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>
39 #include <lu_target.h>
41 #include "lfsck_internal.h"
43 #define LFSCK_NAMESPACE_MAGIC_V1 0xA0629D03
44 #define LFSCK_NAMESPACE_MAGIC_V2 0xA0621A0B
45 #define LFSCK_NAMESPACE_MAGIC_V3 0xA06249FF
47 /* For Lustre-2.x (x <= 6), the namespace LFSCK used LFSCK_NAMESPACE_MAGIC_V1
48 * as the trace file magic. When downgrade to such old release, the old LFSCK
49 * will not recognize the new LFSCK_NAMESPACE_MAGIC_V2 in the new trace file,
50 * then it will reset the whole LFSCK, and will not cause start failure. The
51 * similar case will happen when upgrade from such old release. */
52 #define LFSCK_NAMESPACE_MAGIC LFSCK_NAMESPACE_MAGIC_V3
54 enum lfsck_nameentry_check {
55 LFSCK_NAMEENTRY_DEAD = 1, /* The object has been unlinked. */
56 LFSCK_NAMEENTRY_REMOVED = 2, /* The entry has been removed. */
57 LFSCK_NAMEENTRY_RECREATED = 3, /* The entry has been recreated. */
60 static struct lfsck_namespace_req *
61 lfsck_namespace_assistant_req_init(struct lfsck_instance *lfsck,
62 struct lfsck_assistant_object *lso,
63 struct lu_dirent *ent, __u16 type)
65 struct lfsck_namespace_req *lnr;
68 size = sizeof(*lnr) + (ent->lde_namelen & ~3) + 4;
71 return ERR_PTR(-ENOMEM);
73 INIT_LIST_HEAD(&lnr->lnr_lar.lar_list);
74 lnr->lnr_lar.lar_parent = lfsck_assistant_object_get(lso);
75 lnr->lnr_lmv = lfsck_lmv_get(lfsck->li_lmv);
76 lnr->lnr_fid = ent->lde_fid;
77 lnr->lnr_dir_cookie = ent->lde_hash;
78 lnr->lnr_attr = ent->lde_attrs;
81 lnr->lnr_namelen = ent->lde_namelen;
82 memcpy(lnr->lnr_name, ent->lde_name, ent->lde_namelen);
87 static void lfsck_namespace_assistant_req_fini(const struct lu_env *env,
88 struct lfsck_assistant_req *lar)
90 struct lfsck_namespace_req *lnr =
91 container_of(lar, struct lfsck_namespace_req, lnr_lar);
93 if (lnr->lnr_lmv != NULL)
94 lfsck_lmv_put(env, lnr->lnr_lmv);
96 lfsck_assistant_object_put(env, lar->lar_parent);
97 OBD_FREE(lnr, lnr->lnr_size);
100 static void lfsck_namespace_le_to_cpu(struct lfsck_namespace *dst,
101 struct lfsck_namespace *src)
103 dst->ln_magic = le32_to_cpu(src->ln_magic);
104 dst->ln_status = le32_to_cpu(src->ln_status);
105 dst->ln_flags = le32_to_cpu(src->ln_flags);
106 dst->ln_success_count = le32_to_cpu(src->ln_success_count);
107 dst->ln_run_time_phase1 = le64_to_cpu(src->ln_run_time_phase1);
108 dst->ln_run_time_phase2 = le64_to_cpu(src->ln_run_time_phase2);
109 dst->ln_time_last_complete = le64_to_cpu(src->ln_time_last_complete);
110 dst->ln_time_latest_start = le64_to_cpu(src->ln_time_latest_start);
111 dst->ln_time_last_checkpoint =
112 le64_to_cpu(src->ln_time_last_checkpoint);
113 lfsck_position_le_to_cpu(&dst->ln_pos_latest_start,
114 &src->ln_pos_latest_start);
115 lfsck_position_le_to_cpu(&dst->ln_pos_last_checkpoint,
116 &src->ln_pos_last_checkpoint);
117 lfsck_position_le_to_cpu(&dst->ln_pos_first_inconsistent,
118 &src->ln_pos_first_inconsistent);
119 dst->ln_items_checked = le64_to_cpu(src->ln_items_checked);
120 dst->ln_items_repaired = le64_to_cpu(src->ln_items_repaired);
121 dst->ln_items_failed = le64_to_cpu(src->ln_items_failed);
122 dst->ln_dirs_checked = le64_to_cpu(src->ln_dirs_checked);
123 dst->ln_objs_checked_phase2 = le64_to_cpu(src->ln_objs_checked_phase2);
124 dst->ln_objs_repaired_phase2 =
125 le64_to_cpu(src->ln_objs_repaired_phase2);
126 dst->ln_objs_failed_phase2 = le64_to_cpu(src->ln_objs_failed_phase2);
127 dst->ln_objs_nlink_repaired = le64_to_cpu(src->ln_objs_nlink_repaired);
128 fid_le_to_cpu(&dst->ln_fid_latest_scanned_phase2,
129 &src->ln_fid_latest_scanned_phase2);
130 dst->ln_dirent_repaired = le64_to_cpu(src->ln_dirent_repaired);
131 dst->ln_linkea_repaired = le64_to_cpu(src->ln_linkea_repaired);
132 dst->ln_mul_linked_checked = le64_to_cpu(src->ln_mul_linked_checked);
133 dst->ln_mul_linked_repaired = le64_to_cpu(src->ln_mul_linked_repaired);
134 dst->ln_unknown_inconsistency =
135 le64_to_cpu(src->ln_unknown_inconsistency);
136 dst->ln_unmatched_pairs_repaired =
137 le64_to_cpu(src->ln_unmatched_pairs_repaired);
138 dst->ln_dangling_repaired = le64_to_cpu(src->ln_dangling_repaired);
139 dst->ln_mul_ref_repaired = le64_to_cpu(src->ln_mul_ref_repaired);
140 dst->ln_bad_type_repaired = le64_to_cpu(src->ln_bad_type_repaired);
141 dst->ln_lost_dirent_repaired =
142 le64_to_cpu(src->ln_lost_dirent_repaired);
143 dst->ln_striped_dirs_scanned =
144 le64_to_cpu(src->ln_striped_dirs_scanned);
145 dst->ln_striped_dirs_repaired =
146 le64_to_cpu(src->ln_striped_dirs_repaired);
147 dst->ln_striped_dirs_failed =
148 le64_to_cpu(src->ln_striped_dirs_failed);
149 dst->ln_striped_dirs_disabled =
150 le64_to_cpu(src->ln_striped_dirs_disabled);
151 dst->ln_striped_dirs_skipped =
152 le64_to_cpu(src->ln_striped_dirs_skipped);
153 dst->ln_striped_shards_scanned =
154 le64_to_cpu(src->ln_striped_shards_scanned);
155 dst->ln_striped_shards_repaired =
156 le64_to_cpu(src->ln_striped_shards_repaired);
157 dst->ln_striped_shards_failed =
158 le64_to_cpu(src->ln_striped_shards_failed);
159 dst->ln_striped_shards_skipped =
160 le64_to_cpu(src->ln_striped_shards_skipped);
161 dst->ln_name_hash_repaired = le64_to_cpu(src->ln_name_hash_repaired);
162 dst->ln_local_lpf_scanned = le64_to_cpu(src->ln_local_lpf_scanned);
163 dst->ln_local_lpf_moved = le64_to_cpu(src->ln_local_lpf_moved);
164 dst->ln_local_lpf_skipped = le64_to_cpu(src->ln_local_lpf_skipped);
165 dst->ln_local_lpf_failed = le64_to_cpu(src->ln_local_lpf_failed);
166 dst->ln_bitmap_size = le32_to_cpu(src->ln_bitmap_size);
167 dst->ln_time_latest_reset = le64_to_cpu(src->ln_time_latest_reset);
168 dst->ln_linkea_overflow_cleared =
169 le64_to_cpu(src->ln_linkea_overflow_cleared);
170 dst->ln_agent_entries_repaired =
171 le64_to_cpu(src->ln_agent_entries_repaired);
174 static void lfsck_namespace_cpu_to_le(struct lfsck_namespace *dst,
175 struct lfsck_namespace *src)
177 dst->ln_magic = cpu_to_le32(src->ln_magic);
178 dst->ln_status = cpu_to_le32(src->ln_status);
179 dst->ln_flags = cpu_to_le32(src->ln_flags);
180 dst->ln_success_count = cpu_to_le32(src->ln_success_count);
181 dst->ln_run_time_phase1 = cpu_to_le64(src->ln_run_time_phase1);
182 dst->ln_run_time_phase2 = cpu_to_le64(src->ln_run_time_phase2);
183 dst->ln_time_last_complete = cpu_to_le64(src->ln_time_last_complete);
184 dst->ln_time_latest_start = cpu_to_le64(src->ln_time_latest_start);
185 dst->ln_time_last_checkpoint =
186 cpu_to_le64(src->ln_time_last_checkpoint);
187 lfsck_position_cpu_to_le(&dst->ln_pos_latest_start,
188 &src->ln_pos_latest_start);
189 lfsck_position_cpu_to_le(&dst->ln_pos_last_checkpoint,
190 &src->ln_pos_last_checkpoint);
191 lfsck_position_cpu_to_le(&dst->ln_pos_first_inconsistent,
192 &src->ln_pos_first_inconsistent);
193 dst->ln_items_checked = cpu_to_le64(src->ln_items_checked);
194 dst->ln_items_repaired = cpu_to_le64(src->ln_items_repaired);
195 dst->ln_items_failed = cpu_to_le64(src->ln_items_failed);
196 dst->ln_dirs_checked = cpu_to_le64(src->ln_dirs_checked);
197 dst->ln_objs_checked_phase2 = cpu_to_le64(src->ln_objs_checked_phase2);
198 dst->ln_objs_repaired_phase2 =
199 cpu_to_le64(src->ln_objs_repaired_phase2);
200 dst->ln_objs_failed_phase2 = cpu_to_le64(src->ln_objs_failed_phase2);
201 dst->ln_objs_nlink_repaired = cpu_to_le64(src->ln_objs_nlink_repaired);
202 fid_cpu_to_le(&dst->ln_fid_latest_scanned_phase2,
203 &src->ln_fid_latest_scanned_phase2);
204 dst->ln_dirent_repaired = cpu_to_le64(src->ln_dirent_repaired);
205 dst->ln_linkea_repaired = cpu_to_le64(src->ln_linkea_repaired);
206 dst->ln_mul_linked_checked = cpu_to_le64(src->ln_mul_linked_checked);
207 dst->ln_mul_linked_repaired = cpu_to_le64(src->ln_mul_linked_repaired);
208 dst->ln_unknown_inconsistency =
209 cpu_to_le64(src->ln_unknown_inconsistency);
210 dst->ln_unmatched_pairs_repaired =
211 cpu_to_le64(src->ln_unmatched_pairs_repaired);
212 dst->ln_dangling_repaired = cpu_to_le64(src->ln_dangling_repaired);
213 dst->ln_mul_ref_repaired = cpu_to_le64(src->ln_mul_ref_repaired);
214 dst->ln_bad_type_repaired = cpu_to_le64(src->ln_bad_type_repaired);
215 dst->ln_lost_dirent_repaired =
216 cpu_to_le64(src->ln_lost_dirent_repaired);
217 dst->ln_striped_dirs_scanned =
218 cpu_to_le64(src->ln_striped_dirs_scanned);
219 dst->ln_striped_dirs_repaired =
220 cpu_to_le64(src->ln_striped_dirs_repaired);
221 dst->ln_striped_dirs_failed =
222 cpu_to_le64(src->ln_striped_dirs_failed);
223 dst->ln_striped_dirs_disabled =
224 cpu_to_le64(src->ln_striped_dirs_disabled);
225 dst->ln_striped_dirs_skipped =
226 cpu_to_le64(src->ln_striped_dirs_skipped);
227 dst->ln_striped_shards_scanned =
228 cpu_to_le64(src->ln_striped_shards_scanned);
229 dst->ln_striped_shards_repaired =
230 cpu_to_le64(src->ln_striped_shards_repaired);
231 dst->ln_striped_shards_failed =
232 cpu_to_le64(src->ln_striped_shards_failed);
233 dst->ln_striped_shards_skipped =
234 cpu_to_le64(src->ln_striped_shards_skipped);
235 dst->ln_name_hash_repaired = cpu_to_le64(src->ln_name_hash_repaired);
236 dst->ln_local_lpf_scanned = cpu_to_le64(src->ln_local_lpf_scanned);
237 dst->ln_local_lpf_moved = cpu_to_le64(src->ln_local_lpf_moved);
238 dst->ln_local_lpf_skipped = cpu_to_le64(src->ln_local_lpf_skipped);
239 dst->ln_local_lpf_failed = cpu_to_le64(src->ln_local_lpf_failed);
240 dst->ln_bitmap_size = cpu_to_le32(src->ln_bitmap_size);
241 dst->ln_time_latest_reset = cpu_to_le64(src->ln_time_latest_reset);
242 dst->ln_linkea_overflow_cleared =
243 cpu_to_le64(src->ln_linkea_overflow_cleared);
244 dst->ln_agent_entries_repaired =
245 cpu_to_le64(src->ln_agent_entries_repaired);
248 static void lfsck_namespace_record_failure(const struct lu_env *env,
249 struct lfsck_instance *lfsck,
250 struct lfsck_namespace *ns)
252 struct lfsck_position pos;
254 ns->ln_items_failed++;
255 lfsck_pos_fill(env, lfsck, &pos, false);
256 if (lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent) ||
257 lfsck_pos_is_eq(&pos, &ns->ln_pos_first_inconsistent) < 0) {
258 ns->ln_pos_first_inconsistent = pos;
260 CDEBUG(D_LFSCK, "%s: namespace LFSCK hit first non-repaired "
261 "inconsistency at the pos [%llu, "DFID", %#llx]\n",
262 lfsck_lfsck2name(lfsck),
263 ns->ln_pos_first_inconsistent.lp_oit_cookie,
264 PFID(&ns->ln_pos_first_inconsistent.lp_dir_parent),
265 ns->ln_pos_first_inconsistent.lp_dir_cookie);
270 * Load the MDT bitmap from the lfsck_namespace trace file.
