, true);
4629 /* Record multiple-linked object. */
4630 if (ldata.ld_leh->leh_reccount > 1) {
4631 rc = lfsck_namespace_trace_update(env, com, fid,
4632 LNTF_CHECK_LINKEA, true);
4638 rc = lfsck_namespace_trace_update(env, com, fid,
4639 LNTF_CHECK_LINKEA, true);
4641 rc = lfsck_namespace_check_for_double_scan(env, com, obj);
4646 down_write(&com->lc_sem);
4647 if (S_ISDIR(lfsck_object_type(obj)))
4648 ns->ln_dirs_checked++;
4650 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
4651 up_write(&com->lc_sem);
4656 static int lfsck_namespace_exec_dir(const struct lu_env *env,
4657 struct lfsck_component *com,
4658 struct lfsck_assistant_object *lso,
4659 struct lu_dirent *ent, __u16 type)
4661 struct lfsck_assistant_data *lad = com->lc_data;
4662 struct lfsck_instance *lfsck = com->lc_lfsck;
4663 struct lfsck_namespace_req *lnr;
4664 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4665 struct ptlrpc_thread *mthread = &lfsck->li_thread;
4666 struct ptlrpc_thread *athread = &lad->lad_thread;
4667 bool wakeup = false;
4669 wait_event_idle(mthread->t_ctl_waitq,
4670 lad->lad_prefetched < bk->lb_async_windows ||
4671 !thread_is_running(mthread) ||
4672 !thread_is_running(athread));
4674 if (unlikely(!thread_is_running(mthread) ||
4675 !thread_is_running(athread)))
4678 if (unlikely(lfsck_is_dead_obj(lfsck->li_obj_dir)))
4681 lnr = lfsck_namespace_assistant_req_init(com->lc_lfsck, lso, ent, type);
4683 struct lfsck_namespace *ns = com->lc_file_ram;
4685 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
4686 return PTR_ERR(lnr);
4689 spin_lock(&lad->lad_lock);
4690 if (lad->lad_assistant_status < 0 ||
4691 unlikely(!thread_is_running(mthread) ||
4692 !thread_is_running(athread))) {
4693 spin_unlock(&lad->lad_lock);
4694 lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar);
4695 return lad->lad_assistant_status;
4698 list_add_tail(&lnr->lnr_lar.lar_list, &lad->lad_req_list);
4699 if (lad->lad_prefetched == 0)
4702 lad->lad_prefetched++;
4703 spin_unlock(&lad->lad_lock);
4705 wake_up_all(&lad->lad_thread.t_ctl_waitq);
4707 down_write(&com->lc_sem);
4708 com->lc_new_checked++;
4709 up_write(&com->lc_sem);
4714 static int lfsck_namespace_post(const struct lu_env *env,
4715 struct lfsck_component *com,
4716 int result, bool init)
4718 struct lfsck_instance *lfsck = com->lc_lfsck;
4719 struct lfsck_namespace *ns = com->lc_file_ram;
4723 lfsck_post_generic(env, com, &result);
4725 down_write(&com->lc_sem);
4726 lfsck_namespace_release_lmv(env, com);
4728 spin_lock(&lfsck->li_lock);
4730 ns->ln_pos_last_checkpoint = lfsck->li_pos_checkpoint;
4732 ns->ln_status = LS_SCANNING_PHASE2;
4733 ns->ln_flags |= LF_SCANNED_ONCE;
4734 ns->ln_flags &= ~LF_UPGRADE;
4735 list_del_init(&com->lc_link_dir);
4736 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
4737 } else if (result == 0) {
4738 if (lfsck->li_status != 0)
4739 ns->ln_status = lfsck->li_status;
4741 ns->ln_status = LS_STOPPED;
4742 if (ns->ln_status != LS_PAUSED) {
4743 list_del_init(&com->lc_link_dir);
4744 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4747 ns->ln_status = LS_FAILED;
4748 list_del_init(&com->lc_link_dir);
4749 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4751 spin_unlock(&lfsck->li_lock);
4754 ns->ln_run_time_phase1 += ktime_get_seconds() -
4755 lfsck->li_time_last_checkpoint;
4756 ns->ln_time_last_checkpoint = ktime_get_real_seconds();
4757 ns->ln_items_checked += com->lc_new_checked;
4758 com->lc_new_checked = 0;
4761 rc = lfsck_namespace_store(env, com);
4762 up_write(&com->lc_sem);
4764 CDEBUG(D_LFSCK, "%s: namespace LFSCK post done: rc = %d\n",
4765 lfsck_lfsck2name(lfsck), rc);
4771 lfsck_namespace_dump(const struct lu_env *env, struct lfsck_component *com,
4774 struct lfsck_instance *lfsck = com->lc_lfsck;
4775 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4776 struct lfsck_namespace *ns = com->lc_file_ram;
4778 down_read(&com->lc_sem);
4779 seq_printf(m, "name: lfsck_namespace\n"
4785 lfsck_status2name(ns->ln_status));
4787 lfsck_bits_dump(m, ns->ln_flags, lfsck_flags_names, "flags");
4789 lfsck_bits_dump(m, bk->lb_param, lfsck_param_names, "param");
4791 lfsck_time_dump(m, ns->ln_time_last_complete, "last_completed");
4793 lfsck_time_dump(m, ns->ln_time_latest_start, "latest_start");
4795 lfsck_time_dump(m, ns->ln_time_last_checkpoint, "last_checkpoint");
4797 lfsck_pos_dump(m, &ns->ln_pos_latest_start, "latest_start_position");
4799 lfsck_pos_dump(m, &ns->ln_pos_last_checkpoint,
4800 "last_checkpoint_position");
4802 lfsck_pos_dump(m, &ns->ln_pos_first_inconsistent,
4803 "first_failure_position");
4805 if (ns->ln_status == LS_SCANNING_PHASE1) {
4806 struct lfsck_position pos;
4807 time64_t duration = ktime_get_seconds() -
4808 lfsck->li_time_last_checkpoint;
4809 u64 checked = ns->ln_items_checked + com->lc_new_checked;
4810 u64 speed = checked;
4811 u64 new_checked = com->lc_new_checked;
4812 time64_t rtime = ns->ln_run_time_phase1 + duration;
4815 new_checked = div64_s64(new_checked, duration);
4818 speed = div64_s64(speed, rtime);
4820 lfsck_namespace_dump_statistics(m, ns, checked, 0, rtime, 0,
4821 bk->lb_param & LPF_DRYRUN);
4822 seq_printf(m, "average_speed_phase1: %llu items/sec\n"
4823 "average_speed_phase2: N/A\n"
4824 "average_speed_total: %llu items/sec\n"
4825 "real_time_speed_phase1: %llu items/sec\n"
4826 "real_time_speed_phase2: N/A\n",
4831 if (likely(lfsck->li_di_oit)) {
4832 const struct dt_it_ops *iops =
4833 &lfsck->li_obj_oit->do_index_ops->dio_it;
4835 /* The low layer otable-based iteration position may NOT
4836 * exactly match the namespace-based directory traversal
4837 * cookie. Generally, it is not a serious issue. But the
4838 * caller should NOT make assumption on that. */
4839 pos.lp_oit_cookie = iops->store(env, lfsck->li_di_oit);
4840 if (!lfsck->li_current_oit_processed)
4841 pos.lp_oit_cookie--;
4843 spin_lock(&lfsck->li_lock);
4844 if (lfsck->li_di_dir) {
4845 pos.lp_dir_cookie = lfsck->li_cookie_dir;
4846 if (pos.lp_dir_cookie >= MDS_DIR_END_OFF) {
4847 fid_zero(&pos.lp_dir_parent);
4848 pos.lp_dir_cookie = 0;
4851 *lfsck_dto2fid(lfsck->li_obj_dir);
4854 fid_zero(&pos.lp_dir_parent);
4855 pos.lp_dir_cookie = 0;
4857 spin_unlock(&lfsck->li_lock);
4859 pos = ns->ln_pos_last_checkpoint;
4862 lfsck_pos_dump(m, &pos, "current_position");
4863 } else if (ns->ln_status == LS_SCANNING_PHASE2) {
4864 time64_t duration = ktime_get_seconds() -
4865 com->lc_time_last_checkpoint;
4866 __u64 checked = ns->ln_objs_checked_phase2 +
4867 com->lc_new_checked;
4868 __u64 speed1 = ns->ln_items_checked;
4869 __u64 speed2 = checked;
4870 __u64 speed0 = speed1 + speed2;
4871 __u64 new_checked = com->lc_new_checked;
4872 time64_t rtime = ns->ln_run_time_phase2 + duration;
4873 time64_t time0 = ns->ln_run_time_phase1 + rtime;
4876 new_checked = div64_s64(new_checked, duration);
4878 if (ns->ln_run_time_phase1 != 0)
4879 speed1 = div64_s64(speed1, ns->ln_run_time_phase1);
4880 else if (ns->ln_items_checked != 0)
4884 speed2 = div64_s64(speed2, rtime);
4885 else if (checked != 0)
4889 speed0 = div64_s64(speed0, time0);
4891 lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked,
4893 ns->ln_run_time_phase1, rtime,
4894 bk->lb_param & LPF_DRYRUN);
4895 seq_printf(m, "average_speed_phase1: %llu items/sec\n"
4896 "average_speed_phase2: %llu objs/sec\n"
4897 "average_speed_total: %llu items/sec\n"
4898 "real_time_speed_phase1: N/A\n"
4899 "real_time_speed_phase2: %llu objs/sec\n"
4900 "current_position: "DFID"\n",
4905 PFID(&ns->ln_fid_latest_scanned_phase2));
4907 __u64 speed1 = ns->ln_items_checked;
4908 __u64 speed2 = ns->ln_objs_checked_phase2;
4909 __u64 speed0 = speed1 + speed2;
4910 time64_t time0 = ns->ln_run_time_phase1 + ns->ln_run_time_phase2;
4912 if (ns->ln_run_time_phase1 != 0)
4913 speed1 = div64_s64(speed1, ns->ln_run_time_phase1);
4914 else if (ns->ln_items_checked != 0)
4917 if (ns->ln_run_time_phase2 != 0)
4918 speed2 = div64_s64(speed2, ns->ln_run_time_phase2);
4919 else if (ns->ln_objs_checked_phase2 != 0)
4923 speed0 = div64_s64(speed0, time0);
4925 lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked,
4926 ns->ln_objs_checked_phase2,
4927 ns->ln_run_time_phase1,
4928 ns->ln_run_time_phase2,
4929 bk->lb_param & LPF_DRYRUN);
4930 seq_printf(m, "average_speed_phase1: %llu items/sec\n"
4931 "average_speed_phase2: %llu objs/sec\n"
4932 "average_speed_total: %llu items/sec\n"
4933 "real_time_speed_phase1: N/A\n"
4934 "real_time_speed_phase2: N/A\n"
4935 "current_position: N/A\n",
4941 up_read(&com->lc_sem);
4944 static int lfsck_namespace_double_scan(const struct lu_env *env,
4945 struct lfsck_component *com)
4947 struct lfsck_namespace *ns = com->lc_file_ram;
4948 struct lfsck_assistant_data *lad = com->lc_data;
4949 struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
4950 struct lfsck_tgt_desc *ltd;
4951 struct lfsck_tgt_desc *next;
4954 rc = lfsck_double_scan_generic(env, com, ns->ln_status);
4955 if (thread_is_stopped(&lad->lad_thread)) {
4956 LASSERT(list_empty(&lad->lad_req_list));
4957 LASSERT(list_empty(&lad->lad_mdt_phase1_list));
4959 spin_lock(<ds->ltd_lock);
4960 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4961 ltd_namespace_phase_list) {
4962 list_del_init(<d->ltd_namespace_phase_list);
4964 spin_unlock(<ds->ltd_lock);
4970 static void lfsck_namespace_data_release(const struct lu_env *env,
4971 struct lfsck_component *com)
4973 struct lfsck_assistant_data *lad = com->lc_data;
4974 struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
4975 struct lfsck_tgt_desc *ltd;
4976 struct lfsck_tgt_desc *next;
4978 LASSERT(lad != NULL);
4979 LASSERT(thread_is_init(&lad->lad_thread) ||
4980 thread_is_stopped(&lad->lad_thread));
4981 LASSERT(list_empty(&lad->lad_req_list));
4983 com->lc_data = NULL;
4984 lfsck_namespace_release_lmv(env, com);
4986 spin_lock(<ds->ltd_lock);
4987 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
4988 ltd_namespace_phase_list) {
4989 list_del_init(<d->ltd_namespace_phase_list);
4991 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4992 ltd_namespace_phase_list) {
4993 list_del_init(<d->ltd_namespace_phase_list);
4995 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_list,
4996 ltd_namespace_list) {
4997 list_del_init(<d->ltd_namespace_list);
4999 spin_unlock(<ds->ltd_lock);
5001 if (likely(lad->lad_bitmap != NULL))
5002 CFS_FREE_BITMAP(lad->lad_bitmap);
5007 static void lfsck_namespace_quit(const struct lu_env *env,
5008 struct lfsck_component *com)
5010 struct lfsck_assistant_data *lad = com->lc_data;
5011 struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
5012 struct lfsck_tgt_desc *ltd;
5013 struct lfsck_tgt_desc *next;
5015 LASSERT(lad != NULL);
5017 lfsck_quit_generic(env, com);
5019 LASSERT(thread_is_init(&lad->lad_thread) ||
5020 thread_is_stopped(&lad->lad_thread));
5021 LASSERT(list_empty(&lad->lad_req_list));
5023 lfsck_namespace_release_lmv(env, com);
5025 spin_lock(<ds->ltd_lock);
5026 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
5027 ltd_namespace_phase_list) {
5028 list_del_init(<d->ltd_namespace_phase_list);
5030 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
5031 ltd_namespace_phase_list) {
5032 list_del_init(<d->ltd_namespace_phase_list);
5034 spin_unlock(<ds->ltd_lock);
5037 static int lfsck_namespace_in_notify(const struct lu_env *env,
5038 struct lfsck_component *com,
5039 struct lfsck_request *lr)
5041 struct lfsck_instance *lfsck = com->lc_lfsck;
5042 struct lfsck_namespace *ns = com->lc_file_ram;
5043 struct lfsck_assistant_data *lad = com->lc_data;
5044 struct lfsck_tgt_descs *ltds = &lfsck->li_mdt_descs;
5045 struct lfsck_tgt_desc *ltd;
5050 switch (lr->lr_event) {
5051 case LE_SET_LMV_MASTER: {
5052 struct dt_object *obj;
5054 obj = lfsck_object_find_bottom(env, lfsck, &lr->lr_fid);
5056 RETURN(PTR_ERR(obj));
5058 if (likely(dt_object_exists(obj)))
5059 rc = lfsck_namespace_notify_lmv_master_local(env, com,
5062 lfsck_object_put(env, obj);
5064 RETURN(rc > 0 ? 0 : rc);
5066 case LE_SET_LMV_SLAVE: {
5067 if (!(lr->lr_flags & LEF_RECHECK_NAME_HASH))
5068 ns->ln_striped_shards_repaired++;
5070 rc = lfsck_namespace_trace_update(env, com, &lr->lr_fid,
5071 LNTF_RECHECK_NAME_HASH, true);
5073 RETURN(rc > 0 ? 0 : rc);
5075 case LE_PHASE1_DONE:
5076 case LE_PHASE2_DONE:
5083 CDEBUG(D_LFSCK, "%s: namespace LFSCK handles notify %u from MDT %x, "
5084 "status %d, flags %x\n", lfsck_lfsck2name(lfsck), lr->lr_event,
5085 lr->lr_index, lr->lr_status, lr->lr_flags2);
5087 spin_lock(<ds->ltd_lock);
5088 ltd = lfsck_ltd2tgt(ltds, lr->lr_index);
5090 spin_unlock(<ds->ltd_lock);
5095 list_del_init(<d->ltd_namespace_phase_list);
5096 switch (lr->lr_event) {
5097 case LE_PHASE1_DONE:
5098 if (lr->lr_status <= 0) {
5099 ltd->ltd_namespace_done = 1;
5100 list_del_init(<d->ltd_namespace_list);
5101 CDEBUG(D_LFSCK, "%s: MDT %x failed/stopped at "
5102 "phase1 for namespace LFSCK: rc = %d.\n",
5103 lfsck_lfsck2name(lfsck),
5104 ltd->ltd_index, lr->lr_status);
5105 ns->ln_flags |= LF_INCOMPLETE;
5110 if (lr->lr_flags2 & LF_INCOMPLETE)
5111 ns->ln_flags |= LF_INCOMPLETE;
5113 if (list_empty(<d->ltd_namespace_list))
5114 list_add_tail(<d->ltd_namespace_list,
5115 &lad->lad_mdt_list);
5116 list_add_tail(<d->ltd_namespace_phase_list,
5117 &lad->lad_mdt_phase2_list);
5119 case LE_PHASE2_DONE:
5120 ltd->ltd_namespace_done = 1;
5121 list_del_init(<d->ltd_namespace_list);
5125 ltd->ltd_namespace_done = 1;
5126 list_del_init(<d->ltd_namespace_list);
5127 if (!(lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT)) {
5129 "%s: the peer MDT %x exit namespace LFSCK\n",
5130 lfsck_lfsck2name(lfsck), ltd->ltd_index);
5131 ns->ln_flags |= LF_INCOMPLETE;
5137 spin_unlock(<ds->ltd_lock);
5139 if (fail && lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) {
5140 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
5142 memset(stop, 0, sizeof(*stop));
5143 stop->ls_status = lr->lr_status;
5144 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
5145 lfsck_stop(env, lfsck->li_bottom, stop);
5146 } else if (lfsck_phase2_next_ready(lad)) {
5147 wake_up_all(&lad->lad_thread.t_ctl_waitq);
5153 static void lfsck_namespace_repaired(struct lfsck_namespace *ns, __u64 *count)
5155 *count += ns->ln_objs_nlink_repaired;
5156 *count += ns->ln_dirent_repaired;
5157 *count += ns->ln_linkea_repaired;
5158 *count += ns->ln_mul_linked_repaired;
5159 *count += ns->ln_unmatched_pairs_repaired;
5160 *count += ns->ln_dangling_repaired;
5161 *count += ns->ln_mul_ref_repaired;
5162 *count += ns->ln_bad_type_repaired;
5163 *count += ns->ln_lost_dirent_repaired;
5164 *count += ns->ln_striped_dirs_disabled;
5165 *count += ns->ln_striped_dirs_repaired;
5166 *count += ns->ln_striped_shards_repaired;
5167 *count += ns->ln_name_hash_repaired;
5168 *count += ns->ln_local_lpf_moved;
5171 static int lfsck_namespace_query_all(const struct lu_env *env,
5172 struct lfsck_component *com,
5173 __u32 *mdts_count, __u64 *repaired)
5175 struct lfsck_namespace *ns = com->lc_file_ram;
5176 struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
5177 struct lfsck_tgt_desc *ltd;
5182 rc = lfsck_query_all(env, com);
5186 down_read(<ds->ltd_rw_sem);
5187 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
5188 ltd = lfsck_ltd2tgt(ltds, idx);
5189 LASSERT(ltd != NULL);
5191 mdts_count[ltd->ltd_namespace_status]++;
5192 *repaired += ltd->ltd_namespace_repaired;
5194 up_read(<ds->ltd_rw_sem);
5196 down_read(&com->lc_sem);
5197 mdts_count[ns->ln_status]++;
5198 lfsck_namespace_repaired(ns, repaired);
5199 up_read(&com->lc_sem);
5204 static int lfsck_namespace_query(const struct lu_env *env,
5205 struct lfsck_component *com,
5206 struct lfsck_request *req,
5207 struct lfsck_reply *rep,
5208 struct lfsck_query *que, int idx)
5210 struct lfsck_namespace *ns = com->lc_file_ram;
5214 LASSERT(com->lc_lfsck->li_master);
5216 rc = lfsck_namespace_query_all(env, com,
5217 que->lu_mdts_count[idx],
5218 &que->lu_repaired[idx]);
5220 down_read(&com->lc_sem);
5221 rep->lr_status = ns->ln_status;
5222 if (req->lr_flags & LEF_QUERY_ALL)
5223 lfsck_namespace_repaired(ns, &rep->lr_repaired);
5224 up_read(&com->lc_sem);
5230 static struct lfsck_operations lfsck_namespace_ops = {
5231 .lfsck_reset = lfsck_namespace_reset,
5232 .lfsck_fail = lfsck_namespace_fail,
5233 .lfsck_close_dir = lfsck_namespace_close_dir,
5234 .lfsck_open_dir = lfsck_namespace_open_dir,
5235 .lfsck_checkpoint = lfsck_namespace_checkpoint,
5236 .lfsck_prep = lfsck_namespace_prep,
5237 .lfsck_exec_oit = lfsck_namespace_exec_oit,
5238 .lfsck_exec_dir = lfsck_namespace_exec_dir,
5239 .lfsck_post = lfsck_namespace_post,
5240 .lfsck_dump = lfsck_namespace_dump,
5241 .lfsck_double_scan = lfsck_namespace_double_scan,
5242 .lfsck_data_release = lfsck_namespace_data_release,
5243 .lfsck_quit = lfsck_namespace_quit,
5244 .lfsck_in_notify = lfsck_namespace_in_notify,
5245 .lfsck_query = lfsck_namespace_query,
5249 * Repair dangling name entry.