272 * \param[in] env pointer to the thread context
273 * \param[in] com pointer to the lfsck component
275 * \retval 0 for success
276 * \retval negative error number on failure or data corruption
278 static int lfsck_namespace_load_bitmap(const struct lu_env *env,
279 struct lfsck_component *com)
281 struct dt_object *obj = com->lc_obj;
282 struct lfsck_assistant_data *lad = com->lc_data;
283 struct lfsck_namespace *ns = com->lc_file_ram;
284 unsigned long *bitmap = lad->lad_bitmap;
290 if (com->lc_lfsck->li_mdt_descs.ltd_tgts_mask_len >
292 nbits = com->lc_lfsck->li_mdt_descs.ltd_tgts_mask_len;
294 nbits = ns->ln_bitmap_size;
296 if (unlikely(nbits < BITS_PER_LONG))
297 nbits = BITS_PER_LONG;
299 if (nbits > lad->lad_bitmap_count) {
300 u32 new_bits = lad->lad_bitmap_count;
301 unsigned long *new_bitmap;
303 while (new_bits < nbits)
306 new_bitmap = bitmap_zalloc(new_bits, GFP_KERNEL);
307 if (new_bitmap == NULL)
310 lad->lad_bitmap = new_bitmap;
311 lad->lad_bitmap_count = new_bits;
316 if (ns->ln_bitmap_size == 0) {
317 clear_bit(LAD_INCOMPLETE, &lad->lad_flags);
318 bitmap_zero(bitmap, lad->lad_bitmap_count);
322 size = (ns->ln_bitmap_size + 7) >> 3;
323 rc = dt_xattr_get(env, obj,
324 lfsck_buf_get(env, bitmap, size),
325 XATTR_NAME_LFSCK_BITMAP);
327 RETURN(rc >= 0 ? -EINVAL : rc);
329 if (bitmap_empty(bitmap, lad->lad_bitmap_count))
330 clear_bit(LAD_INCOMPLETE, &lad->lad_flags);
332 set_bit(LAD_INCOMPLETE, &lad->lad_flags);
338 * Load namespace LFSCK statistics information from the trace file.
340 * \param[in] env pointer to the thread context
341 * \param[in] com pointer to the lfsck component
343 * \retval 0 for success
344 * \retval negative error number on failure
346 static int lfsck_namespace_load(const struct lu_env *env,
347 struct lfsck_component *com)
349 int len = com->lc_file_size;
352 rc = dt_xattr_get(env, com->lc_obj,
353 lfsck_buf_get(env, com->lc_file_disk, len),
354 XATTR_NAME_LFSCK_NAMESPACE);
356 struct lfsck_namespace *ns = com->lc_file_ram;
358 lfsck_namespace_le_to_cpu(ns,
359 (struct lfsck_namespace *)com->lc_file_disk);
360 if (ns->ln_magic != LFSCK_NAMESPACE_MAGIC) {
361 CDEBUG(D_LFSCK, "%s: invalid lfsck_namespace magic "
362 "%#x != %#x\n", lfsck_lfsck2name(com->lc_lfsck),
363 ns->ln_magic, LFSCK_NAMESPACE_MAGIC);
368 } else if (rc != -ENODATA) {
369 CDEBUG(D_LFSCK, "%s: fail to load lfsck_namespace, "
370 "expected = %d: rc = %d\n",
371 lfsck_lfsck2name(com->lc_lfsck), len, rc);
379 static int lfsck_namespace_store(const struct lu_env *env,
380 struct lfsck_component *com)
382 struct dt_object *obj = com->lc_obj;
383 struct lfsck_instance *lfsck = com->lc_lfsck;
384 struct lfsck_namespace *ns = com->lc_file_ram;
385 struct lfsck_assistant_data *lad = com->lc_data;
386 struct dt_device *dev = lfsck_obj2dev(obj);
387 unsigned long *bitmap = NULL;
388 struct thandle *handle;
390 int len = com->lc_file_size;
395 bitmap = lad->lad_bitmap;
396 nbits = lad->lad_bitmap_count;
399 LASSERTF((nbits & 7) == 0, "Invalid nbits %u\n", nbits);
402 ns->ln_bitmap_size = nbits;
403 lfsck_namespace_cpu_to_le((struct lfsck_namespace *)com->lc_file_disk,
405 handle = dt_trans_create(env, dev);
407 GOTO(log, rc = PTR_ERR(handle));
409 rc = dt_declare_xattr_set(env, obj,
410 lfsck_buf_get(env, com->lc_file_disk, len),
411 XATTR_NAME_LFSCK_NAMESPACE, 0, handle);
415 if (bitmap != NULL) {
416 rc = dt_declare_xattr_set(env, obj,
417 lfsck_buf_get(env, bitmap, nbits >> 3),
418 XATTR_NAME_LFSCK_BITMAP, 0, handle);
423 rc = dt_trans_start_local(env, dev, handle);
427 rc = dt_xattr_set(env, obj,
428 lfsck_buf_get(env, com->lc_file_disk, len),
429 XATTR_NAME_LFSCK_NAMESPACE, 0, handle);
430 if (rc == 0 && bitmap != NULL)
431 rc = dt_xattr_set(env, obj,
432 lfsck_buf_get(env, bitmap, nbits >> 3),
433 XATTR_NAME_LFSCK_BITMAP, 0, handle);
438 dt_trans_stop(env, dev, handle);
442 CDEBUG(D_LFSCK, "%s: fail to store lfsck_namespace: rc = %d\n",
443 lfsck_lfsck2name(lfsck), rc);
447 static int lfsck_namespace_init(const struct lu_env *env,
448 struct lfsck_component *com)
450 struct lfsck_namespace *ns = com->lc_file_ram;
453 memset(ns, 0, sizeof(*ns));
454 ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
455 ns->ln_status = LS_INIT;
456 ns->ln_time_latest_reset = ktime_get_real_seconds();
457 down_write(&com->lc_sem);
458 rc = lfsck_namespace_store(env, com);
460 rc = lfsck_load_sub_trace_files(env, com,
461 &dt_lfsck_namespace_features, LFSCK_NAMESPACE, true);
462 up_write(&com->lc_sem);
468 * Update the namespace LFSCK trace file for the given @fid
470 * \param[in] env pointer to the thread context
471 * \param[in] com pointer to the lfsck component
472 * \param[in] fid the fid which flags to be updated in the lfsck
474 * \param[in] add true if add new flags, otherwise remove flags
476 * \retval 0 for success or nothing to be done
477 * \retval negative error number on failure
479 int lfsck_namespace_trace_update(const struct lu_env *env,
480 struct lfsck_component *com,
481 const struct lu_fid *fid,
482 const __u8 flags, bool add)
484 struct lfsck_instance *lfsck = com->lc_lfsck;
485 struct dt_object *obj;
486 struct lu_fid *key = &lfsck_env_info(env)->lti_fid3;
487 struct dt_device *dev;
488 struct thandle *th = NULL;
497 if (unlikely(!fid_is_sane(fid)))
500 idx = lfsck_sub_trace_file_fid2idx(fid);
501 mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
502 obj = com->lc_sub_trace_objs[idx].lsto_obj;
503 if (unlikely(obj == NULL)) {
504 mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
508 lfsck_object_get(obj);
509 dev = lfsck_obj2dev(obj);
510 fid_cpu_to_be(key, fid);
511 rc = dt_lookup(env, obj, (struct dt_rec *)&old,
512 (const struct dt_key *)key);
515 GOTO(unlock, rc = 0);
519 } else if (rc == 0) {
521 if ((old & flags) == flags)
522 GOTO(unlock, rc = 0);
526 if ((old & flags) == 0)
527 GOTO(unlock, rc = 0);
535 th = lfsck_trans_create(env, dev, lfsck);
537 GOTO(log, rc = PTR_ERR(th));
540 rc = dt_declare_delete(env, obj,
541 (const struct dt_key *)key, th);
547 rc = dt_declare_insert(env, obj,
548 (const struct dt_rec *)&new,
549 (const struct dt_key *)key, th);
554 rc = dt_trans_start_local(env, dev, th);
559 rc = dt_delete(env, obj, (const struct dt_key *)key, th);
565 rc = dt_insert(env, obj, (const struct dt_rec *)&new,
566 (const struct dt_key *)key, th);
574 if (th != NULL && !IS_ERR(th))
575 dt_trans_stop(env, dev, th);
577 CDEBUG(D_LFSCK, "%s: namespace LFSCK %s flags for "DFID" in the "
578 "trace file, flags %x, old %x, new %x: rc = %d\n",
579 lfsck_lfsck2name(lfsck), add ? "add" : "del", PFID(fid),
580 (__u32)flags, (__u32)old, (__u32)new, rc);
583 mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
584 lfsck_object_put(env, obj);
589 int lfsck_namespace_check_exist(const struct lu_env *env,
590 struct dt_object *dir,
591 struct dt_object *obj, const char *name)
593 struct lu_fid *fid = &lfsck_env_info(env)->lti_fid;
597 if (unlikely(lfsck_is_dead_obj(obj)))
598 RETURN(LFSCK_NAMEENTRY_DEAD);
600 rc = dt_lookup_dir(env, dir, name, fid);
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 = lfsck_trans_create(env, dev, lfsck);
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, false);
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_dir(env, parent, info->lti_key, &tfid);
950 if (rc != 0 && rc != -ENOENT)
953 if (unlikely(rc == 0 && lu_fid_eq(cfid, &tfid)))
955 } while (rc == 0 && !exist);
957 rc = lfsck_lock(env, lfsck, parent, info->lti_key, pllh,
958 MDS_INODELOCK_UPDATE, LCK_PW);
962 /* Re-check whether the name conflict with othrs after taken
964 rc = dt_lookup_dir(env, parent, info->lti_key, &tfid);
966 if (!lu_fid_eq(cfid, &tfid)) {
973 } else if (rc != -ENOENT) {
979 cname->ln_name = info->lti_key;
980 cname->ln_namelen = namelen;
981 rc = linkea_links_new(&ldata2, &info->lti_linkea_buf2,
986 rc = lfsck_ibits_lock(env, lfsck, orphan, &clh,
987 MDS_INODELOCK_UPDATE | MDS_INODELOCK_LOOKUP |
988 MDS_INODELOCK_XATTR, LCK_EX);
992 lfsck_buf_init(&linkea_buf, ldata2.ld_buf->lb_buf,
993 ldata2.ld_leh->leh_len);
994 th = lfsck_trans_create(env, dev, lfsck);
996 GOTO(log, rc = PTR_ERR(th));
998 if (S_ISDIR(lfsck_object_type(orphan))) {
999 rc = dt_declare_delete(env, orphan,
1000 (const struct dt_key *)dotdot, th);
1004 rec->rec_type = S_IFDIR;
1005 rec->rec_fid = pfid;
1006 rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
1007 (const struct dt_key *)dotdot, th);
1012 rc = dt_declare_xattr_set(env, orphan, &linkea_buf,
1013 XATTR_NAME_LINK, 0, th);
1018 rec->rec_type = lfsck_object_type(orphan) & S_IFMT;
1019 rec->rec_fid = cfid;
1020 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1021 (const struct dt_key *)cname->ln_name,
1026 if (S_ISDIR(rec->rec_type)) {
1027 rc = dt_declare_ref_add(env, parent, th);
1033 memset(la, 0, sizeof(*la));
1034 la->la_ctime = ktime_get_real_seconds();
1035 la->la_valid = LA_CTIME;
1036 rc = dt_declare_attr_set(env, orphan, la, th);
1040 rc = dt_trans_start_local(env, dev, th);
1044 dt_write_lock(env, orphan, 0);
1045 rc = lfsck_links_read2_with_rec(env, orphan, &ldata2);
1046 if (likely(rc == -ENODATA || rc == -EINVAL)) {
1047 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1048 GOTO(unlock, rc = 1);
1050 if (S_ISDIR(lfsck_object_type(orphan))) {
1051 rc = dt_delete(env, orphan,
1052 (const struct dt_key *)dotdot, th);
1056 rec->rec_type = S_IFDIR;
1057 rec->rec_fid = pfid;
1058 rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1059 (const struct dt_key *)dotdot, th);
1064 rc = dt_xattr_set(env, orphan, &linkea_buf, XATTR_NAME_LINK, 0,
1067 if (rc == 0 && count != NULL)
1068 *count = ldata2.ld_leh->leh_reccount;
1072 dt_write_unlock(env, orphan);
1074 if (rc == 0 && !exist) {
1075 rec->rec_type = lfsck_object_type(orphan) & S_IFMT;
1076 rec->rec_fid = cfid;
1077 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1078 (const struct dt_key *)cname->ln_name, th);
1079 if (rc == 0 && S_ISDIR(rec->rec_type)) {
1080 dt_write_lock(env, parent, 0);
1081 rc = dt_ref_add(env, parent, th);
1082 dt_write_unlock(env, parent);
1087 rc = dt_attr_set(env, orphan, la, th);
1089 GOTO(stop, rc = (rc == 0 ? 1 : rc));
1092 dt_write_unlock(env, orphan);
1095 dt_trans_stop(env, dev, th);
1098 lfsck_ibits_unlock(&clh, LCK_EX);
1100 CDEBUG(D_LFSCK, "%s: namespace LFSCK insert orphan for the "
1101 "object "DFID", name = %s: rc = %d\n",
1102 lfsck_lfsck2name(lfsck), PFID(cfid),
1103 cname->ln_name != NULL ? cname->ln_name : "<NULL>", rc);
1106 struct lfsck_namespace *ns = com->lc_file_ram;
1108 ns->ln_flags |= LF_INCONSISTENT;
1114 static int lfsck_lmv_set(const struct lu_env *env,
1115 struct lfsck_instance *lfsck,
1116 struct dt_object *obj,
1117 struct lmv_mds_md_v1 *lmv)
1119 struct dt_device *dev = lfsck->li_next;
1120 struct thandle *th = NULL;
1121 struct lu_buf buf = { lmv, sizeof(*lmv) };
1126 th = lfsck_trans_create(env, dev, lfsck);
1128 RETURN(PTR_ERR(th));
1130 rc = dt_declare_xattr_set(env, obj, &buf, XATTR_NAME_LMV, 0, th);
1134 rc = dt_trans_start_local(env, dev, th);
1138 rc = dt_xattr_set(env, obj, &buf, XATTR_NAME_LMV, 0, th);
1144 dt_trans_stop(env, dev, th);
1149 static int lfsck_lmv_delete(const struct lu_env *env,
1150 struct lfsck_instance *lfsck,
1151 struct dt_object *obj)
1153 struct dt_device *dev = lfsck->li_next;
1154 struct thandle *th = NULL;
1159 th = lfsck_trans_create(env, dev, lfsck);
1161 RETURN(PTR_ERR(th));
1163 rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LMV, th);
1167 rc = dt_trans_start_local(env, dev, th);
1171 rc = dt_xattr_del(env, obj, XATTR_NAME_LMV, th);
1177 dt_trans_stop(env, dev, th);
1182 static inline int lfsck_object_is_shard(const struct lu_env *env,
1183 struct lfsck_instance *lfsck,
1184 struct dt_object *obj,
1185 const struct lu_name *lname)
1187 struct lfsck_thread_info *info = lfsck_env_info(env);
1188 struct lmv_mds_md_v1 *lmv = &info->lti_lmv;
1191 rc = lfsck_shard_name_to_index(env, lname->ln_name, lname->ln_namelen,
1192 lfsck_object_type(obj),
1193 lfsck_dto2fid(obj));
1197 rc = lfsck_read_stripe_lmv(env, lfsck, obj, lmv);
1201 if (!rc && lmv->lmv_magic == LMV_MAGIC_STRIPE)
1208 * Add the specified name entry back to namespace.
1210 * If there is a linkEA entry that back references a name entry under
1211 * some parent directory, but such parent directory does not have the
1212 * claimed name entry. On the other hand, the linkEA entries count is
1213 * not larger than the MDT-object's hard link count. Under such case,
1214 * it is quite possible that the name entry is lost. Then the LFSCK
1215 * should add the name entry back to the namespace.
1217 * If \a child is shard, which means \a parent is a striped directory,
1218 * if \a parent has LMV, we need to delete it before insertion because
1219 * now parent's striping is broken and can't be parsed correctly.
1221 * \param[in] env pointer to the thread context
1222 * \param[in] com pointer to the lfsck component
1223 * \param[in] parent pointer to the directory under which the name entry
1224 * will be inserted into
1225 * \param[in] child pointer to the object referenced by the name entry
1226 * that to be inserted into the parent
1227 * \param[in] lname the name for the child in the parent directory
1229 * \retval positive number for repaired cases
1230 * \retval 0 if nothing to be repaired
1231 * \retval negative error number on failure
1233 static int lfsck_namespace_insert_normal(const struct lu_env *env,
1234 struct lfsck_component *com,
1235 struct dt_object *parent,
1236 struct dt_object *child,
1237 const struct lu_name *lname)
1239 struct lfsck_thread_info *info = lfsck_env_info(env);
1240 struct lu_attr *la = &info->lti_la;
1241 struct dt_insert_rec *rec = &info->lti_dt_rec;
1242 struct lfsck_instance *lfsck = com->lc_lfsck;
1243 /* The child and its name may be on different MDTs. */
1244 const struct lu_fid *pfid = lfsck_dto2fid(parent);
1245 const struct lu_fid *cfid = lfsck_dto2fid(child);
1246 struct dt_device *dev = lfsck->li_next;
1247 struct thandle *th = NULL;
1248 struct lfsck_lock_handle *llh = &info->lti_llh;
1249 struct lmv_mds_md_v1 *lmv = &info->lti_lmv;
1250 struct lu_buf buf = { lmv, sizeof(*lmv) };
1251 /* whether parent's LMV is deleted before insertion */
1252 bool parent_lmv_deleted = false;
1253 /* whether parent's LMV is missing */
1254 bool parent_lmv_lost = false;
1259 /* @parent/@child may be based on lfsck->li_bottom,
1260 * but here we need the object based on the lfsck->li_next. */
1262 parent = lfsck_object_locate(dev, parent);
1264 GOTO(log, rc = PTR_ERR(parent));
1266 child = lfsck_object_locate(dev, child);
1268 GOTO(log, rc = PTR_ERR(child));
1270 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1273 rc = lfsck_lock(env, lfsck, parent, lname->ln_name, llh,
1274 MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
1275 MDS_INODELOCK_XATTR, LCK_EX);
1279 rc = lfsck_object_is_shard(env, lfsck, child, lname);
1284 rc = lfsck_read_stripe_lmv(env, lfsck, parent, lmv);
1287 * To add a shard, we need to convert parent to a
1288 * plain directory by deleting its LMV, and after
1289 * insertion set it back.
1291 rc = lfsck_lmv_delete(env, lfsck, parent);
1294 parent_lmv_deleted = true;
1295 lmv->lmv_layout_version++;
1296 lfsck_lmv_header_cpu_to_le(lmv, lmv);
1297 } else if (rc == -ENODATA) {
1298 struct lu_seq_range *range = &info->lti_range;
1299 struct seq_server_site *ss = lfsck_dev_site(lfsck);
1301 rc = lfsck_read_stripe_lmv(env, lfsck, child, lmv);
1305 fld_range_set_mdt(range);
1306 rc = fld_server_lookup(env, ss->ss_server_fld,
1307 fid_seq(lfsck_dto2fid(parent)), range);
1311 parent_lmv_lost = true;
1312 lmv->lmv_magic = LMV_MAGIC;
1313 lmv->lmv_master_mdt_index = range->lsr_index;
1314 lmv->lmv_layout_version++;
1315 lfsck_lmv_header_cpu_to_le(lmv, lmv);
1321 if (unlikely(!dt_try_as_dir(env, parent, true)))
1322 GOTO(unlock, rc = -ENOTDIR);
1324 th = lfsck_trans_create(env, dev, lfsck);
1326 GOTO(unlock, rc = PTR_ERR(th));
1328 rec->rec_type = lfsck_object_type(child) & S_IFMT;
1329 rec->rec_fid = cfid;
1330 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1331 (const struct dt_key *)lname->ln_name, th);
1335 if (S_ISDIR(rec->rec_type)) {
1336 rc = dt_declare_ref_add(env, parent, th);
1341 if (parent_lmv_lost) {
1342 rc = dt_declare_xattr_set(env, parent, &buf, XATTR_NAME_LMV,
1348 la->la_ctime = ktime_get_real_seconds();
1349 la->la_valid = LA_CTIME;
1350 rc = dt_declare_attr_set(env, parent, la, th);
1354 rc = dt_declare_attr_set(env, child, la, th);
1358 rc = dt_trans_start_local(env, dev, th);
1362 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1363 (const struct dt_key *)lname->ln_name, th);
1367 if (S_ISDIR(rec->rec_type)) {
1368 dt_write_lock(env, parent, 0);
1369 rc = dt_ref_add(env, parent, th);
1370 dt_write_unlock(env, parent);
1375 if (parent_lmv_lost) {
1376 rc = dt_xattr_set(env, parent, &buf, XATTR_NAME_LMV, 0, th);
1381 rc = dt_attr_set(env, parent, la, th);
1385 rc = dt_attr_set(env, child, la, th);
1387 GOTO(stop, rc = (rc == 0 ? 1 : rc));
1390 dt_trans_stop(env, dev, th);
1393 if (parent_lmv_deleted)
1394 lfsck_lmv_set(env, lfsck, parent, lmv);
1399 CDEBUG(D_LFSCK, "%s: namespace LFSCK insert object "DFID" with "
1400 "the name %s and type %o to the parent "DFID": rc = %d\n",
1401 lfsck_lfsck2name(lfsck), PFID(cfid), lname->ln_name,
1402 lfsck_object_type(child) & S_IFMT, PFID(pfid), rc);
1405 struct lfsck_namespace *ns = com->lc_file_ram;
1407 ns->ln_flags |= LF_INCONSISTENT;
1409 ns->ln_lost_dirent_repaired++;
1416 * Create the specified orphan directory.
1418 * For the case that the parent MDT-object stored in some MDT-object's
1419 * linkEA entry is lost, the LFSCK will re-create the parent object as
1420 * an orphan and insert it into .lustre/lost+found/MDTxxxx/ directory
1421 * with the name ${FID}-P-${conflict_version}.