5251 * For the name entry with dangling reference, we need to repare the
5252 * inconsistency according to the LFSCK sponsor's requirement:
5254 * 1) Keep the inconsistency there and report the inconsistency case,
5255 * then give the chance to the application to find related issues,
5256 * and the users can make the decision about how to handle it with
5257 * more human knownledge. (by default)
5259 * 2) Re-create the missing MDT-object with the FID information.
5261 * \param[in] env pointer to the thread context
5262 * \param[in] com pointer to the lfsck component
5263 * \param[in] parent pointer to the dir object that contains the dangling
5265 * \param[in] child pointer to the object corresponding to the dangling
5267 * \param[in] lnr pointer to the namespace request that contains the
5268 * name's name, parent object, parent's LMV, and ect.
5270 * \retval positive number if no need to repair
5271 * \retval zero for repaired successfully
5272 * \retval negative error number on failure
5274 int lfsck_namespace_repair_dangling(const struct lu_env *env,
5275 struct lfsck_component *com,
5276 struct dt_object *parent,
5277 struct dt_object *child,
5278 struct lfsck_namespace_req *lnr)
5280 struct lfsck_thread_info *info = lfsck_env_info(env);
5281 struct lu_attr *la = &info->lti_la;
5282 struct dt_allocation_hint *hint = &info->lti_hint;
5283 struct dt_object_format *dof = &info->lti_dof;
5284 struct dt_insert_rec *rec = &info->lti_dt_rec;
5285 struct lmv_mds_md_v1 *lmv2 = &info->lti_lmv2;
5286 const struct lu_name *cname;
5287 const struct lu_fid *pfid = lfsck_dto2fid(parent);
5288 const struct lu_fid *cfid = lfsck_dto2fid(child);
5289 struct linkea_data ldata = { NULL };
5290 struct lfsck_lock_handle *llh = &info->lti_llh;
5291 struct lustre_handle rlh = { 0 };
5292 struct lustre_handle clh = { 0 };
5293 struct lu_buf linkea_buf;
5294 struct lu_buf lmv_buf;
5295 struct lfsck_instance *lfsck = com->lc_lfsck;
5296 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5297 struct dt_device *dev = lfsck->li_next;
5298 struct thandle *th = NULL;
5300 __u16 type = lnr->lnr_type;
5304 cname = lfsck_name_get_const(env, lnr->lnr_name, lnr->lnr_namelen);
5305 if (bk->lb_param & LPF_CREATE_MDTOBJ)
5310 if (!create || bk->lb_param & LPF_DRYRUN)
5313 /* We may need to create the sub-objects of the @child via LOD,
5314 * so make the modification based on lfsck->li_next. */
5316 parent = lfsck_object_locate(dev, parent);
5318 GOTO(log, rc = PTR_ERR(parent));
5320 if (unlikely(!dt_try_as_dir(env, parent)))
5321 GOTO(log, rc = -ENOTDIR);
5323 child = lfsck_object_locate(dev, child);
5325 GOTO(log, rc = PTR_ERR(child));
5327 rc = linkea_links_new(&ldata, &info->lti_linkea_buf2,
5332 rc = lfsck_lock(env, lfsck, parent, lnr->lnr_name, llh,
5333 MDS_INODELOCK_UPDATE, LCK_PW);
5337 rc = lfsck_namespace_check_exist(env, parent, child, lnr->lnr_name);
5341 if (dt_object_remote(child)) {
5342 rc = lfsck_remote_lookup_lock(env, lfsck, parent, child, &rlh,
5348 rc = lfsck_ibits_lock(env, lfsck, child, &clh,
5349 MDS_INODELOCK_UPDATE | MDS_INODELOCK_LOOKUP |
5350 MDS_INODELOCK_XATTR, LCK_EX);
5352 GOTO(unlock_remote_lookup, rc);
5354 /* Set the ctime as zero, then others can know it is created for
5355 * repairing dangling name entry by LFSCK. And if the LFSCK made
5356 * wrong decision and the real MDT-object has been found later,
5357 * then the LFSCK has chance to fix the incosistency properly. */
5358 memset(la, 0, sizeof(*la));
5360 la->la_mode = (type & S_IFMT) | 0700;
5362 la->la_mode = (type & S_IFMT) | 0600;
5363 la->la_valid = LA_TYPE | LA_MODE | LA_CTIME;
5366 * if it's directory, skip do_ah_init() to create a plain directory
5367 * because it may have shards already, which will be inserted back
5368 * later, besides, it may be remote, and creating stripe directory
5369 * remotely is not supported.
5372 child->do_ops->do_ah_init(env, hint, parent, child, type);
5373 else if (S_ISDIR(type))
5374 child->do_ops->do_ah_init(env, hint, NULL, child, type);
5376 memset(dof, 0, sizeof(*dof));
5377 dof->dof_type = dt_mode_to_dft(type);
5378 /* If the target is a regular file, then the LFSCK will only create
5379 * the MDT-object without stripes (dof->dof_reg.striped = 0). related
5380 * OST-objects will be created when write open. */
5382 th = dt_trans_create(env, dev);
5384 GOTO(unlock_child, rc = PTR_ERR(th));
5386 /* 1a. create child. */
5387 rc = dt_declare_create(env, child, la, hint, dof, th);
5391 if (S_ISDIR(type)) {
5392 if (unlikely(!dt_try_as_dir(env, child)))
5393 GOTO(stop, rc = -ENOTDIR);
5395 /* 2a. increase child nlink */
5396 rc = dt_declare_ref_add(env, child, th);
5400 /* 3a. insert dot into child dir */
5401 rec->rec_type = S_IFDIR;
5402 rec->rec_fid = cfid;
5403 rc = dt_declare_insert(env, child,
5404 (const struct dt_rec *)rec,
5405 (const struct dt_key *)dot, th);
5409 /* 4a. insert dotdot into child dir */
5410 rec->rec_fid = pfid;
5411 rc = dt_declare_insert(env, child,
5412 (const struct dt_rec *)rec,
5413 (const struct dt_key *)dotdot, th);
5417 /* 5a. generate slave LMV EA. */
5418 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
5421 idx = lfsck_shard_name_to_index(env,
5422 lnr->lnr_name, lnr->lnr_namelen,
5424 if (unlikely(idx < 0))
5425 GOTO(stop, rc = idx);
5427 *lmv2 = lnr->lnr_lmv->ll_lmv;
5428 lmv2->lmv_magic = LMV_MAGIC_STRIPE;
5429 lmv2->lmv_master_mdt_index = idx;
5431 lfsck_lmv_header_cpu_to_le(lmv2, lmv2);
5432 lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
5433 rc = dt_declare_xattr_set(env, child, &lmv_buf,
5434 XATTR_NAME_LMV, 0, th);
5440 /* 6a. insert linkEA for child */
5441 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
5442 ldata.ld_leh->leh_len);
5443 rc = dt_declare_xattr_set(env, child, &linkea_buf,
5444 XATTR_NAME_LINK, 0, th);
5448 /* 7a. if child is remote, delete and insert to generate local agent */
5449 if (dt_object_remote(child)) {
5450 rc = dt_declare_delete(env, parent,
5451 (const struct dt_key *)lnr->lnr_name,
5456 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
5457 (const struct dt_key *)lnr->lnr_name,
5463 rc = dt_trans_start_local(env, dev, th);
5465 GOTO(stop, rc = (rc == -EEXIST ? 1 : rc));
5467 dt_write_lock(env, child, 0);
5468 /* 1b. create child */
5469 rc = dt_create(env, child, la, hint, dof, th);
5471 GOTO(unlock, rc = (rc == -EEXIST ? 1 : rc));
5473 if (S_ISDIR(type)) {
5474 /* 2b. increase child nlink */
5475 rc = dt_ref_add(env, child, th);
5479 /* 3b. insert dot into child dir */
5480 rec->rec_type = S_IFDIR;
5481 rec->rec_fid = cfid;
5482 rc = dt_insert(env, child, (const struct dt_rec *)rec,
5483 (const struct dt_key *)dot, th);
5487 /* 4b. insert dotdot into child dir */
5488 rec->rec_fid = pfid;
5489 rc = dt_insert(env, child, (const struct dt_rec *)rec,
5490 (const struct dt_key *)dotdot, th);
5494 /* 5b. generate slave LMV EA. */
5495 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
5496 rc = dt_xattr_set(env, child, &lmv_buf, XATTR_NAME_LMV,
5503 /* 6b. insert linkEA for child. */