1423 * \param[in] env pointer to the thread context
1424 * \param[in] com pointer to the lfsck component
1425 * \param[in] orphan pointer to the orphan MDT-object to be created
1426 * \param[in] lmv pointer to master LMV EA that will be set to the orphan
1428 * \retval positive number for repaired cases
1429 * \retval negative error number on failure
1431 static int lfsck_namespace_create_orphan_dir(const struct lu_env *env,
1432 struct lfsck_component *com,
1433 struct dt_object *orphan,
1434 struct lmv_mds_md_v1 *lmv)
1436 struct lfsck_thread_info *info = lfsck_env_info(env);
1437 struct lu_attr *la = &info->lti_la;
1438 struct dt_allocation_hint *hint = &info->lti_hint;
1439 struct dt_object_format *dof = &info->lti_dof;
1440 struct lu_name *cname = &info->lti_name2;
1441 struct dt_insert_rec *rec = &info->lti_dt_rec;
1442 struct lmv_mds_md_v1 *lmv2 = &info->lti_lmv2;
1443 const struct lu_fid *cfid = lfsck_dto2fid(orphan);
1445 struct lfsck_instance *lfsck = com->lc_lfsck;
1446 struct lfsck_namespace *ns = com->lc_file_ram;
1447 struct dt_device *dev = lfsck_obj2dev(orphan);
1448 struct dt_object *parent = NULL;
1449 struct thandle *th = NULL;
1450 struct lfsck_lock_handle *llh = &info->lti_llh;
1451 struct linkea_data ldata = { NULL };
1452 struct lu_buf linkea_buf;
1453 struct lu_buf lmv_buf;
1461 LASSERT(!dt_object_exists(orphan));
1463 cname->ln_name = NULL;
1464 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1467 if (dt_object_remote(orphan)) {
1468 if (lfsck->li_lpf_root_obj == NULL)
1469 GOTO(log, rc = -EBADF);
1471 idx = lfsck_find_mdt_idx_by_fid(env, lfsck, cfid);
1473 GOTO(log, rc = idx);
1475 snprintf(name, 8, "MDT%04x", idx);
1476 rc = dt_lookup_dir(env, lfsck->li_lpf_root_obj, name, &tfid);
1478 GOTO(log, rc = (rc == -ENOENT ? -ENXIO : rc));
1480 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
1482 GOTO(log, rc = PTR_ERR(parent));
1484 if (unlikely(!dt_try_as_dir(env, parent, true)))
1485 GOTO(log, rc = -ENOTDIR);
1487 if (unlikely(lfsck->li_lpf_obj == NULL))
1488 GOTO(log, rc = -ENXIO);
1490 parent = lfsck->li_lpf_obj;
1493 dev = lfsck_find_dev_by_fid(env, lfsck, cfid);
1495 GOTO(log, rc = PTR_ERR(dev));
1501 namelen = snprintf(name, 31, DFID"-P-%d",
1503 rc = dt_lookup_dir(env, parent, name, &tfid);
1504 if (rc != 0 && rc != -ENOENT)
1508 rc = lfsck_lock(env, lfsck, parent, name, llh,
1509 MDS_INODELOCK_UPDATE, LCK_PW);
1513 /* Re-check whether the name conflict with othrs after taken
1515 rc = dt_lookup_dir(env, parent, name, &tfid);
1516 if (unlikely(rc == 0)) {
1524 cname->ln_name = name;
1525 cname->ln_namelen = namelen;
1527 memset(la, 0, sizeof(*la));
1528 la->la_mode = S_IFDIR | 0700;
1529 la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
1530 LA_ATIME | LA_MTIME | LA_CTIME;
1532 orphan->do_ops->do_ah_init(env, hint, parent, orphan,
1533 la->la_mode & S_IFMT);
1535 memset(dof, 0, sizeof(*dof));
1536 dof->dof_type = dt_mode_to_dft(S_IFDIR);
1538 rc = linkea_links_new(&ldata, &info->lti_linkea_buf2,
1539 cname, lfsck_dto2fid(parent));
1543 th = lfsck_trans_create(env, dev, lfsck);
1545 GOTO(unlock1, rc = PTR_ERR(th));
1547 /* Sync the remote transaction to guarantee that the subsequent
1548 * lock against the @orphan can find the @orphan in time. */
1549 if (dt_object_remote(orphan))
1552 rc = dt_declare_create(env, orphan, la, hint, dof, th);
1556 if (unlikely(!dt_try_as_dir(env, orphan, false)))
1557 GOTO(stop, rc = -ENOTDIR);
1559 rc = dt_declare_ref_add(env, orphan, th);
1563 rec->rec_type = S_IFDIR;
1564 rec->rec_fid = cfid;
1565 rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
1566 (const struct dt_key *)dot, th);
1570 rec->rec_fid = lfsck_dto2fid(parent);
1571 rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
1572 (const struct dt_key *)dotdot, th);
1577 lmv->lmv_magic = LMV_MAGIC;
1578 lmv->lmv_master_mdt_index = lfsck_dev_idx(lfsck);
1579 lfsck_lmv_header_cpu_to_le(lmv2, lmv);
1580 lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
1581 rc = dt_declare_xattr_set(env, orphan, &lmv_buf, XATTR_NAME_LMV,
1587 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
1588 ldata.ld_leh->leh_len);
1589 rc = dt_declare_xattr_set(env, orphan, &linkea_buf,
1590 XATTR_NAME_LINK, 0, th);
1594 rec->rec_fid = cfid;
1595 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1596 (const struct dt_key *)name, th);
1598 rc = dt_declare_ref_add(env, parent, th);
1603 rc = dt_trans_start_local(env, dev, th);
1607 dt_write_lock(env, orphan, 0);
1608 rc = dt_create(env, orphan, la, hint, dof, th);
1612 rc = dt_ref_add(env, orphan, th);
1616 rec->rec_fid = cfid;
1617 rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1618 (const struct dt_key *)dot, th);
1622 rec->rec_fid = lfsck_dto2fid(parent);
1623 rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1624 (const struct dt_key *)dotdot, th);
1629 rc = dt_xattr_set(env, orphan, &lmv_buf, XATTR_NAME_LMV, 0, th);
1634 rc = dt_xattr_set(env, orphan, &linkea_buf,
1635 XATTR_NAME_LINK, 0, th);
1636 dt_write_unlock(env, orphan);
1640 rec->rec_fid = cfid;
1641 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1642 (const struct dt_key *)name, th);
1644 dt_write_lock(env, parent, 0);
1645 rc = dt_ref_add(env, parent, th);
1646 dt_write_unlock(env, parent);
1649 GOTO(stop, rc = (rc == 0 ? 1 : rc));
1652 dt_write_unlock(env, orphan);
1655 rc1 = dt_trans_stop(env, dev, th);
1656 if (rc1 != 0 && rc > 0)
1663 CDEBUG(D_LFSCK, "%s: namespace LFSCK create orphan dir for "
1664 "the object "DFID", name = %s: rc = %d\n",
1665 lfsck_lfsck2name(lfsck), PFID(cfid),
1666 cname->ln_name != NULL ? cname->ln_name : "<NULL>", rc);
1668 if (parent != NULL && !IS_ERR(parent) && parent != lfsck->li_lpf_obj)
1669 lfsck_object_put(env, parent);
1672 ns->ln_flags |= LF_INCONSISTENT;
1678 * Remove the specified entry from the linkEA.
1680 * Locate the linkEA entry with the given @cname and @pfid, then
1681 * remove this entry or the other entries those are repeated with
1684 * \param[in] env pointer to the thread context
1685 * \param[in] com pointer to the lfsck component
1686 * \param[in] obj pointer to the dt_object to be handled
1687 * \param[in,out]ldata pointer to the buffer that holds the linkEA
1688 * \param[in] cname the name for the child in the parent directory
1689 * \param[in] pfid the parent directory's FID for the linkEA
1690 * \param[in] next if true, then remove the first found linkEA
1691 * entry, and move the ldata->ld_lee to next entry
1693 * \retval positive number for repaired cases
1694 * \retval 0 if nothing to be repaired
1695 * \retval negative error number on failure
1697 static int lfsck_namespace_shrink_linkea(const struct lu_env *env,
1698 struct lfsck_component *com,
1699 struct dt_object *obj,
1700 struct linkea_data *ldata,
1701 struct lu_name *cname,
1702 struct lu_fid *pfid,
1705 struct lfsck_instance *lfsck = com->lc_lfsck;
1706 struct dt_device *dev = lfsck_obj2dev(obj);
1707 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
1708 struct thandle *th = NULL;
1709 struct lustre_handle lh = { 0 };
1710 struct linkea_data ldata_new = { NULL };
1711 struct lu_buf linkea_buf;
1716 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
1717 MDS_INODELOCK_UPDATE | MDS_INODELOCK_XATTR,
1723 lfsck_linkea_del_buf(ldata, cname);
1725 lfsck_namespace_filter_linkea_entry(ldata, cname, pfid,
1727 if (ldata->ld_leh->leh_reccount > 0 ||
1728 unlikely(ldata->ld_leh->leh_overflow_time)) {
1729 lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
1730 ldata->ld_leh->leh_len);
1731 buflen = linkea_buf.lb_len;
1735 th = lfsck_trans_create(env, dev, lfsck);
1737 GOTO(unlock1, rc = PTR_ERR(th));
1740 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
1741 XATTR_NAME_LINK, 0, th);
1743 rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, th);
1747 rc = dt_trans_start_local(env, dev, th);
1751 dt_write_lock(env, obj, 0);
1752 if (unlikely(lfsck_is_dead_obj(obj)))
1753 GOTO(unlock2, rc = -ENOENT);
1755 rc = lfsck_links_read2_with_rec(env, obj, &ldata_new);
1757 GOTO(unlock2, rc = (rc == -ENODATA ? 0 : rc));
1759 /* The specified linkEA entry has been removed by race. */
1760 rc = linkea_links_find(&ldata_new, cname, pfid);
1762 GOTO(unlock2, rc = 0);
1764 if (bk->lb_param & LPF_DRYRUN)
1765 GOTO(unlock2, rc = 1);
1768 lfsck_linkea_del_buf(&ldata_new, cname);
1770 lfsck_namespace_filter_linkea_entry(&ldata_new, cname, pfid,
1774 * linkea may change because it doesn't take lock in the first read, if
1775 * it becomes larger, restart from beginning.
1777 if ((ldata_new.ld_leh->leh_reccount > 0 ||
1778 unlikely(ldata_new.ld_leh->leh_overflow_time)) &&
1779 buflen < ldata_new.ld_leh->leh_len) {
1780 dt_write_unlock(env, obj);
1781 dt_trans_stop(env, dev, th);
1782 lfsck_buf_init(&linkea_buf, ldata_new.ld_buf->lb_buf,
1783 ldata_new.ld_leh->leh_len);
1784 buflen = linkea_buf.lb_len;
1789 rc = lfsck_links_write(env, obj, &ldata_new, th);
1791 rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, th);
1793 GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
1796 dt_write_unlock(env, obj);
1799 dt_trans_stop(env, dev, th);
1802 lfsck_ibits_unlock(&lh, LCK_EX);
1805 CDEBUG(D_LFSCK, "%s: namespace LFSCK remove %s linkEA entry "
1806 "for the object: "DFID", parent "DFID", name %.*s\n",
1807 lfsck_lfsck2name(lfsck), next ? "invalid" : "redundant",
1808 PFID(lfsck_dto2fid(obj)), PFID(pfid), cname->ln_namelen,
1812 struct lfsck_namespace *ns = com->lc_file_ram;
1814 ns->ln_flags |= LF_INCONSISTENT;
1821 * Conditionally remove the specified entry from the linkEA.
1823 * Take the parent lock firstly, then check whether the specified
1824 * name entry exists or not: if yes, do nothing; otherwise, call
1825 * lfsck_namespace_shrink_linkea() to remove the linkea entry.
1827 * \param[in] env pointer to the thread context
1828 * \param[in] com pointer to the lfsck component
1829 * \param[in] parent pointer to the parent directory
1830 * \param[in] child pointer to the child object that holds the linkEA
1831 * \param[in,out]ldata pointer to the buffer that holds the linkEA
1832 * \param[in] cname the name for the child in the parent directory
1833 * \param[in] pfid the parent directory's FID for the linkEA
1835 * \retval positive number for repaired cases
1836 * \retval 0 if nothing to be repaired
1837 * \retval negative error number on failure
1839 static int lfsck_namespace_shrink_linkea_cond(const struct lu_env *env,
1840 struct lfsck_component *com,
1841 struct dt_object *parent,
1842 struct dt_object *child,
1843 struct linkea_data *ldata,
1844 struct lu_name *cname,
1845 struct lu_fid *pfid)
1847 struct lfsck_thread_info *info = lfsck_env_info(env);
1848 struct lu_fid *cfid = &info->lti_fid3;
1849 struct lfsck_lock_handle *llh = &info->lti_llh;
1853 rc = lfsck_lock(env, com->lc_lfsck, parent, cname->ln_name, llh,
1854 MDS_INODELOCK_UPDATE, LCK_PR);
1858 dt_read_lock(env, parent, 0);
1859 if (unlikely(lfsck_is_dead_obj(parent))) {
1860 dt_read_unlock(env, parent);
1862 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
1868 rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
1869 (const struct dt_key *)cname->ln_name);
1870 dt_read_unlock(env, parent);
1872 /* It is safe to release the ldlm lock, because when the logic come
1873 * here, we have got all the needed information above whether the
1874 * linkEA entry is valid or not. It is not important that others
1875 * may add new linkEA entry after the ldlm lock released. If other
1876 * has removed the specified linkEA entry by race, then it is OK,
1877 * because the subsequent lfsck_namespace_shrink_linkea() can handle
1880 if (rc == -ENOENT) {
1881 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
1890 /* The LFSCK just found some internal status of cross-MDTs
1891 * create operation. That is normal. */
1892 if (lu_fid_eq(cfid, lfsck_dto2fid(child))) {
1893 linkea_next_entry(ldata);
1898 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata, cname,
1905 * Conditionally replace name entry in the parent.
1907 * As required, the LFSCK may re-create the lost MDT-object for dangling
1908 * name entry, but such repairing may be wrong because of bad FID in the
1909 * name entry. As the LFSCK processing, the real MDT-object may be found,
1910 * then the LFSCK should check whether the former re-created MDT-object
1911 * has been modified or not, if not, then destroy it and update the name
1912 * entry in the parent to reference the real MDT-object.