5504 rc = dt_xattr_set(env, child, &linkea_buf,
5505 XATTR_NAME_LINK, 0, th);
5509 /* 7b. if child is remote, delete and insert to generate local agent */
5510 if (dt_object_remote(child)) {
5511 rc = dt_delete(env, parent,
5512 (const struct dt_key *)lnr->lnr_name, th);
5516 rec->rec_type = type;
5517 rec->rec_fid = cfid;
5518 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
5519 (const struct dt_key *)lnr->lnr_name, th);
5527 dt_write_unlock(env, child);
5530 dt_trans_stop(env, dev, th);
5533 lfsck_ibits_unlock(&clh, LCK_EX);
5534 unlock_remote_lookup:
5535 if (dt_object_remote(child))
5536 lfsck_ibits_unlock(&rlh, LCK_EX);
5539 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found dangling "
5540 "reference for: parent "DFID", child "DFID", type %u, "
5541 "name %s. %s: rc = %d\n", lfsck_lfsck2name(lfsck),
5542 PFID(pfid), PFID(cfid), type, cname->ln_name,
5543 create ? "Create the lost MDT-object as required" :
5544 "Keep the MDT-object there by default", rc);
5547 struct lfsck_namespace *ns = com->lc_file_ram;
5549 ns->ln_flags |= LF_INCONSISTENT;
5555 static int lfsck_namespace_assistant_handler_p1(const struct lu_env *env,
5556 struct lfsck_component *com,
5557 struct lfsck_assistant_req *lar)
5559 struct lfsck_thread_info *info = lfsck_env_info(env);
5560 struct lu_attr *la = &info->lti_la;
5561 struct lfsck_instance *lfsck = com->lc_lfsck;
5562 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5563 struct lfsck_namespace *ns = com->lc_file_ram;
5564 struct lfsck_assistant_data *lad = com->lc_data;
5565 struct linkea_data ldata = { NULL };
5566 const struct lu_name *cname;
5567 struct thandle *handle = NULL;
5568 struct lfsck_namespace_req *lnr =
5569 container_of(lar, struct lfsck_namespace_req, lnr_lar);
5570 struct dt_object *dir = NULL;
5571 struct dt_object *obj = NULL;
5572 struct lfsck_assistant_object *lso = lar->lar_parent;
5573 const struct lu_fid *pfid = &lso->lso_fid;
5574 struct dt_device *dev = NULL;
5575 struct lustre_handle lh = { 0 };
5576 bool repaired = false;
5577 bool dtlocked = false;
5581 bool bad_hash = false;
5582 bool bad_linkea = false;
5586 enum lfsck_namespace_inconsistency_type type = LNIT_NONE;
5593 if (lnr->lnr_attr & LUDA_UPGRADE) {
5594 ns->ln_flags |= LF_UPGRADE;
5595 ns->ln_dirent_repaired++;
5597 } else if (lnr->lnr_attr & LUDA_REPAIR) {
5598 ns->ln_flags |= LF_INCONSISTENT;
5599 ns->ln_dirent_repaired++;
5603 if (unlikely(fid_is_zero(&lnr->lnr_fid))) {
5604 if (strcmp(lnr->lnr_name, dotdot) != 0)
5607 rc = lfsck_namespace_trace_update(env, com, pfid,
5608 LNTF_CHECK_PARENT, true);
5613 if (unlikely(!fid_is_sane(&lnr->lnr_fid))) {
5614 CDEBUG(D_LFSCK, "%s: dir scan find invalid FID "DFID
5615 " for the name entry %.*s under "DFID"\n",
5616 lfsck_lfsck2name(lfsck), PFID(&lnr->lnr_fid),
5617 lnr->lnr_namelen, lnr->lnr_name, PFID(pfid));
5619 if (strcmp(lnr->lnr_name, dotdot) != 0)
5620 /* invalid FID means bad name entry, remove it. */
5621 type = LNIT_BAD_DIRENT;
5623 /* If the parent FID is invalid, we cannot remove
5624 * the ".." entry directly. */
5625 rc = lfsck_namespace_trace_update(env, com, pfid,
5626 LNTF_CHECK_PARENT, true);
5631 if (unlikely(lnr->lnr_dir_cookie == MDS_DIR_END_OFF)) {
5632 rc = lfsck_namespace_striped_dir_rescan(env, com, lnr);
5637 if (fid_seq_is_dot(fid_seq(&lnr->lnr_fid)))
5640 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
5641 rc = lfsck_namespace_handle_striped_master(env, com, lnr);
5646 idx = lfsck_find_mdt_idx_by_fid(env, lfsck, &lnr->lnr_fid);
5648 GOTO(out, rc = idx);
5650 if (idx == lfsck_dev_idx(lfsck)) {
5651 if (unlikely(strcmp(lnr->lnr_name, dotdot) == 0))
5654 dev = lfsck->li_bottom;
5656 struct lfsck_tgt_desc *ltd;
5658 /* Usually, some local filesystem consistency verification
5659 * tools can guarantee the local namespace tree consistenct.
5660 * So the LFSCK will only verify the remote directory. */
5661 if (unlikely(strcmp(lnr->lnr_name, dotdot) == 0)) {
5662 rc = lfsck_namespace_trace_update(env, com, pfid,
5663 LNTF_CHECK_PARENT, true);
5668 ltd = lfsck_ltd2tgt(&lfsck->li_mdt_descs, idx);
5669 if (unlikely(ltd == NULL)) {
5670 CDEBUG(D_LFSCK, "%s: cannot talk with MDT %x which "
5671 "did not join the namespace LFSCK\n",
5672 lfsck_lfsck2name(lfsck), idx);
5673 lfsck_lad_set_bitmap(env, com, idx);
5675 GOTO(out, rc = -ENODEV);
5681 obj = lfsck_object_find_by_dev(env, dev, &lnr->lnr_fid);
5683 GOTO(out, rc = PTR_ERR(obj));
5685 cname = lfsck_name_get_const(env, lnr->lnr_name, lnr->lnr_namelen);
5686 if (dt_object_exists(obj) == 0) {
5690 dir = lfsck_assistant_object_load(env, lfsck, lso);
5694 GOTO(trace, rc == -ENOENT ? 0 : rc);
5698 rc = lfsck_namespace_check_exist(env, dir, obj, lnr->lnr_name);
5700 if (!lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5701 lnr->lnr_name, lnr->lnr_namelen)) {
5702 type = LNIT_BAD_DIRENT;
5707 type = LNIT_DANGLING;
5708 rc = lfsck_namespace_repair_dangling(env, com, dir,
5717 if (!(bk->lb_param & LPF_DRYRUN) && lad->lad_advance_lock) {
5720 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
5721 MDS_INODELOCK_UPDATE |
5722 MDS_INODELOCK_XATTR, LCK_EX);
5726 handle = dt_trans_create(env, dev);
5728 GOTO(out, rc = PTR_ERR(handle));
5730 rc = lfsck_declare_namespace_exec_dir(env, obj, handle);
5734 rc = dt_trans_start_local(env, dev, handle);
5738 dt_write_lock(env, obj, 0);
5742 rc = lfsck_links_read(env, obj, &ldata);
5743 if (unlikely(rc == -ENOENT)) {
5744 if (handle != NULL) {
5745 dt_write_unlock(env, obj);
5748 dt_trans_stop(env, dev, handle);
5751 lfsck_ibits_unlock(&lh, LCK_EX);
5754 /* It may happen when the remote object has been removed,
5755 * but the local MDT is not aware of that. */
5757 } else if (rc == 0) {
5758 count = ldata.ld_leh->leh_reccount;
5759 rc = linkea_links_find(&ldata, cname, pfid);
5761 (count == 1 || !S_ISDIR(lfsck_object_type(obj)))) {
5762 if ((lfsck_object_type(obj) & S_IFMT) !=
5764 ns->ln_flags |= LF_INCONSISTENT;
5765 type = LNIT_BAD_TYPE;
5771 /* If the name entry hash does not match the slave striped
5772 * directory, and the name entry does not match also, then
5773 * it is quite possible that name entry is corrupted. */
5774 if (!lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5775 lnr->lnr_name, lnr->lnr_namelen)) {
5776 ns->ln_flags |= LF_INCONSISTENT;
5777 type = LNIT_BAD_DIRENT;
5782 /* If the file type stored in the name entry does not match
5783 * the file type claimed by the object, and the object does
5784 * not recognize the name entry, then it is quite possible
5785 * that the name entry is corrupted. */
5786 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type) {
5787 ns->ln_flags |= LF_INCONSISTENT;
5788 type = LNIT_BAD_DIRENT;
5793 /* For sub-dir object, we cannot make sure whether the sub-dir
5794 * back references the parent via ".." name entry correctly or
5795 * not in the LFSCK first-stage scanning. It may be that the
5796 * (remote) sub-dir ".." name entry has no parent FID after
5797 * file-level backup/restore and its linkEA may be wrong.
5798 * So under such case, we should replace the linkEA according
5799 * to current name entry. But this needs to be done during the
5800 * LFSCK second-stage scanning. The LFSCK will record the name
5801 * entry for further possible using. */
5805 } else if (unlikely(rc == -EINVAL)) {
5806 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type)
5807 type = LNIT_BAD_TYPE;
5810 /* The magic crashed, we are not sure whether there are more
5811 * corrupt data in the linkea, so remove all linkea entries. */
5815 } else if (rc == -ENODATA) {
5816 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type)
5817 type = LNIT_BAD_TYPE;
5824 if (bk->lb_param & LPF_DRYRUN) {
5826 ns->ln_flags |= LF_UPGRADE;
5828 ns->ln_flags |= LF_INCONSISTENT;
5829 ns->ln_linkea_repaired++;
5835 if (!lustre_handle_is_used(&lh)) {
5843 LASSERT(handle != NULL);
5846 dir = lfsck_assistant_object_load(env, lfsck, lso);
5850 GOTO(stop, rc == -ENOENT ? 0 : rc);
5854 rc = lfsck_namespace_check_exist(env, dir, obj, lnr->lnr_name);
5859 if (!remove && newdata)
5860 ns->ln_flags |= LF_UPGRADE;
5861 else if (remove || !(ns->ln_flags & LF_UPGRADE))
5862 ns->ln_flags |= LF_INCONSISTENT;
5867 rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, handle);
5868 if (rc != 0 && rc != -ENOENT && rc != -ENODATA)
5873 rc = linkea_data_new(&ldata,
5874 &lfsck_env_info(env)->lti_linkea_buf);
5879 rc = linkea_add_buf(&ldata, cname, pfid);
5881 rc = lfsck_links_write(env, obj, &ldata, handle);
5885 count = ldata.ld_leh->leh_reccount;
5886 if (!S_ISDIR(lfsck_object_type(obj)) ||
5887 !dt_object_remote(obj)) {
5888 ns->ln_linkea_repaired++;
5898 dt_write_unlock(env, obj);
5900 if (handle != NULL && !IS_ERR(handle))
5901 dt_trans_stop(env, dev, handle);
5904 lfsck_ibits_unlock(&lh, LCK_EX);
5906 if (!name_is_dot_or_dotdot(lnr->lnr_name, lnr->lnr_namelen) &&
5907 !lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5908 lnr->lnr_name, lnr->lnr_namelen) &&
5909 type != LNIT_BAD_DIRENT) {
5910 ns->ln_flags |= LF_INCONSISTENT;
5914 dir = lfsck_assistant_object_load(env, lfsck, lso);
5918 GOTO(trace, rc == -ENOENT ? 0 : rc);
5922 rc = lfsck_namespace_repair_bad_name_hash(env, com, dir,
5923 lnr->lnr_lmv, lnr->lnr_name);
5929 if (type != LNIT_NONE && dir == NULL) {
5930 dir = lfsck_assistant_object_load(env, lfsck, lso);
5934 GOTO(trace, rc == -ENOENT ? 0 : rc);
5941 rc = lfsck_namespace_repair_dirent(env, com, dir,
5942 obj, lnr->lnr_name, lnr->lnr_name,
5943 lnr->lnr_type, true, false);
5947 case LNIT_BAD_DIRENT:
5949 /* XXX: This is a bad dirent, we do not know whether
5950 * the original name entry reference a regular
5951 * file or a directory, then keep the parent's
5952 * nlink count unchanged here. */
5953 rc = lfsck_namespace_repair_dirent(env, com, dir,
5954 obj, lnr->lnr_name, lnr->lnr_name,
5955 lnr->lnr_type, false, false);
5963 if (obj != NULL && count == 1 &&
5964 S_ISREG(lfsck_object_type(obj)))
5965 dt_attr_get(env, obj, la);
5969 down_write(&com->lc_sem);
5971 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail to handle "
5972 "the entry: "DFID", parent "DFID", name %.*s: rc = %d\n",
5973 lfsck_lfsck2name(lfsck), PFID(&lnr->lnr_fid), PFID(pfid),
5974 lnr->lnr_namelen, lnr->lnr_name, rc);
5976 lfsck_namespace_record_failure(env, lfsck, ns);
5977 if ((rc == -ENOTCONN || rc == -ESHUTDOWN || rc == -EREMCHG ||
5978 rc == -ETIMEDOUT || rc == -EHOSTDOWN ||
5979 rc == -EHOSTUNREACH || rc == -EINPROGRESS) &&
5980 dev != NULL && dev != lfsck->li_bottom)
5981 lfsck_lad_set_bitmap(env, com, idx);
5983 if (!(bk->lb_param & LPF_FAILOUT))
5987 ns->ln_items_repaired++;
5989 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant "
5990 "repaired the entry: "DFID", parent "DFID
5991 ", name %.*s, type %d\n",
5992 lfsck_lfsck2name(lfsck),
5993 PFID(&lnr->lnr_fid), PFID(pfid),
5994 lnr->lnr_namelen, lnr->lnr_name, type);
5998 ns->ln_dangling_repaired++;
6001 ns->ln_bad_type_repaired++;
6003 case LNIT_BAD_DIRENT:
6004 ns->ln_dirent_repaired++;
6010 if (bk->lb_param & LPF_DRYRUN &&
6011 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent))
6012 lfsck_pos_fill(env, lfsck,
6013 &ns->ln_pos_first_inconsistent,
6018 ns->ln_name_hash_repaired++;
6020 /* Not count repeatedly. */
6022 ns->ln_items_repaired++;
6024 CDEBUG(D_LFSCK, "%s: namespace LFSCK "
6025 "assistant repaired the entry: "
6028 lfsck_lfsck2name(lfsck),
6029 PFID(&lnr->lnr_fid), PFID(pfid),
6030 lnr->lnr_namelen, lnr->lnr_name);
6033 if (bk->lb_param & LPF_DRYRUN &&
6034 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent))
6035 lfsck_pos_fill(env, lfsck,
6036 &ns->ln_pos_first_inconsistent,
6043 if (count > 1 || la->la_nlink > 1)
6044 ns->ln_mul_linked_checked++;
6046 up_write(&com->lc_sem);
6048 if (obj != NULL && !IS_ERR(obj))
6049 lfsck_object_put(env, obj);
6051 if (dir != NULL && !IS_ERR(dir))
6052 lfsck_object_put(env, dir);
6054 lad->lad_advance_lock = bad_linkea;
6060 * Handle one orphan under the backend /lost+found directory
6062 * Insert the orphan FID into the namespace LFSCK trace file for further
6063 * processing (via the subsequent namespace LFSCK second-stage scanning).
6064 * At the same time, remove the orphan name entry from backend /lost+found
6065 * directory. There is an interval between the orphan name entry removed
6066 * from the backend /lost+found directory and the orphan FID in the LFSCK
6067 * trace file handled. In such interval, the LFSCK can be reset, then
6068 * all the FIDs recorded in the namespace LFSCK trace file will be dropped.
6069 * To guarantee that the orphans can be found when LFSCK run next time
6070 * without e2fsck again, when remove the orphan name entry, the LFSCK
6071 * will set the orphan's ctime attribute as 1. Since normal applications
6072 * cannot change the object's ctime attribute as 1. Then when LFSCK run
6073 * next time, it can record the object (that ctime is 1) in the namespace
6074 * LFSCK trace file during the first-stage scanning.
6076 * \param[in] env pointer to the thread context
6077 * \param[in] com pointer to the lfsck component
6078 * \param[in] parent pointer to the object for the backend /lost+found
6079 * \param[in] ent pointer to the name entry for the target under the
6080 * backend /lost+found
6082 * \retval positive for repaired
6083 * \retval 0 if needs to repair nothing
6084 * \retval negative error number on failure
6086 static int lfsck_namespace_scan_local_lpf_one(const struct lu_env *env,
6087 struct lfsck_component *com,
6088 struct dt_object *parent,
6089 struct lu_dirent *ent)
6091 struct lfsck_thread_info *info = lfsck_env_info(env);
6092 struct lu_fid *key = &info->lti_fid;
6093 struct lu_attr *la = &info->lti_la;
6094 struct lfsck_instance *lfsck = com->lc_lfsck;
6095 struct dt_object *obj;
6096 struct dt_device *dev = lfsck->li_bottom;
6097 struct dt_object *child = NULL;
6098 struct thandle *th = NULL;
6105 child = lfsck_object_find_by_dev(env, dev, &ent->lde_fid);
6107 RETURN(PTR_ERR(child));
6109 LASSERT(dt_object_exists(child));
6110 LASSERT(!dt_object_remote(child));
6112 idx = lfsck_sub_trace_file_fid2idx(&ent->lde_fid);
6113 obj = com->lc_sub_trace_objs[idx].lsto_obj;
6114 fid_cpu_to_be(key, &ent->lde_fid);
6115 rc = dt_lookup(env, obj, (struct dt_rec *)&flags,
6116 (const struct dt_key *)key);
6119 flags |= LNTF_CHECK_ORPHAN;
6120 } else if (rc == -ENOENT) {
6121 flags = LNTF_CHECK_ORPHAN;
6126 th = dt_trans_create(env, dev);
6128 GOTO(out, rc = PTR_ERR(th));
6130 /* a1. remove name entry from backend /lost+found */
6131 rc = dt_declare_delete(env, parent,
6132 (const struct dt_key *)ent->lde_name, th);
6136 if (S_ISDIR(lfsck_object_type(child))) {
6137 /* a2. decrease parent's nlink */
6138 rc = dt_declare_ref_del(env, parent, th);
6144 /* a3. remove child's FID from the LFSCK trace file. */
6145 rc = dt_declare_delete(env, obj,
6146 (const struct dt_key *)key, th);
6150 /* a4. set child's ctime as 1 */
6151 memset(la, 0, sizeof(*la));
6153 la->la_valid = LA_CTIME;
6154 rc = dt_declare_attr_set(env, child, la, th);
6159 /* a5. insert child's FID into the LFSCK trace file. */
6160 rc = dt_declare_insert(env, obj, (const struct dt_rec *)&flags,
6161 (const struct dt_key *)key, th);
6165 rc = dt_trans_start_local(env, dev, th);
6169 /* b1. remove name entry from backend /lost+found */
6170 rc = dt_delete(env, parent, (const struct dt_key *)ent->lde_name, th);
6174 if (S_ISDIR(lfsck_object_type(child))) {
6175 /* b2. decrease parent's nlink */
6176 dt_write_lock(env, parent, 0);
6177 rc = dt_ref_del(env, parent, th);
6178 dt_write_unlock(env, parent);