1914 * \param[in] env pointer to the thread context
1915 * \param[in] com pointer to the lfsck component
1916 * \param[in] parent pointer to the parent directory
1917 * \param[in] child pointer to the MDT-object that may be the real
1918 * MDT-object corresponding to the name entry in parent
1919 * \param[in] cfid the current FID in the name entry
1920 * \param[in] cname contains the name of the child in the parent directory
1922 * \retval positive number for repaired cases
1923 * \retval 0 if nothing to be repaired
1924 * \retval negative error number on failure
1926 static int lfsck_namespace_replace_cond(const struct lu_env *env,
1927 struct lfsck_component *com,
1928 struct dt_object *parent,
1929 struct dt_object *child,
1930 const struct lu_fid *cfid,
1931 const struct lu_name *cname)
1933 struct lfsck_thread_info *info = lfsck_env_info(env);
1934 struct lu_attr *la = &info->lti_la;
1935 struct dt_insert_rec *rec = &info->lti_dt_rec;
1937 struct lfsck_instance *lfsck = com->lc_lfsck;
1938 /* The child and its name may be on different MDTs. */
1939 struct dt_device *dev = lfsck->li_next;
1940 const char *name = cname->ln_name;
1941 const struct lu_fid *pfid = lfsck_dto2fid(parent);
1942 struct dt_object *cobj = NULL;
1943 struct lfsck_lock_handle *pllh = &info->lti_llh;
1944 struct lustre_handle clh = { 0 };
1945 struct linkea_data ldata = { NULL };
1946 struct thandle *th = NULL;
1951 /* @parent/@child may be based on lfsck->li_bottom,
1952 * but here we need the object based on the lfsck->li_next. */
1954 parent = lfsck_object_locate(dev, parent);
1956 GOTO(log, rc = PTR_ERR(parent));
1958 if (unlikely(!dt_try_as_dir(env, parent, true)))
1959 GOTO(log, rc = -ENOTDIR);
1961 rc = lfsck_lock(env, lfsck, parent, name, pllh,
1962 MDS_INODELOCK_UPDATE, LCK_PW);
1966 if (!fid_is_sane(cfid)) {
1971 cobj = lfsck_object_find_by_dev(env, dev, cfid);
1974 if (rc == -ENOENT) {
1982 if (!dt_object_exists(cobj)) {
1987 rc = dt_lookup_dir(env, parent, name, &tfid);
1988 if (rc == -ENOENT) {
1996 /* Someone changed the name entry, cannot replace it. */
1997 if (!lu_fid_eq(cfid, &tfid))
2000 /* lock the object to be destroyed. */
2001 rc = lfsck_ibits_lock(env, lfsck, cobj, &clh,
2002 MDS_INODELOCK_UPDATE |
2003 MDS_INODELOCK_UPDATE | MDS_INODELOCK_XATTR,
2008 if (unlikely(lfsck_is_dead_obj(cobj))) {
2013 rc = dt_attr_get(env, cobj, la);
2017 /* The object has been modified by other(s), or it is not created by
2018 * LFSCK, the two cases are indistinguishable. So cannot replace it. */
2019 if (la->la_ctime != 0)
2022 if (S_ISREG(la->la_mode)) {
2023 rc = dt_xattr_get(env, cobj, &LU_BUF_NULL, XATTR_NAME_LOV);
2024 /* If someone has created related OST-object(s),
2026 if ((rc > 0) || (rc < 0 && rc != -ENODATA))
2027 GOTO(log, rc = (rc > 0 ? 0 : rc));
2031 dt_read_lock(env, child, 0);
2032 rc = lfsck_links_read2_with_rec(env, child, &ldata);
2033 dt_read_unlock(env, child);
2035 /* Someone changed the child, no need to replace. */
2042 rc = linkea_links_find(&ldata, cname, pfid);
2043 /* Someone moved the child, no need to replace. */
2047 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2050 th = lfsck_trans_create(env, dev, lfsck);
2052 GOTO(log, rc = PTR_ERR(th));
2055 rc = dt_declare_destroy(env, cobj, th);
2060 rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
2064 rec->rec_type = S_IFDIR;
2065 rec->rec_fid = lfsck_dto2fid(child);
2066 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
2067 (const struct dt_key *)name, th);
2071 rc = dt_trans_start_local(env, dev, th);
2076 rc = dt_destroy(env, cobj, th);
2081 /* The old name entry maybe not exist. */
2082 rc = dt_delete(env, parent, (const struct dt_key *)name, th);
2083 if (rc != 0 && rc != -ENOENT)
2086 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
2087 (const struct dt_key *)name, th);
2089 GOTO(stop, rc = (rc == 0 ? 1 : rc));
2092 dt_trans_stop(env, dev, th);
2095 lfsck_ibits_unlock(&clh, LCK_EX);
2098 if (cobj != NULL && !IS_ERR(cobj))
2099 lfsck_object_put(env, cobj);
2101 CDEBUG(D_LFSCK, "%s: namespace LFSCK conditionally destroy the "
2102 "object "DFID" because of conflict with the object "DFID
2103 " under the parent "DFID" with name %s: rc = %d\n",
2104 lfsck_lfsck2name(lfsck), PFID(cfid),
2105 PFID(lfsck_dto2fid(child)), PFID(pfid), name, rc);
2111 * Overwrite the linkEA for the object with the given ldata.
2113 * The caller should take the ldlm lock before the calling.
2115 * \param[in] env pointer to the thread context
2116 * \param[in] com pointer to the lfsck component
2117 * \param[in] obj pointer to the dt_object to be handled
2118 * \param[in] ldata pointer to the new linkEA data
2120 * \retval positive number for repaired cases
2121 * \retval 0 if nothing to be repaired
2122 * \retval negative error number on failure
2124 int lfsck_namespace_rebuild_linkea(const struct lu_env *env,
2125 struct lfsck_component *com,
2126 struct dt_object *obj,
2127 struct linkea_data *ldata)
2129 struct lfsck_instance *lfsck = com->lc_lfsck;
2130 struct dt_device *dev = lfsck_obj2dev(obj);
2131 struct thandle *th = NULL;
2132 struct lu_buf linkea_buf;
2136 th = lfsck_trans_create(env, dev, lfsck);
2138 GOTO(log, rc = PTR_ERR(th));
2140 lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
2141 ldata->ld_leh->leh_len);
2142 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
2143 XATTR_NAME_LINK, 0, th);
2147 rc = dt_trans_start_local(env, dev, th);
2151 dt_write_lock(env, obj, 0);
2152 if (unlikely(lfsck_is_dead_obj(obj)))
2153 GOTO(unlock, rc = 0);
2155 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2156 GOTO(unlock, rc = 1);
2158 rc = dt_xattr_set(env, obj, &linkea_buf,
2159 XATTR_NAME_LINK, 0, th);
2161 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2164 dt_write_unlock(env, obj);
2167 dt_trans_stop(env, dev, th);
2170 CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild linkEA for the "
2171 "object "DFID": rc = %d\n",
2172 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
2175 struct lfsck_namespace *ns = com->lc_file_ram;
2177 ns->ln_flags |= LF_INCONSISTENT;
2184 * Repair invalid name entry.
2186 * If the name entry contains invalid information, such as bad file type
2187 * or (and) corrupted object FID, then either remove the name entry or
2188 * udpate the name entry with the given (right) information.
2190 * \param[in] env pointer to the thread context
2191 * \param[in] com pointer to the lfsck component
2192 * \param[in] parent pointer to the parent directory
2193 * \param[in] child pointer to the object referenced by the name entry
2194 * \param[in] name the old name of the child under the parent directory
2195 * \param[in] name2 the new name of the child under the parent directory
2196 * \param[in] type the type claimed by the name entry
2197 * \param[in] update update the name entry if true; otherwise, remove it
2198 * \param[in] dec decrease the parent nlink count if true
2200 * \retval positive number for repaired successfully
2201 * \retval 0 if nothing to be repaired
2202 * \retval negative error number on failure
2204 int lfsck_namespace_repair_dirent(const struct lu_env *env,
2205 struct lfsck_component *com,
2206 struct dt_object *parent,
2207 struct dt_object *child,
2208 const char *name, const char *name2,
2209 __u16 type, bool update, bool dec)
2211 struct lfsck_thread_info *info = lfsck_env_info(env);
2212 struct dt_insert_rec *rec = &info->lti_dt_rec;
2213 const struct lu_fid *pfid = lfsck_dto2fid(parent);
2214 struct lu_fid cfid = {0};
2216 struct lfsck_instance *lfsck = com->lc_lfsck;
2217 struct dt_device *dev = lfsck->li_next;
2218 struct thandle *th = NULL;
2219 struct lfsck_lock_handle *llh = &info->lti_llh;
2220 struct lustre_handle lh = { 0 };
2225 cfid = *lfsck_dto2fid(child);
2226 parent = lfsck_object_locate(dev, parent);
2228 GOTO(log, rc = PTR_ERR(parent));
2230 if (unlikely(!dt_try_as_dir(env, parent, true)))
2231 GOTO(log, rc = -ENOTDIR);
2233 if (!update || strcmp(name, name2) == 0)
2234 rc = lfsck_lock(env, lfsck, parent, name, llh,
2235 MDS_INODELOCK_UPDATE, LCK_PW);
2237 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
2238 MDS_INODELOCK_UPDATE, LCK_PW);
2242 th = lfsck_trans_create(env, dev, lfsck);
2244 GOTO(unlock1, rc = PTR_ERR(th));
2246 rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
2251 rec->rec_type = lfsck_object_type(child) & S_IFMT;
2252 LASSERT(!fid_is_zero(&cfid));
2253 rec->rec_fid = &cfid;
2254 rc = dt_declare_insert(env, parent,
2255 (const struct dt_rec *)rec,
2256 (const struct dt_key *)name2, th);
2261 if (dec && S_ISDIR(type)) {
2262 rc = dt_declare_ref_del(env, parent, th);
2267 rc = dt_trans_start_local(env, dev, th);
2272 dt_write_lock(env, parent, 0);
2273 rc = dt_lookup_dir(env, dt_object_child(parent), name, &tfid);
2274 /* Someone has removed the bad name entry by race. */
2276 GOTO(unlock2, rc = 0);
2281 /* Someone has removed the bad name entry and reused it for other
2282 * object by race. */
2283 if (!lu_fid_eq(&tfid, &cfid))
2284 GOTO(unlock2, rc = 0);
2286 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2287 GOTO(unlock2, rc = 1);
2289 rc = dt_delete(env, parent, (const struct dt_key *)name, th);
2294 rc = dt_insert(env, parent,
2295 (const struct dt_rec *)rec,
2296 (const struct dt_key *)name2, th);
2301 if (dec && S_ISDIR(type)) {
2302 rc = dt_ref_del(env, parent, th);
2307 GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
2310 dt_write_unlock(env, parent);
2313 dt_trans_stop(env, dev, th);
2315 /* We are not sure whether the child will become orphan or not.
2316 * Record it in the LFSCK trace file for further checking in
2317 * the second-stage scanning. */
2318 if (!update && !dec && child && rc == 0)
2319 lfsck_namespace_trace_update(env, com, &cfid,
2320 LNTF_CHECK_LINKEA, true);
2323 /* It is harmless even if unlock the unused lock_handle */
2324 lfsck_ibits_unlock(&lh, LCK_PW);
2328 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found bad name "
2329 "entry for: parent "DFID", child "DFID", name %s, type "
2330 "in name entry %o, type claimed by child %o. repair it "
2331 "by %s with new name2 %s: rc = %d\n",
2332 lfsck_lfsck2name(lfsck), PFID(pfid), PFID(&cfid),
2333 name, type, update ? lfsck_object_type(child) : 0,
2334 update ? "updating" : "removing", name2, rc);
2337 struct lfsck_namespace *ns = com->lc_file_ram;
2339 ns->ln_flags |= LF_INCONSISTENT;
2346 * Update the ".." name entry for the given object.
2348 * The object's ".." is corrupted, this function will update the ".." name
2349 * entry with the given pfid, and the linkEA with the given ldata.
2351 * The caller should take the ldlm lock before the calling.
2353 * \param[in] env pointer to the thread context
2354 * \param[in] com pointer to the lfsck component
2355 * \param[in] obj pointer to the dt_object to be handled
2356 * \param[in] pfid the new fid for the object's ".." name entry
2357 * \param[in] cname the name for the @obj in the parent directory
2359 * \retval positive number for repaired cases
2360 * \retval 0 if nothing to be repaired
2361 * \retval negative error number on failure
2363 static int lfsck_namespace_repair_unmatched_pairs(const struct lu_env *env,
2364 struct lfsck_component *com,
2365 struct dt_object *obj,
2366 const struct lu_fid *pfid,
2367 struct lu_name *cname)
2369 struct lfsck_thread_info *info = lfsck_env_info(env);
2370 struct dt_insert_rec *rec = &info->lti_dt_rec;
2371 struct lfsck_instance *lfsck = com->lc_lfsck;
2372 struct dt_device *dev = lfsck_obj2dev(obj);
2373 struct thandle *th = NULL;
2374 struct linkea_data ldata = { NULL };
2375 struct lu_buf linkea_buf;
2379 LASSERT(!dt_object_remote(obj));
2380 LASSERT(S_ISDIR(lfsck_object_type(obj)));
2382 rc = linkea_links_new(&ldata, &info->lti_big_buf, cname, pfid);
2386 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
2387 ldata.ld_leh->leh_len);
2389 th = lfsck_trans_create(env, dev, lfsck);
2391 GOTO(log, rc = PTR_ERR(th));
2393 rc = dt_declare_delete(env, obj, (const struct dt_key *)dotdot, th);
2397 rec->rec_type = S_IFDIR;
2398 rec->rec_fid = pfid;
2399 rc = dt_declare_insert(env, obj, (const struct dt_rec *)rec,
2400 (const struct dt_key *)dotdot, th);
2404 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
2405 XATTR_NAME_LINK, 0, th);
2409 rc = dt_trans_start_local(env, dev, th);
2413 dt_write_lock(env, obj, 0);
2414 if (unlikely(lfsck_is_dead_obj(obj)))
2415 GOTO(unlock, rc = 0);
2417 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2418 GOTO(unlock, rc = 1);
2420 /* The old ".." name entry maybe not exist. */
2421 dt_delete(env, obj, (const struct dt_key *)dotdot, th);
2423 rc = dt_insert(env, obj, (const struct dt_rec *)rec,
2424 (const struct dt_key *)dotdot, th);
2428 rc = lfsck_links_write(env, obj, &ldata, th);
2430 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2433 dt_write_unlock(env, obj);
2436 dt_trans_stop(env, dev, th);
2439 CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild dotdot name entry for "
2440 "the object "DFID", new parent "DFID": rc = %d\n",
2441 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)),
2445 struct lfsck_namespace *ns = com->lc_file_ram;
2447 ns->ln_flags |= LF_INCONSISTENT;
2454 * Handle orphan @obj during Double Scan Directory.
2456 * Remove the @obj's current (invalid) linkEA entries, and insert
2457 * it in the directory .lustre/lost+found/MDTxxxx/ with the name:
2458 * ${FID}-${PFID}-D-${conflict_version}
2460 * The caller should take the ldlm lock before the calling.
2462 * \param[in] env pointer to the thread context
2463 * \param[in] com pointer to the lfsck component
2464 * \param[in] obj pointer to the orphan object to be handled
2465 * \param[in] pfid the new fid for the object's ".." name entry
2466 * \param[in,out] lh ldlm lock handler for the given @obj
2467 * \param[out] type to tell the caller what the inconsistency is
2469 * \retval positive number for repaired cases
2470 * \retval 0 if nothing to be repaired
2471 * \retval negative error number on failure
2474 lfsck_namespace_dsd_orphan(const struct lu_env *env,
2475 struct lfsck_component *com,
2476 struct dt_object *obj,
2477 const struct lu_fid *pfid,
2478 struct lustre_handle *lh,
2479 enum lfsck_namespace_inconsistency_type *type)
2481 struct lfsck_thread_info *info = lfsck_env_info(env);
2482 struct lfsck_namespace *ns = com->lc_file_ram;
2486 /* Remove the unrecognized linkEA. */
2487 rc = lfsck_namespace_links_remove(env, com, obj);
2488 lfsck_ibits_unlock(lh, LCK_EX);
2489 if (rc < 0 && rc != -ENODATA)
2492 *type = LNIT_MUL_REF;
2494 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
2495 * ever tried to verify some remote MDT-object that resides on this
2496 * MDT, but this MDT failed to respond such request. So means there
2497 * may be some remote name entry on other MDT that references this
2498 * object with another name, so we cannot know whether this linkEA
2499 * is valid or not. So keep it there and maybe resolved when next
2501 if (ns->ln_flags & LF_INCOMPLETE)
2504 /* The unique linkEA is invalid, even if the ".." name entry may be
2505 * valid, we still cannot know via which name entry this directory
2506 * will be referenced. Then handle it as pure orphan. */
2507 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2508 "-"DFID, PFID(pfid));
2509 rc = lfsck_namespace_insert_orphan(env, com, obj,
2510 info->lti_tmpbuf, "D", NULL);
2516 * Double Scan Directory object for single linkEA entry case.
2518 * The given @child has unique linkEA entry. If the linkEA entry is valid,
2519 * then check whether the name is in the namespace or not, if not, add the
2520 * missing name entry back to namespace. If the linkEA entry is invalid,
2521 * then remove it and insert the @child in the .lustre/lost+found/MDTxxxx/
2524 * \param[in] env pointer to the thread context
2525 * \param[in] com pointer to the lfsck component
2526 * \param[in] child pointer to the directory to be double scanned
2527 * \param[in] pfid the FID corresponding to the ".." entry
2528 * \param[in] ldata pointer to the linkEA data for the given @child
2529 * \param[in,out] lh ldlm lock handler for the given @child
2530 * \param[out] type to tell the caller what the inconsistency is
2531 * \param[in] retry if found inconsistency, but the caller does not hold
2532 * ldlm lock on the @child, then set @retry as true
2533 * \param[in] unknown set if does not know how to repair the inconsistency
2535 * \retval positive number for repaired cases
2536 * \retval 0 if nothing to be repaired
2537 * \retval negative error number on failure
2540 lfsck_namespace_dsd_single(const struct lu_env *env,
2541 struct lfsck_component *com,
2542 struct dt_object *child,
2543 const struct lu_fid *pfid,
2544 struct linkea_data *ldata,
2545 struct lustre_handle *lh,
2546 enum lfsck_namespace_inconsistency_type *type,
2547 bool *retry, bool *unknown)
2549 struct lfsck_thread_info *info = lfsck_env_info(env);
2550 struct lu_name *cname = &info->lti_name;
2551 const struct lu_fid *cfid = lfsck_dto2fid(child);
2553 struct lfsck_namespace *ns = com->lc_file_ram;
2554 struct lfsck_instance *lfsck = com->lc_lfsck;
2555 struct dt_object *parent = NULL;
2556 struct lmv_mds_md_v1 *lmv;
2560 rc = lfsck_namespace_unpack_linkea_entry(ldata, cname, &tfid,
2562 sizeof(info->lti_key));
2563 /* The unique linkEA entry with bad parent will be handled as orphan. */
2565 if (!lustre_handle_is_used(lh) && retry != NULL)
2568 rc = lfsck_namespace_dsd_orphan(env, com, child,
2574 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
2576 GOTO(out, rc = PTR_ERR(parent));
2578 /* We trust the unique linkEA entry in spite of whether it matches the
2579 * ".." name entry or not. Because even if the linkEA entry is wrong
2580 * and the ".." name entry is right, we still cannot know via which
2581 * name entry the child will be referenced, since all known entries
2582 * have been verified during the first-stage scanning. */
2583 if (!dt_object_exists(parent)) {
2584 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2585 * has ever tried to verify some remote MDT-object that resides
2586 * on this MDT, but this MDT failed to respond such request. So
2587 * means there may be some remote name entry on other MDT that
2588 * references this object with another name, so we cannot know
2589 * whether this linkEA is valid or not. So keep it there and
2590 * maybe resolved when next LFSCK run. */
2591 if (ns->ln_flags & LF_INCOMPLETE)
2594 if (!lustre_handle_is_used(lh) && retry != NULL) {
2600 lfsck_ibits_unlock(lh, LCK_EX);
2603 lmv = &info->lti_lmv;
2604 rc = lfsck_read_stripe_lmv(env, lfsck, child, lmv);
2605 if (rc != 0 && rc != -ENODATA)
2608 if (rc == -ENODATA || lmv->lmv_magic != LMV_MAGIC_STRIPE) {
2610 } else if (lfsck_shard_name_to_index(env,
2611 cname->ln_name, cname->ln_namelen,
2612 S_IFDIR, cfid) < 0) {
2613 /* It is an invalid name entry, we
2614 * cannot trust the parent also. */
2615 rc = lfsck_namespace_shrink_linkea(env, com, child,
2616 ldata, cname, &tfid, true);
2620 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2621 "-"DFID, PFID(pfid));
2622 rc = lfsck_namespace_insert_orphan(env, com, child,
2623 info->lti_tmpbuf, "S", NULL);
2628 /* Create the lost parent as an orphan. */
2629 rc = lfsck_namespace_create_orphan_dir(env, com, parent, lmv);
2631 /* Add the missing name entry to the parent. */
2632 rc = lfsck_namespace_insert_normal(env, com, parent,
2634 if (unlikely(rc == -EEXIST)) {
2635 /* Unfortunately, someone reused the name
2636 * under the parent by race. So we have
2637 * to remove the linkEA entry from
2638 * current child object. It means that the
2639 * LFSCK cannot recover the system
2640 * totally back to its original status,
2641 * but it is necessary to make the
2642 * current system to be consistent. */
2643 rc = lfsck_namespace_shrink_linkea(env,
2645 cname, &tfid, true);
2647 snprintf(info->lti_tmpbuf,
2648 sizeof(info->lti_tmpbuf),
2649 "-"DFID, PFID(pfid));
2650 rc = lfsck_namespace_insert_orphan(env,
2651 com, child, info->lti_tmpbuf,
2658 } /* !dt_object_exists(parent) */
2660 /* The unique linkEA entry with bad parent will be handled as orphan. */
2661 if (unlikely(!dt_try_as_dir(env, parent, true))) {
2662 if (!lustre_handle_is_used(lh) && retry != NULL)
2665 rc = lfsck_namespace_dsd_orphan(env, com, child,
2671 rc = dt_lookup_dir(env, parent, cname->ln_name, &tfid);
2672 if (rc == -ENOENT) {
2673 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2674 * has ever tried to verify some remote MDT-object that resides
2675 * on this MDT, but this MDT failed to respond such request. So
2676 * means there may be some remote name entry on other MDT that
2677 * references this object with another name, so we cannot know
2678 * whether this linkEA is valid or not. So keep it there and
2679 * maybe resolved when next LFSCK run. */
2680 if (ns->ln_flags & LF_INCOMPLETE)
2683 if (!lustre_handle_is_used(lh) && retry != NULL) {
2689 lfsck_ibits_unlock(lh, LCK_EX);
2690 rc = lfsck_namespace_check_name(env, lfsck, parent, child,
2698 /* It is an invalid name entry, drop it. */
2699 if (unlikely(rc > 0)) {
2700 rc = lfsck_namespace_shrink_linkea(env, com, child,
2701 ldata, cname, &tfid, true);
2703 snprintf(info->lti_tmpbuf,
2704 sizeof(info->lti_tmpbuf),
2705 "-"DFID, PFID(pfid));
2706 rc = lfsck_namespace_insert_orphan(env, com,
2707 child, info->lti_tmpbuf, "D", NULL);
2713 /* Add the missing name entry back to the namespace. */
2714 rc = lfsck_namespace_insert_normal(env, com, parent, child,
2716 if (unlikely(rc == -ESTALE))
2717 /* It may happen when the remote object has been
2718 * removed, but the local MDT is not aware of that. */
2721 if (unlikely(rc == -EEXIST)) {
2722 /* Unfortunately, someone reused the name under the
2723 * parent by race. So we have to remove the linkEA
2724 * entry from current child object. It means that the
2725 * LFSCK cannot recover the system totally back to
2726 * its original status, but it is necessary to make
2727 * the current system to be consistent.
2729 * It also may be because of the LFSCK found some
2730 * internal status of create operation. Under such
2731 * case, nothing to be done. */
2732 rc = lfsck_namespace_shrink_linkea_cond(env, com,
2733 parent, child, ldata, cname, &tfid);
2735 snprintf(info->lti_tmpbuf,
2736 sizeof(info->lti_tmpbuf),
2737 "-"DFID, PFID(pfid));
2738 rc = lfsck_namespace_insert_orphan(env, com,
2739 child, info->lti_tmpbuf, "D", NULL);
2744 } /* rc == -ENOENT */
2749 if (!lu_fid_eq(&tfid, cfid)) {
2750 if (!lustre_handle_is_used(lh) && retry != NULL) {
2756 lfsck_ibits_unlock(lh, LCK_EX);
2757 /* The name entry references another MDT-object that
2758 * may be created by the LFSCK for repairing dangling
2759 * name entry. Try to replace it. */
2760 rc = lfsck_namespace_replace_cond(env, com, parent, child,
2763 rc = lfsck_namespace_dsd_orphan(env, com, child,
2769 /* Zero FID may because the remote directroy object has invalid linkEA,
2770 * or lost linkEA. Under such case, the LFSCK on this MDT does not know
2771 * how to repair the inconsistency, but the namespace LFSCK on the MDT
2772 * where its name entry resides may has more information (name, FID) to
2773 * repair such inconsistency. So here, keep the inconsistency to avoid
2774 * some imporper repairing. */
2775 if (fid_is_zero(pfid)) {
2782 /* The ".." name entry is wrong, update it. */
2783 if (!lu_fid_eq(pfid, lfsck_dto2fid(parent))) {
2784 if (!lustre_handle_is_used(lh) && retry != NULL) {
2790 *type = LNIT_UNMATCHED_PAIRS;
2791 rc = lfsck_namespace_repair_unmatched_pairs(env, com, child,
2792 lfsck_dto2fid(parent), cname);
2798 if (parent != NULL && !IS_ERR(parent))
2799 lfsck_object_put(env, parent);
2805 * Double Scan Directory object for multiple linkEA entries case.
2807 * The given @child has multiple linkEA entries. There is at most one linkEA
2808 * entry will be valid, all the others will be removed. Firstly, the function
2809 * will try to find out the linkEA entry for which the name entry exists under
2810 * the given parent (@pfid). If there is no linkEA entry that matches the given
2811 * ".." name entry, then tries to find out the first linkEA entry that both the
2812 * parent and the name entry exist to rebuild a new ".." name entry.
2814 * \param[in] env pointer to the thread context
2815 * \param[in] com pointer to the lfsck component
2816 * \param[in] child pointer to the directory to be double scanned
2817 * \param[in] pfid the FID corresponding to the ".." entry
2818 * \param[in] ldata pointer to the linkEA data for the given @child
2819 * \param[in,out] lh ldlm lock handler for the given @child
2820 * \param[out] type to tell the caller what the inconsistency is
2821 * \param[in] lpf true if the ".." entry is under lost+found/MDTxxxx/
2822 * \param[in] unknown set if does not know how to repair the inconsistency
2824 * \retval positive number for repaired cases
2825 * \retval 0 if nothing to be repaired
2826 * \retval negative error number on failure
2829 lfsck_namespace_dsd_multiple(const struct lu_env *env,
2830 struct lfsck_component *com,
2831 struct dt_object *child,
2832 const struct lu_fid *pfid,
2833 struct linkea_data *ldata,
2834 struct lustre_handle *lh,
2835 enum lfsck_namespace_inconsistency_type *type,
2836 bool lpf, bool *unknown)
2838 struct lfsck_thread_info *info = lfsck_env_info(env);
2839 struct lu_name *cname = &info->lti_name;
2840 const struct lu_fid *cfid = lfsck_dto2fid(child);
2841 struct lu_fid *pfid2 = &info->lti_fid3;
2843 struct lfsck_namespace *ns = com->lc_file_ram;
2844 struct lfsck_instance *lfsck = com->lc_lfsck;
2845 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
2846 struct dt_object *parent = NULL;
2847 struct linkea_data ldata_new = { NULL };
2848 int dirent_count = 0;
2854 while (ldata->ld_lee != NULL) {
2855 rc = lfsck_namespace_unpack_linkea_entry(ldata, cname, &tfid,
2857 sizeof(info->lti_key));
2858 /* Drop invalid linkEA entry. */
2860 lfsck_linkea_del_buf(ldata, cname);
2864 /* Drop repeated linkEA entries. */
2865 lfsck_namespace_filter_linkea_entry(ldata, cname, &tfid, true);
2867 /* If current dotdot is the .lustre/lost+found/MDTxxxx/,
2868 * then it is possible that: the directry object has ever
2869 * been lost, but its name entry was there. In the former
2870 * LFSCK run, during the first-stage scanning, the LFSCK
2871 * found the dangling name entry, but it did not recreate
2872 * the lost object, and when moved to the second-stage
2873 * scanning, some children objects of the lost directory
2874 * object were found, then the LFSCK recreated such lost
2875 * directory object as an orphan.