6184 /* a3. remove child's FID from the LFSCK trace file. */
6185 rc = dt_delete(env, obj, (const struct dt_key *)key, th);
6189 /* b4. set child's ctime as 1 */
6190 rc = dt_attr_set(env, child, la, th);
6195 /* b5. insert child's FID into the LFSCK trace file. */
6196 rc = dt_insert(env, obj, (const struct dt_rec *)&flags,
6197 (const struct dt_key *)key, th);
6199 GOTO(stop, rc = (rc == 0 ? 1 : rc));
6202 dt_trans_stop(env, dev, th);
6205 lfsck_object_put(env, child);
6211 * Handle orphans under the backend /lost+found directory
6213 * Some backend checker, such as e2fsck for ldiskfs may find some orphans
6214 * and put them under the backend /lost+found directory that is invisible
6215 * to client. The LFSCK will scan such directory, for the original client
6216 * visible orphans, add their fids into the namespace LFSCK trace file,
6217 * then the subsenquent namespace LFSCK second-stage scanning can handle
6218 * them as other objects to be double scanned: either move back to normal
6219 * namespace, or to the global visible orphan directory:
6220 * /ROOT/.lustre/lost+found/MDTxxxx/
6222 * \param[in] env pointer to the thread context
6223 * \param[in] com pointer to the lfsck component
6225 static void lfsck_namespace_scan_local_lpf(const struct lu_env *env,
6226 struct lfsck_component *com)
6228 struct lfsck_thread_info *info = lfsck_env_info(env);
6229 struct lu_dirent *ent =
6230 (struct lu_dirent *)info->lti_key;
6231 struct lu_seq_range *range = &info->lti_range;
6232 struct lfsck_instance *lfsck = com->lc_lfsck;
6233 struct ptlrpc_thread *thread = &lfsck->li_thread;
6234 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
6235 struct lfsck_namespace *ns = com->lc_file_ram;
6236 struct dt_object *parent;
6237 const struct dt_it_ops *iops;
6239 struct seq_server_site *ss = lfsck_dev_site(lfsck);
6241 __u32 idx = lfsck_dev_idx(lfsck);
6246 parent = lfsck_object_find_by_dev(env, lfsck->li_bottom,
6247 &LU_BACKEND_LPF_FID);
6248 if (IS_ERR(parent)) {
6249 CERROR("%s: fail to find backend /lost+found: rc = %ld\n",
6250 lfsck_lfsck2name(lfsck), PTR_ERR(parent));
6254 /* It is normal that the /lost+found does not exist for ZFS backend. */
6255 if (!dt_object_exists(parent))
6258 if (unlikely(!dt_try_as_dir(env, parent)))
6259 GOTO(out, rc = -ENOTDIR);
6261 CDEBUG(D_LFSCK, "%s: start to scan backend /lost+found\n",
6262 lfsck_lfsck2name(lfsck));
6264 com->lc_new_scanned = 0;
6265 iops = &parent->do_index_ops->dio_it;
6266 di = iops->init(env, parent, LUDA_64BITHASH | LUDA_TYPE);
6268 GOTO(out, rc = PTR_ERR(di));
6270 rc = iops->load(env, di, 0);
6272 rc = iops->next(env, di);
6277 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
6278 unlikely(!thread_is_running(thread)))
6281 rc = iops->rec(env, di, (struct dt_rec *)ent,
6282 LUDA_64BITHASH | LUDA_TYPE);
6284 rc = lfsck_unpack_ent(ent, &cookie, &type);
6286 if (unlikely(rc != 0)) {
6287 CDEBUG(D_LFSCK, "%s: fail to iterate backend "
6288 "/lost+found: rc = %d\n",
6289 lfsck_lfsck2name(lfsck), rc);
6294 /* skip dot and dotdot entries */
6295 if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen))
6298 if (!fid_seq_in_fldb(fid_seq(&ent->lde_fid)))
6301 if (fid_is_norm(&ent->lde_fid)) {
6302 fld_range_set_mdt(range);
6303 rc = fld_local_lookup(env, ss->ss_server_fld,
6304 fid_seq(&ent->lde_fid), range);
6307 } else if (idx != 0) {
6308 /* If the returned FID is IGIF, then there are three
6311 * 1) The object is upgraded from old Lustre-1.8 with
6312 * IGIF assigned to such object.
6313 * 2) The object is a backend local object and is
6314 * invisible to client.
6315 * 3) The object lost its LMV EA, and since there is
6316 * no FID-in-dirent for the orphan in the backend
6317 * /lost+found directory, then the low layer will
6318 * return IGIF for such object.
6320 * For MDTx (x != 0), it is either case 2) or case 3),
6321 * but from the LFSCK view, they are indistinguishable.
6322 * To be safe, the LFSCK will keep it there and report
6323 * some message, then the adminstrator can handle that
6326 * For MDT0, it is more possible the case 1). The LFSCK
6327 * will handle the orphan as an upgraded object. */
6328 CDEBUG(D_LFSCK, "%s: the orphan %.*s with IGIF "DFID
6329 "in the backend /lost+found on the MDT %04x, "
6330 "to be safe, skip it.\n",
6331 lfsck_lfsck2name(lfsck), ent->lde_namelen,
6332 ent->lde_name, PFID(&ent->lde_fid), idx);
6336 rc = lfsck_namespace_scan_local_lpf_one(env, com, parent, ent);
6339 down_write(&com->lc_sem);
6340 com->lc_new_scanned++;
6341 ns->ln_local_lpf_scanned++;
6343 ns->ln_local_lpf_moved++;
6345 ns->ln_local_lpf_skipped++;
6347 ns->ln_local_lpf_failed++;
6348 up_write(&com->lc_sem);
6350 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
6354 lfsck_control_speed_by_self(com);
6355 if (unlikely(!thread_is_running(thread))) {
6360 rc = iops->next(env, di);
6364 iops->fini(env, di);
6369 CDEBUG(D_LFSCK, "%s: stop to scan backend /lost+found: rc = %d\n",
6370 lfsck_lfsck2name(lfsck), rc);
6372 lfsck_object_put(env, parent);
6376 * Rescan the striped directory after the master LMV EA reset.
6378 * Sometimes, the master LMV EA of the striped directory maybe lost, so when
6379 * the namespace LFSCK engine scan the striped directory for the first time,
6380 * it will be regarded as a normal directory. As the LFSCK processing, some
6381 * other LFSCK instance on other MDT will find the shard of this striped dir,
6382 * and find that the master MDT-object of the striped directory lost its LMV
6383 * EA, then such remote LFSCK instance will regenerate the master LMV EA and
6384 * notify the LFSCK instance on this MDT to rescan the striped directory.
6386 * \param[in] env pointer to the thread context
6387 * \param[in] com pointer to the lfsck component
6388 * \param[in] llu the lfsck_lmv_unit that contains the striped directory
6391 * \retval positive number for success
6392 * \retval 0 for LFSCK stopped/paused
6393 * \retval negative error number on failure
6395 static int lfsck_namespace_rescan_striped_dir(const struct lu_env *env,
6396 struct lfsck_component *com,
6397 struct lfsck_lmv_unit *llu)
6399 struct lfsck_thread_info *info = lfsck_env_info(env);
6400 struct lfsck_instance *lfsck = com->lc_lfsck;
6401 struct lfsck_assistant_data *lad = com->lc_data;
6402 struct dt_object *dir;
6403 const struct dt_it_ops *iops;
6405 struct lu_dirent *ent =
6406 (struct lu_dirent *)info->lti_key;
6407 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
6408 struct ptlrpc_thread *thread = &lfsck->li_thread;
6409 struct lfsck_assistant_object *lso = NULL;
6410 struct lfsck_namespace_req *lnr;
6411 struct lfsck_assistant_req *lar;
6416 LASSERT(list_empty(&lad->lad_req_list));
6418 lfsck->li_lmv = &llu->llu_lmv;
6419 lfsck->li_obj_dir = lfsck_object_get(llu->llu_obj);
6420 rc = lfsck_open_dir(env, lfsck, 0);
6424 dir = lfsck->li_obj_dir;
6425 di = lfsck->li_di_dir;
6426 iops = &dir->do_index_ops->dio_it;
6428 rc = iops->rec(env, di, (struct dt_rec *)ent,
6429 lfsck->li_args_dir);
6431 rc = lfsck_unpack_ent(ent, &lfsck->li_cookie_dir,
6435 if (bk->lb_param & LPF_FAILOUT)
6441 if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen))
6445 lso = lfsck_assistant_object_init(env,
6446 lfsck_dto2fid(dir), NULL,
6447 lfsck->li_pos_current.lp_oit_cookie, true);
6449 if (bk->lb_param & LPF_FAILOUT)
6450 GOTO(out, rc = PTR_ERR(lso));
6457 lnr = lfsck_namespace_assistant_req_init(lfsck, lso, ent, type);
6459 if (bk->lb_param & LPF_FAILOUT)
6460 GOTO(out, rc = PTR_ERR(lnr));
6465 lar = &lnr->lnr_lar;
6466 rc = lfsck_namespace_assistant_handler_p1(env, com, lar);
6467 lfsck_namespace_assistant_req_fini(env, lar);
6468 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
6471 if (unlikely(!thread_is_running(thread)))
6475 rc = iops->next(env, di);
6479 if (lso != NULL && !IS_ERR(lso))
6480 lfsck_assistant_object_put(env, lso);
6482 lfsck_close_dir(env, lfsck, rc);
6486 /* The close_dir() may insert a dummy lnr in the lad->lad_req_list. */
6487 if (list_empty(&lad->lad_req_list))
6490 spin_lock(&lad->lad_lock);
6491 lar = list_entry(lad->lad_req_list.next, struct lfsck_assistant_req,
6493 list_del_init(&lar->lar_list);
6494 spin_unlock(&lad->lad_lock);
6496 rc = lfsck_namespace_assistant_handler_p1(env, com, lar);
6497 lfsck_namespace_assistant_req_fini(env, lar);
6499 RETURN(rc == 0 ? 1 : rc);
6503 lfsck_namespace_double_scan_one_trace_file(const struct lu_env *env,
6504 struct lfsck_component *com,
6505 struct dt_object *obj, bool first)
6507 struct lfsck_instance *lfsck = com->lc_lfsck;
6508 struct ptlrpc_thread *thread = &lfsck->li_thread;
6509 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
6510 struct lfsck_namespace *ns = com->lc_file_ram;
6511 const struct dt_it_ops *iops = &obj->do_index_ops->dio_it;
6512 struct dt_object *target;
6520 di = iops->init(env, obj, 0);
6522 RETURN(PTR_ERR(di));
6525 fid_cpu_to_be(&fid, &ns->ln_fid_latest_scanned_phase2);
6528 rc = iops->get(env, di, (const struct dt_key *)&fid);
6533 /* The start one either has been processed or does not exist,
6535 rc = iops->next(env, di);
6541 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
6542 unlikely(!thread_is_running(thread)))
6545 key = iops->key(env, di);
6554 fid_be_to_cpu(&fid, (const struct lu_fid *)key);
6555 if (!fid_is_sane(&fid)) {
6560 target = lfsck_object_find_bottom(env, lfsck, &fid);
6561 if (IS_ERR(target)) {
6562 rc = PTR_ERR(target);
6566 if (dt_object_exists(target)) {
6567 rc = iops->rec(env, di, (struct dt_rec *)&flags, 0);
6569 rc = lfsck_namespace_double_scan_one(env, com,
6576 lfsck_object_put(env, target);
6579 down_write(&com->lc_sem);
6580 com->lc_new_checked++;
6581 com->lc_new_scanned++;
6583 ns->ln_fid_latest_scanned_phase2 = fid;
6586 ns->ln_objs_repaired_phase2++;
6588 ns->ln_objs_failed_phase2++;
6589 up_write(&com->lc_sem);
6591 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
6594 if (unlikely(com->lc_time_next_checkpoint <=
6595 ktime_get_seconds()) &&
6596 com->lc_new_checked != 0) {
6597 down_write(&com->lc_sem);
6598 ns->ln_run_time_phase2 += ktime_get_seconds() -
6599 com->lc_time_last_checkpoint;
6600 ns->ln_time_last_checkpoint = ktime_get_real_seconds();
6601 ns->ln_objs_checked_phase2 += com->lc_new_checked;
6602 com->lc_new_checked = 0;
6603 lfsck_namespace_store(env, com);
6604 up_write(&com->lc_sem);
6606 com->lc_time_last_checkpoint = ktime_get_seconds();
6607 com->lc_time_next_checkpoint =
6608 com->lc_time_last_checkpoint +
6609 LFSCK_CHECKPOINT_INTERVAL;
6612 lfsck_control_speed_by_self(com);
6613 if (unlikely(!thread_is_running(thread)))
6616 rc = iops->next(env, di);
6625 iops->fini(env, di);
6630 static int lfsck_namespace_assistant_handler_p2(const struct lu_env *env,
6631 struct lfsck_component *com)
6633 struct lfsck_instance *lfsck = com->lc_lfsck;
6634 struct lfsck_namespace *ns = com->lc_file_ram;
6639 while (!list_empty(&lfsck->li_list_lmv)) {
6640 struct lfsck_lmv_unit *llu;
6642 spin_lock(&lfsck->li_lock);
6643 llu = list_entry(lfsck->li_list_lmv.next,
6644 struct lfsck_lmv_unit, llu_link);
6645 list_del_init(&llu->llu_link);
6646 spin_unlock(&lfsck->li_lock);
6648 rc = lfsck_namespace_rescan_striped_dir(env, com, llu);
6653 CDEBUG(D_LFSCK, "%s: namespace LFSCK phase2 scan start\n",
6654 lfsck_lfsck2name(lfsck));
6656 lfsck_namespace_scan_local_lpf(env, com);
6658 com->lc_new_checked = 0;
6659 com->lc_new_scanned = 0;
6660 com->lc_time_last_checkpoint = ktime_get_seconds();
6661 com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
6662 LFSCK_CHECKPOINT_INTERVAL;
6664 i = lfsck_sub_trace_file_fid2idx(&ns->ln_fid_latest_scanned_phase2);
6665 rc = lfsck_namespace_double_scan_one_trace_file(env, com,
6666 com->lc_sub_trace_objs[i].lsto_obj, true);
6667 while (rc > 0 && ++i < LFSCK_STF_COUNT)
6668 rc = lfsck_namespace_double_scan_one_trace_file(env, com,
6669 com->lc_sub_trace_objs[i].lsto_obj, false);
6671 CDEBUG(D_LFSCK, "%s: namespace LFSCK phase2 scan stop at the No. %d "
6672 "trace file: rc = %d\n", lfsck_lfsck2name(lfsck), i, rc);
6677 static void lfsck_namespace_assistant_fill_pos(const struct lu_env *env,
6678 struct lfsck_component *com,
6679 struct lfsck_position *pos)
6681 struct lfsck_assistant_data *lad = com->lc_data;
6682 struct lfsck_namespace_req *lnr;
6684 if (((struct lfsck_namespace *)(com->lc_file_ram))->ln_status !=
6688 if (list_empty(&lad->lad_req_list))
6691 lnr = list_entry(lad->lad_req_list.next,
6692 struct lfsck_namespace_req,
6694 pos->lp_oit_cookie = lnr->lnr_lar.lar_parent->lso_oit_cookie;
6695 pos->lp_dir_cookie = lnr->lnr_dir_cookie - 1;
6696 pos->lp_dir_parent = lnr->lnr_lar.lar_parent->lso_fid;
6699 static int lfsck_namespace_double_scan_result(const struct lu_env *env,
6700 struct lfsck_component *com,
6703 struct lfsck_instance *lfsck = com->lc_lfsck;
6704 struct lfsck_namespace *ns = com->lc_file_ram;
6706 down_write(&com->lc_sem);
6707 ns->ln_run_time_phase2 += ktime_get_seconds() -
6708 com->lc_time_last_checkpoint;
6709 ns->ln_time_last_checkpoint = ktime_get_real_seconds();
6710 ns->ln_objs_checked_phase2 += com->lc_new_checked;
6711 com->lc_new_checked = 0;
6714 if (ns->ln_flags & LF_INCOMPLETE)
6715 ns->ln_status = LS_PARTIAL;
6717 ns->ln_status = LS_COMPLETED;
6718 ns->ln_flags &= ~LF_SCANNED_ONCE;
6719 if (!(lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN))
6720 ns->ln_flags &= ~LF_INCONSISTENT;
6721 ns->ln_time_last_complete = ns->ln_time_last_checkpoint;
6722 ns->ln_success_count++;
6723 } else if (rc == 0) {
6724 if (lfsck->li_status != 0)
6725 ns->ln_status = lfsck->li_status;
6727 ns->ln_status = LS_STOPPED;
6729 ns->ln_status = LS_FAILED;
6732 rc = lfsck_namespace_store(env, com);
6733 up_write(&com->lc_sem);
6739 lfsck_namespace_assistant_sync_failures_interpret(const struct lu_env *env,
6740 struct ptlrpc_request *req,
6744 struct lfsck_async_interpret_args *laia = args;
6745 struct lfsck_tgt_desc *ltd = laia->laia_ltd;
6747 ltd->ltd_synced_failures = 1;
6754 * Notify remote LFSCK instances about former failures.
6756 * The local LFSCK instance has recorded which MDTs have ever failed to respond
6757 * some LFSCK verification requests (maybe because of network issues or the MDT
6758 * itself trouble). During the respond gap the MDT may missed some name entries
6759 * verification, then the MDT cannot know whether related MDT-objects have been
6760 * referenced by related name entries or not, then in the second-stage scanning,
6761 * these MDT-objects will be regarded as orphan, if the MDT-object contains bad
6762 * linkEA for back reference, then it will misguide the LFSCK to generate wrong
6763 * name entry for repairing the orphan.
6765 * To avoid above trouble, when layout LFSCK finishes the first-stage scanning,
6766 * it will scan the bitmap for the ever failed MDTs, and notify them that they
6767 * have ever missed some name entries verification and should skip the handling
6768 * for orphan MDT-objects.
6770 * \param[in] env pointer to the thread context
6771 * \param[in] com pointer to the lfsck component
6772 * \param[in] lr pointer to the lfsck request
6774 static void lfsck_namespace_assistant_sync_failures(const struct lu_env *env,
6775 struct lfsck_component *com,
6776 struct lfsck_request *lr)
6778 struct lfsck_async_interpret_args *laia =
6779 &lfsck_env_info(env)->lti_laia2;
6780 struct lfsck_assistant_data *lad = com->lc_data;
6781 struct lfsck_namespace *ns = com->lc_file_ram;
6782 struct lfsck_instance *lfsck = com->lc_lfsck;
6783 struct lfsck_tgt_descs *ltds = &lfsck->li_mdt_descs;
6784 struct lfsck_tgt_desc *ltd;
6785 struct ptlrpc_request_set *set;
6790 if (!test_bit(LAD_INCOMPLETE, &lad->lad_flags))
6793 set = ptlrpc_prep_set();
6795 GOTO(out, rc = -ENOMEM);
6797 lr->lr_flags2 = ns->ln_flags | LF_INCOMPLETE;
6798 memset(laia, 0, sizeof(*laia));
6799 lad->lad_touch_gen++;
6801 down_read(<ds->ltd_rw_sem);
6802 cfs_foreach_bit(lad->lad_bitmap, idx) {
6803 ltd = lfsck_ltd2tgt(ltds, idx);
6807 laia->laia_ltd = ltd;
6808 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
6809 lfsck_namespace_assistant_sync_failures_interpret,
6810 laia, LFSCK_NOTIFY);
6812 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail "
6813 "to sync failure with MDT %x: rc = %d\n",
6814 lfsck_lfsck2name(lfsck), ltd->ltd_index, rc);
6816 up_read(<ds->ltd_rw_sem);
6818 rc = ptlrpc_set_wait(env, set);
6819 ptlrpc_set_destroy(set);
6825 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail "
6826 "to sync failure with MDTs, and related MDTs "
6827 "may handle orphan improperly: rc = %d\n",
6828 lfsck_lfsck2name(lfsck), rc);
6833 struct lfsck_assistant_operations lfsck_namespace_assistant_ops = {
6834 .la_handler_p1 = lfsck_namespace_assistant_handler_p1,
6835 .la_handler_p2 = lfsck_namespace_assistant_handler_p2,
6836 .la_fill_pos = lfsck_namespace_assistant_fill_pos,
6837 .la_double_scan_result = lfsck_namespace_double_scan_result,
6838 .la_req_fini = lfsck_namespace_assistant_req_fini,
6839 .la_sync_failures = lfsck_namespace_assistant_sync_failures,
6843 * Verify the specified linkEA entry for the given directory object.