2877 * When the LFSCK runs again, if the dangling name is still
2878 * there, the LFSCK should move the orphan directory object
2879 * back to the normal namespace. */
2880 if (!lpf && !fid_is_zero(pfid) &&
2881 !lu_fid_eq(pfid, &tfid) && once) {
2882 linkea_next_entry(ldata);
2886 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
2887 if (IS_ERR(parent)) {
2888 rc = PTR_ERR(parent);
2889 /* if @pfid doesn't have a valid OI mapping, it will
2890 * trigger OI scrub, and -ENONET is is returned if it's
2891 * remote, -EINPROGRESS if local.
2893 if ((rc == -ENOENT || rc == -EINPROGRESS) &&
2894 ldata->ld_leh->leh_reccount > 1) {
2895 lfsck_linkea_del_buf(ldata, cname);
2902 if (!dt_object_exists(parent)) {
2903 lfsck_object_put(env, parent);
2904 if (ldata->ld_leh->leh_reccount > 1) {
2905 /* If it is NOT the last linkEA entry, then
2906 * there is still other chance to make the
2907 * child to be visible via other parent, then
2908 * remove this linkEA entry. */
2909 lfsck_linkea_del_buf(ldata, cname);
2916 /* The linkEA entry with bad parent will be removed. */
2917 if (unlikely(!dt_try_as_dir(env, parent, true))) {
2918 lfsck_object_put(env, parent);
2919 lfsck_linkea_del_buf(ldata, cname);
2923 rc = dt_lookup_dir(env, parent, cname->ln_name, &tfid);
2924 *pfid2 = *lfsck_dto2fid(parent);
2925 if (rc == -ENOENT) {
2926 lfsck_object_put(env, parent);
2927 linkea_next_entry(ldata);
2932 lfsck_object_put(env, parent);
2937 if (lu_fid_eq(&tfid, cfid)) {
2938 lfsck_object_put(env, parent);
2939 /* If the parent (that is declared via linkEA entry)
2940 * directory contains the specified child, but such
2941 * parent does not match the dotdot name entry, then
2942 * trust the linkEA. */
2943 if (!fid_is_zero(pfid) && !lu_fid_eq(pfid, pfid2)) {
2944 *type = LNIT_UNMATCHED_PAIRS;
2945 rc = lfsck_namespace_repair_unmatched_pairs(env,
2946 com, child, pfid2, cname);
2952 /* It is the most common case that we find the
2953 * name entry corresponding to the linkEA entry
2954 * that matches the ".." name entry. */
2955 rc = linkea_links_new(&ldata_new, &info->lti_big_buf,
2960 rc = lfsck_namespace_rebuild_linkea(env, com, child,
2965 lfsck_linkea_del_buf(ldata, cname);
2966 linkea_first_entry(ldata);
2967 /* There may be some invalid dangling name entries under
2968 * other parent directories, remove all of them. */
2969 while (ldata->ld_lee != NULL) {
2970 rc = lfsck_namespace_unpack_linkea_entry(ldata,
2971 cname, &tfid, info->lti_key,
2972 sizeof(info->lti_key));
2976 parent = lfsck_object_find_bottom(env, lfsck,
2978 if (IS_ERR(parent)) {
2979 rc = PTR_ERR(parent);
2980 if (rc != -ENOENT &&
2981 bk->lb_param & LPF_FAILOUT)
2987 if (!dt_object_exists(parent)) {
2988 lfsck_object_put(env, parent);
2992 rc = lfsck_namespace_repair_dirent(env, com,
2993 parent, child, cname->ln_name,
2994 cname->ln_name, S_IFDIR, false, true);
2995 lfsck_object_put(env, parent);
2997 if (bk->lb_param & LPF_FAILOUT)
3006 lfsck_linkea_del_buf(ldata, cname);
3009 ns->ln_dirent_repaired += dirent_count;
3012 } /* lu_fid_eq(&tfid, lfsck_dto2fid(child)) */
3014 lfsck_ibits_unlock(lh, LCK_EX);
3015 /* The name entry references another MDT-object that may be
3016 * created by the LFSCK for repairing dangling name entry.
3017 * Try to replace it. */
3018 rc = lfsck_namespace_replace_cond(env, com, parent, child,
3020 lfsck_object_put(env, parent);
3027 lfsck_linkea_del_buf(ldata, cname);
3028 } /* while (ldata->ld_lee != NULL) */
3030 /* If there is still linkEA overflow, return. */
3031 if (unlikely(ldata->ld_leh->leh_overflow_time))
3034 linkea_first_entry(ldata);
3035 if (ldata->ld_leh->leh_reccount == 1) {
3036 rc = lfsck_namespace_dsd_single(env, com, child, pfid, ldata,
3037 lh, type, NULL, unknown);
3042 /* All linkEA entries are invalid and removed, then handle the @child
3044 if (ldata->ld_leh->leh_reccount == 0) {
3045 rc = lfsck_namespace_dsd_orphan(env, com, child, pfid, lh,
3051 /* If the dangling name entry for the orphan directory object has
3052 * been remvoed, then just check whether the directory object is
3053 * still under the .lustre/lost+found/MDTxxxx/ or not. */
3059 /* There is no linkEA entry that matches the ".." name entry. Find
3060 * the first linkEA entry that both parent and name entry exist to
3061 * rebuild a new ".." name entry. */
3071 * Repair the object's nlink attribute.
3073 * If all the known name entries have been verified, then the object's hard
3074 * link attribute should match the object's linkEA entries count unless the
3075 * object's has too many hard link to be recorded in the linkEA. Such cases
3076 * should have been marked in the LFSCK trace file. Otherwise, trust the
3077 * linkEA to update the object's nlink attribute.
3079 * \param[in] env pointer to the thread context
3080 * \param[in] com pointer to the lfsck component
3081 * \param[in] obj pointer to the dt_object to be handled
3082 * \param[in,out] la pointer to buffer to object's attribute before
3083 * and after the repairing
3085 * \retval positive number for repaired cases
3086 * \retval 0 if nothing to be repaired
3087 * \retval negative error number on failure
3089 static int lfsck_namespace_repair_nlink(const struct lu_env *env,
3090 struct lfsck_component *com,
3091 struct dt_object *obj,
3094 struct lfsck_namespace *ns = com->lc_file_ram;
3095 struct lfsck_instance *lfsck = com->lc_lfsck;
3096 struct dt_device *dev = lfsck_obj2dev(obj);
3097 const struct lu_fid *cfid = lfsck_dto2fid(obj);
3098 struct thandle *th = NULL;
3099 struct linkea_data ldata = { NULL };
3100 struct lustre_handle lh = { 0 };
3101 __u32 old = la->la_nlink;
3105 LASSERT(!dt_object_remote(obj));
3107 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
3108 MDS_INODELOCK_UPDATE, LCK_PW);
3112 th = lfsck_trans_create(env, dev, lfsck);
3114 GOTO(log, rc = PTR_ERR(th));
3116 la->la_valid = LA_NLINK;
3117 rc = dt_declare_attr_set(env, obj, la, th);
3121 rc = dt_trans_start_local(env, dev, th);
3125 dt_write_lock(env, obj, 0);
3126 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
3127 * ever tried to verify some remote MDT-object that resides on this
3128 * MDT, but this MDT failed to respond such request. So means there
3129 * may be some remote name entry on other MDT that references this
3130 * object with another name, so we cannot know whether this linkEA
3131 * is valid or not. So keep it there and maybe resolved when next
3133 if (ns->ln_flags & LF_INCOMPLETE)
3134 GOTO(unlock, rc = 0);
3136 rc = dt_attr_get(env, obj, la);
3138 GOTO(unlock, rc = (rc == -ENOENT ? 0 : rc));
3140 rc = lfsck_links_read2_with_rec(env, obj, &ldata);
3142 GOTO(unlock, rc = (rc == -ENODATA ? 0 : rc));
3144 /* XXX: Currently, we only update the nlink attribute if the known
3145 * linkEA entries is larger than the nlink attribute. That is
3147 if (la->la_nlink >= ldata.ld_leh->leh_reccount ||
3148 unlikely(la->la_nlink == 0 ||
3149 ldata.ld_leh->leh_overflow_time))
3150 GOTO(unlock, rc = 0);
3152 la->la_nlink = ldata.ld_leh->leh_reccount;
3153 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
3154 GOTO(unlock, rc = 1);
3156 rc = dt_attr_set(env, obj, la, th);
3158 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
3161 dt_write_unlock(env, obj);
3164 dt_trans_stop(env, dev, th);
3167 lfsck_ibits_unlock(&lh, LCK_PW);
3168 CDEBUG(D_LFSCK, "%s: namespace LFSCK repaired the object "DFID"'s "
3169 "nlink count from %u to %u: rc = %d\n",
3170 lfsck_lfsck2name(lfsck), PFID(cfid), old, la->la_nlink, rc);
3173 ns->ln_flags |= LF_INCONSISTENT;
3179 * Double scan the directory object for namespace LFSCK.
3181 * This function will verify the <parent, child> pairs in the namespace tree:
3182 * the parent references the child via some name entry that should be in the
3183 * child's linkEA entry, the child should back references the parent via its
3186 * The LFSCK will scan every linkEA entry in turn until find out the first
3187 * matched pairs. If found, then all other linkEA entries will be dropped.
3188 * If all the linkEA entries cannot match the ".." name entry, then there
3189 * are serveral possible cases:
3191 * 1) If there is only one linkEA entry, then trust it as long as the PFID
3192 * in the linkEA entry is valid.
3194 * 2) If there are multiple linkEA entries, then try to find the linkEA
3195 * that matches the ".." name entry. If found, then all other entries
3196 * are invalid; otherwise, it is quite possible that the ".." name entry
3197 * is corrupted. Under such case, the LFSCK will rebuild the ".." name
3198 * entry according to the first valid linkEA entry (both the parent and
3199 * the name entry should exist).
3201 * 3) If the directory object has no (valid) linkEA entry, then the
3202 * directory object will be handled as pure orphan and inserted
3203 * in the .lustre/lost+found/MDTxxxx/ with the name:
3204 * ${self_FID}-${PFID}-D-${conflict_version}
3206 * \param[in] env pointer to the thread context
3207 * \param[in] com pointer to the lfsck component
3208 * \param[in] child pointer to the directory object to be handled
3209 * \param[in] flags to indicate the specical checking on the @child
3211 * \retval positive number for repaired cases
3212 * \retval 0 if nothing to be repaired
3213 * \retval negative error number on failure
3215 static int lfsck_namespace_double_scan_dir(const struct lu_env *env,
3216 struct lfsck_component *com,
3217 struct dt_object *child, __u8 flags)
3219 struct lfsck_thread_info *info = lfsck_env_info(env);
3220 const struct lu_fid *cfid = lfsck_dto2fid(child);
3221 struct lu_fid *pfid = &info->lti_fid2;
3222 struct lfsck_namespace *ns = com->lc_file_ram;
3223 struct lfsck_instance *lfsck = com->lc_lfsck;
3224 struct lustre_handle lh = { 0 };
3225 struct linkea_data ldata = { NULL };
3226 bool unknown = false;
3229 enum lfsck_namespace_inconsistency_type type = LNIT_BAD_LINKEA;
3233 LASSERT(!dt_object_remote(child));
3235 if (flags & LNTF_UNCERTAIN_LMV) {
3236 if (flags & LNTF_RECHECK_NAME_HASH) {
3237 rc = lfsck_namespace_scan_shard(env, com, child);
3241 ns->ln_striped_shards_scanned++;
3243 ns->ln_striped_shards_skipped++;
3247 flags &= ~(LNTF_RECHECK_NAME_HASH | LNTF_UNCERTAIN_LMV);
3251 if (flags & (LNTF_CHECK_LINKEA | LNTF_CHECK_PARENT) &&
3252 !(lfsck->li_bookmark_ram.lb_param & LPF_ALL_TGT)) {
3254 "%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",
3255 lfsck_lfsck2name(lfsck), PFID(cfid));
3260 if (unlikely(!dt_try_as_dir(env, child, true)))
3261 GOTO(out, rc = -ENOTDIR);
3263 /* We only take ldlm lock on the @child when required. When the
3264 * logic comes here for the first time, it is always false. */
3268 rc = lfsck_ibits_lock(env, lfsck, child, &lh,
3269 MDS_INODELOCK_UPDATE |
3270 MDS_INODELOCK_XATTR, LCK_EX);
3275 dt_read_lock(env, child, 0);
3276 if (unlikely(lfsck_is_dead_obj(child))) {
3277 dt_read_unlock(env, child);
3282 rc = dt_lookup_dir(env, child, dotdot, pfid);
3284 if (rc != -ENOENT && rc != -ENODATA && rc != -EINVAL) {
3285 dt_read_unlock(env, child);
3290 if (!lustre_handle_is_used(&lh)) {
3291 dt_read_unlock(env, child);
3296 } else if (lfsck->li_lpf_obj != NULL &&
3297 lu_fid_eq(pfid, lfsck_dto2fid(lfsck->li_lpf_obj))) {
3299 } else if (unlikely(!fid_is_sane(pfid))) {
3303 rc = lfsck_links_read(env, child, &ldata);
3304 dt_read_unlock(env, child);
3306 if (rc != -ENODATA && rc != -EINVAL)
3309 if (!lustre_handle_is_used(&lh))
3312 if (rc == -EINVAL && !fid_is_zero(pfid)) {
3313 /* Remove the corrupted linkEA. */
3314 rc = lfsck_namespace_links_remove(env, com, child);
3316 /* Here, because of the crashed linkEA, we
3317 * cannot know whether there is some parent
3318 * that references the child directory via
3319 * some name entry or not. So keep it there,
3320 * when the LFSCK run next time, if there is
3321 * some parent that references this object,
3322 * then the LFSCK can rebuild the linkEA;
3323 * otherwise, this object will be handled
3324 * as orphan as above. */
3327 /* 1. If we have neither ".." nor linkEA,
3328 * then it is an orphan.