6844 * If the object has no such linkEA entry or it has more other linkEA
6845 * entries, then re-generate the linkEA with the given information.
6847 * \param[in] env pointer to the thread context
6848 * \param[in] obj pointer to the dt_object to be handled
6849 * \param[in] cname the name for the child in the parent directory
6850 * \param[in] pfid the parent directory's FID for the linkEA
6852 * \retval 0 for success
6853 * \retval negative error number on failure
6855 int lfsck_verify_linkea(const struct lu_env *env, struct dt_object *obj,
6856 const struct lu_name *cname, const struct lu_fid *pfid)
6858 struct dt_device *dev = lfsck_obj2dev(obj);
6859 struct linkea_data ldata = { NULL };
6860 struct lu_buf linkea_buf;
6863 int fl = LU_XATTR_CREATE;
6867 LASSERT(S_ISDIR(lfsck_object_type(obj)));
6869 rc = lfsck_links_read_with_rec(env, obj, &ldata);
6870 if (rc == -ENODATA) {
6872 } else if (rc == 0) {
6873 fl = LU_XATTR_REPLACE;
6874 if (ldata.ld_leh->leh_reccount != 1) {
6877 rc = linkea_links_find(&ldata, cname, pfid);
6886 rc = linkea_links_new(&ldata, &lfsck_env_info(env)->lti_linkea_buf,
6891 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
6892 ldata.ld_leh->leh_len);
6893 th = dt_trans_create(env, dev);
6895 RETURN(PTR_ERR(th));
6897 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
6898 XATTR_NAME_LINK, fl, th);
6902 rc = dt_trans_start_local(env, dev, th);
6906 dt_write_lock(env, obj, 0);
6907 rc = dt_xattr_set(env, obj, &linkea_buf,
6908 XATTR_NAME_LINK, fl, th);
6909 dt_write_unlock(env, obj);
6914 dt_trans_stop(env, dev, th);
6919 * Get the name and parent directory's FID from the first linkEA entry.
6921 * \param[in] env pointer to the thread context
6922 * \param[in] obj pointer to the object which get linkEA from
6923 * \param[out] name pointer to the buffer to hold the name
6924 * in the first linkEA entry
6925 * \param[out] pfid pointer to the buffer to hold the parent
6926 * directory's FID in the first linkEA entry
6928 * \retval 0 for success
6929 * \retval negative error number on failure
6931 int lfsck_links_get_first(const struct lu_env *env, struct dt_object *obj,
6932 char *name, struct lu_fid *pfid)
6934 struct lu_name *cname = &lfsck_env_info(env)->lti_name;
6935 struct linkea_data ldata = { NULL };
6938 rc = lfsck_links_read_with_rec(env, obj, &ldata);
6942 linkea_first_entry(&ldata);
6943 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, cname, pfid);
6944 if (!linkea_entry_is_valid(&ldata, cname, pfid))
6947 /* To guarantee the 'name' is terminated with '0'. */
6948 memcpy(name, cname->ln_name, cname->ln_namelen);
6949 name[cname->ln_namelen] = 0;
6955 * Update the object's name entry with the given FID.
6957 * \param[in] env pointer to the thread context
6958 * \param[in] lfsck pointer to the lfsck instance
6959 * \param[in] dir pointer to the directory that holds
6961 * \param[in] name the name for the entry to be updated
6962 * \param[in] fid the new FID for the name entry referenced
6963 * \param[in] type the type for the name entry to be updated
6965 * \retval 0 for success
6966 * \retval negative error number on failure
6968 int lfsck_update_name_entry(const struct lu_env *env,
6969 struct lfsck_instance *lfsck,
6970 struct dt_object *dir, const char *name,
6971 const struct lu_fid *fid, __u32 type)
6973 struct lfsck_thread_info *info = lfsck_env_info(env);
6974 struct dt_insert_rec *rec = &info->lti_dt_rec;
6975 struct lfsck_lock_handle *llh = &info->lti_llh;
6976 struct dt_device *dev = lfsck_obj2dev(dir);
6982 rc = lfsck_lock(env, lfsck, dir, name, llh,
6983 MDS_INODELOCK_UPDATE, LCK_PW);
6987 th = dt_trans_create(env, dev);
6989 GOTO(unlock, rc = PTR_ERR(th));
6991 rc = dt_declare_delete(env, dir, (const struct dt_key *)name, th);
6995 rec->rec_type = type;
6997 rc = dt_declare_insert(env, dir, (const struct dt_rec *)rec,
6998 (const struct dt_key *)name, th);
7002 rc = dt_declare_ref_add(env, dir, th);
7006 rc = dt_trans_start_local(env, dev, th);
7010 rc = dt_delete(env, dir, (const struct dt_key *)name, th);
7011 if (rc == -ENOENT) {
7019 rc = dt_insert(env, dir, (const struct dt_rec *)rec,
7020 (const struct dt_key *)name, th);
7021 if (rc == 0 && S_ISDIR(type) && !exists) {
7022 dt_write_lock(env, dir, 0);
7023 rc = dt_ref_add(env, dir, th);
7024 dt_write_unlock(env, dir);
7030 dt_trans_stop(env, dev, th);
7034 CDEBUG(D_LFSCK, "%s: update name entry "DFID"/%s with the FID "DFID
7035 " and the type %o: rc = %d\n", lfsck_lfsck2name(lfsck),
7036 PFID(lfsck_dto2fid(dir)), name, PFID(fid), type, rc);
7041 int lfsck_namespace_setup(const struct lu_env *env,
7042 struct lfsck_instance *lfsck)
7044 struct lfsck_component *com;
7045 struct lfsck_namespace *ns;
7046 struct dt_object *root = NULL;
7047 struct dt_object *obj;
7052 LASSERT(lfsck->li_master);
7058 INIT_LIST_HEAD(&com->lc_link);
7059 INIT_LIST_HEAD(&com->lc_link_dir);
7060 init_rwsem(&com->lc_sem);
7061 atomic_set(&com->lc_ref, 1);
7062 com->lc_lfsck = lfsck;
7063 com->lc_type = LFSCK_TYPE_NAMESPACE;
7064 com->lc_ops = &lfsck_namespace_ops;
7065 com->lc_data = lfsck_assistant_data_init(
7066 &lfsck_namespace_assistant_ops,
7068 if (com->lc_data == NULL)
7069 GOTO(out, rc = -ENOMEM);
7071 com->lc_file_size = sizeof(struct lfsck_namespace);
7072 OBD_ALLOC(com->lc_file_ram, com->lc_file_size);
7073 if (com->lc_file_ram == NULL)
7074 GOTO(out, rc = -ENOMEM);
7076 OBD_ALLOC(com->lc_file_disk, com->lc_file_size);
7077 if (com->lc_file_disk == NULL)
7078 GOTO(out, rc = -ENOMEM);
7080 for (i = 0; i < LFSCK_STF_COUNT; i++)
7081 mutex_init(&com->lc_sub_trace_objs[i].lsto_mutex);
7083 root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
7085 GOTO(out, rc = PTR_ERR(root));
7087 if (unlikely(!dt_try_as_dir(env, root)))
7088 GOTO(out, rc = -ENOTDIR);
7090 obj = local_index_find_or_create(env, lfsck->li_los, root,
7092 S_IFREG | S_IRUGO | S_IWUSR,
7093 &dt_lfsck_namespace_features);
7095 GOTO(out, rc = PTR_ERR(obj));
7098 rc = lfsck_namespace_load(env, com);
7099 if (rc == -ENODATA) {
7100 rc = lfsck_namespace_init(env, com);
7101 } else if (rc < 0) {
7102 rc = lfsck_namespace_reset(env, com, true);
7104 rc = lfsck_load_sub_trace_files(env, com,
7105 &dt_lfsck_namespace_features, LFSCK_NAMESPACE, false);
7107 rc = lfsck_namespace_reset(env, com, true);
7112 ns = com->lc_file_ram;
7113 switch (ns->ln_status) {
7118 spin_lock(&lfsck->li_lock);
7119 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
7120 spin_unlock(&lfsck->li_lock);
7123 CERROR("%s: unknown lfsck_namespace status %d\n",
7124 lfsck_lfsck2name(lfsck), ns->ln_status);
7126 case LS_SCANNING_PHASE1:
7127 case LS_SCANNING_PHASE2:
7128 /* No need to store the status to disk right now.
7129 * If the system crashed before the status stored,
7130 * it will be loaded back when next time. */
7131 ns->ln_status = LS_CRASHED;
7135 spin_lock(&lfsck->li_lock);
7136 list_add_tail(&com->lc_link, &lfsck->li_list_scan);
7137 list_add_tail(&com->lc_link_dir, &lfsck->li_list_dir);
7138 spin_unlock(&lfsck->li_lock);
7145 if (root != NULL && !IS_ERR(root))
7146 lfsck_object_put(env, root);
7148 lfsck_component_cleanup(env, com);
7149 CERROR("%s: fail to init namespace LFSCK component: rc = %d\n",
7150 lfsck_lfsck2name(lfsck), rc);