3330 * 2. If we only have the ".." name entry,
3331 * but no parent references this child
3332 * directory, then handle it as orphan. */
3333 lfsck_ibits_unlock(&lh, LCK_EX);
3334 type = LNIT_MUL_REF;
3336 /* If the LFSCK is marked as LF_INCOMPLETE,
3337 * then means some MDT has ever tried to
3338 * verify some remote MDT-object that resides
3339 * on this MDT, but this MDT failed to respond
3340 * such request. So means there may be some
3341 * remote name entry on other MDT that
3342 * references this object with another name,
3343 * so we cannot know whether this linkEA is
3344 * valid or not. So keep it there and maybe
3345 * resolved when next LFSCK run. */
3346 if (ns->ln_flags & LF_INCOMPLETE)
3349 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
3350 "-"DFID, PFID(pfid));
3351 rc = lfsck_namespace_insert_orphan(env, com, child,
3352 info->lti_tmpbuf, "D", NULL);
3358 linkea_first_entry(&ldata);
3359 /* This is the most common case: the object has unique linkEA entry. */
3360 if (ldata.ld_leh->leh_reccount == 1) {
3361 rc = lfsck_namespace_dsd_single(env, com, child, pfid, &ldata,
3362 &lh, &type, &retry, &unknown);
3364 LASSERT(!lustre_handle_is_used(&lh));
3373 if (!lustre_handle_is_used(&lh))
3376 if (unlikely(ldata.ld_leh->leh_reccount == 0)) {
3377 rc = lfsck_namespace_dsd_orphan(env, com, child, pfid, &lh,
3383 /* When we come here, the cases usually like that:
3384 * 1) The directory object has a corrupted linkEA entry. During the
3385 * first-stage scanning, the LFSCK cannot know such corruption,
3386 * then it appends the right linkEA entry according to the found
3387 * name entry after the bad one.
3389 * 2) The directory object has a right linkEA entry. During the
3390 * first-stage scanning, the LFSCK finds some bad name entry,
3391 * but the LFSCK cannot aware that at that time, then it adds
3392 * the bad linkEA entry for further processing. */
3393 rc = lfsck_namespace_dsd_multiple(env, com, child, pfid, &ldata,
3394 &lh, &type, lpf, &unknown);
3399 lfsck_ibits_unlock(&lh, LCK_EX);
3402 case LNIT_BAD_LINKEA:
3403 ns->ln_linkea_repaired++;
3405 case LNIT_UNMATCHED_PAIRS:
3406 ns->ln_unmatched_pairs_repaired++;
3409 ns->ln_mul_ref_repaired++;
3417 ns->ln_unknown_inconsistency++;
3423 lfsck_namespace_linkea_stale_overflow(struct linkea_data *ldata,
3424 struct lfsck_namespace *ns)
3426 /* Both the leh_overflow_time and ln_time_latest_reset are
3427 * local time based, so need NOT to care about clock drift
3428 * among the servers. */
3429 return ldata->ld_leh->leh_overflow_time &&
3430 ldata->ld_leh->leh_overflow_time < ns->ln_time_latest_reset;
3434 * Clear the object's linkEA overflow timestamp.
3436 * If the MDT-object has too many hard links as to the linkEA cannot hold
3437 * all of them, then overflow timestamp will be set in the linkEA header.
3438 * If some hard links are removed after that, then it is possible to hold
3439 * other missed linkEA entries. If the namespace LFSCK have added all the
3440 * related linkEA entries, then it will remove the overflow timestamp.
3442 * \param[in] env pointer to the thread context
3443 * \param[in] com pointer to the lfsck component
3444 * \param[in] ldata pointer to the linkEA data for the given @obj
3445 * \param[in] obj pointer to the dt_object to be handled
3447 * \retval positive number for repaired cases
3448 * \retval 0 if nothing to be repaired
3449 * \retval negative error number on failure
3451 static int lfsck_namespace_linkea_clear_overflow(const struct lu_env *env,
3452 struct lfsck_component *com,
3453 struct linkea_data *ldata,
3454 struct dt_object *obj)
3456 struct lfsck_namespace *ns = com->lc_file_ram;
3457 struct lfsck_instance *lfsck = com->lc_lfsck;
3458 struct dt_device *dev = lfsck_obj2dev(obj);
3459 struct thandle *th = NULL;
3460 struct lustre_handle lh = { 0 };
3461 struct lu_buf linkea_buf;
3465 LASSERT(!dt_object_remote(obj));
3467 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
3468 MDS_INODELOCK_UPDATE, LCK_PW);
3472 th = lfsck_trans_create(env, dev, lfsck);
3474 GOTO(log, rc = PTR_ERR(th));
3476 rc = dt_declare_xattr_set(env, obj,
3477 lfsck_buf_get_const(env, NULL, MAX_LINKEA_SIZE),
3478 XATTR_NAME_LINK, 0, th);
3482 rc = dt_trans_start_local(env, dev, th);
3486 dt_write_lock(env, obj, 0);
3487 rc = lfsck_links_read(env, obj, ldata);
3491 if (unlikely(!lfsck_namespace_linkea_stale_overflow(ldata, ns)))
3492 GOTO(unlock, rc = 0);
3494 ldata->ld_leh->leh_overflow_time = 0;
3495 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
3496 GOTO(unlock, rc = 1);
3498 lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
3499 ldata->ld_leh->leh_len);
3500 rc = dt_xattr_set(env, obj, &linkea_buf, XATTR_NAME_LINK, 0, th);
3501 if (unlikely(rc == -ENOSPC))
3509 dt_write_unlock(env, obj);
3512 dt_trans_stop(env, dev, th);
3515 lfsck_ibits_unlock(&lh, LCK_PW);
3516 CDEBUG(D_LFSCK, "%s: clear linkea overflow timestamp for the object "
3518 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
3524 * Verify the object's agent entry.
3526 * If the object claims to have agent entry but the linkEA does not contain
3527 * remote parent, then remove the agent entry. Otherwise, if the object has
3528 * no agent entry but its linkEA contains remote parent, then will generate
3529 * agent entry for it.
3531 * \param[in] env pointer to the thread context
3532 * \param[in] com pointer to the lfsck component
3533 * \param[in] obj pointer to the dt_object to be handled
3535 * \retval positive number for repaired cases
3536 * \retval 0 if nothing to be repaired
3537 * \retval negative error number on failure
3539 static int lfsck_namespace_check_agent_entry(const struct lu_env *env,
3540 struct lfsck_component *com,
3541 struct dt_object *obj)
3543 struct linkea_data ldata = { NULL };
3544 struct lfsck_thread_info *info = lfsck_env_info(env);
3545 struct lfsck_namespace *ns = com->lc_file_ram;
3546 struct lfsck_instance *lfsck = com->lc_lfsck;
3547 struct lu_fid *pfid = &info->lti_fid2;
3548 struct lu_name *cname = &info->lti_name;
3549 struct lu_seq_range *range = &info->lti_range;
3550 struct seq_server_site *ss = lfsck_dev_site(lfsck);
3551 __u32 idx = lfsck_dev_idx(lfsck);
3553 bool remote = false;
3556 if (!(lfsck->li_bookmark_ram.lb_param & LPF_ALL_TGT))
3559 rc = lfsck_links_read_with_rec(env, obj, &ldata);
3560 if (rc == -ENOENT || rc == -ENODATA)
3563 if (rc && rc != -EINVAL)
3566 /* We check the agent entry again after verifying the linkEA
3567 * successfully. So invalid linkEA should be dryrun mode. */
3568 if (rc == -EINVAL || unlikely(!ldata.ld_leh->leh_reccount))
3571 linkea_first_entry(&ldata);
3572 while (ldata.ld_lee != NULL && !remote) {
3573 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
3575 if (!linkea_entry_is_valid(&ldata, cname, pfid))
3578 fld_range_set_mdt(range);
3579 rc = fld_server_lookup(env, ss->ss_server_fld,
3580 fid_seq(pfid), range);
3582 GOTO(out, rc = (rc == -ENOENT ? 0 : rc));
3584 if (range->lsr_index != idx)
3587 linkea_next_entry(&ldata);
3590 if ((lu_object_has_agent_entry(&obj->do_lu) && !remote) ||
3591 (!lu_object_has_agent_entry(&obj->do_lu) && remote)) {
3592 struct dt_device *dev = lfsck_obj2dev(obj);
3593 struct linkea_data ldata2 = { NULL };
3594 struct lustre_handle lh = { 0 };
3595 struct lu_buf linkea_buf;
3596 struct thandle *handle;
3598 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
3601 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
3602 MDS_INODELOCK_UPDATE |
3603 MDS_INODELOCK_XATTR, LCK_EX);
3607 handle = lfsck_trans_create(env, dev, lfsck);
3609 GOTO(unlock, rc = PTR_ERR(handle));
3611 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
3612 ldata.ld_leh->leh_len);
3613 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
3614 XATTR_NAME_LINK, LU_XATTR_REPLACE, handle);
3618 rc = dt_trans_start_local(env, dev, handle);
3622 dt_write_lock(env, obj, 0);
3623 rc = lfsck_links_read2_with_rec(env, obj, &ldata2);
3625 if (rc == -ENOENT || rc == -ENODATA)
3630 /* If someone changed linkEA by race, then the agent
3631 * entry will be updated by lower layer automatically. */
3632 if (ldata.ld_leh->leh_len != ldata2.ld_leh->leh_len ||
3633 memcmp(ldata.ld_buf->lb_buf, ldata2.ld_buf->lb_buf,
3634 ldata.ld_leh->leh_len) != 0)
3635 GOTO(unlock2, rc = 0);
3637 rc = dt_xattr_set(env, obj, &linkea_buf, XATTR_NAME_LINK,
3638 LU_XATTR_REPLACE, handle);
3645 dt_write_unlock(env, obj);
3647 dt_trans_stop(env, dev, handle);
3649 lfsck_ibits_unlock(&lh, LCK_EX);
3656 ns->ln_agent_entries_repaired++;
3658 CDEBUG(D_LFSCK, "%s: repair agent entry for "DFID": rc = %d\n",
3659 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
3664 * Double scan the MDT-object for namespace LFSCK.
3666 * If the MDT-object contains invalid or repeated linkEA entries, then drop
3667 * those entries from the linkEA; if the linkEA becomes empty or the object
3668 * has no linkEA, then it is an orphan and will be added into the directory
3669 * .lustre/lost+found/MDTxxxx/; if the remote parent is lost, then recreate
3670 * the remote parent; if the name entry corresponding to some linkEA entry
3671 * is lost, then add the name entry back to the namespace.
3673 * \param[in] env pointer to the thread context
3674 * \param[in] com pointer to the lfsck component
3675 * \param[in] child pointer to the dt_object to be handled
3676 * \param[in] flags some hints to indicate how the @child should be handled
3678 * \retval positive number for repaired cases
3679 * \retval 0 if nothing to be repaired
3680 * \retval negative error number on failure
3682 static int lfsck_namespace_double_scan_one(const struct lu_env *env,
3683 struct lfsck_component *com,
3684 struct dt_object *child, __u8 flags)
3686 struct lfsck_thread_info *info = lfsck_env_info(env);
3687 struct lu_attr *la = &info->lti_la;
3688 struct lu_name *cname = &info->lti_name;
3689 struct lu_fid *pfid = &info->lti_fid;
3690 struct lu_fid *cfid = &info->lti_fid2;
3691 struct lfsck_instance *lfsck = com->lc_lfsck;
3692 struct lfsck_namespace *ns = com->lc_file_ram;
3693 struct dt_object *parent = NULL;
3694 struct linkea_data ldata = { NULL };
3695 bool repaired = false;
3700 dt_read_lock(env, child, 0);
3701 if (unlikely(lfsck_is_dead_obj(child))) {
3702 dt_read_unlock(env, child);
3707 if (S_ISDIR(lfsck_object_type(child))) {
3708 dt_read_unlock(env, child);
3709 rc = lfsck_namespace_double_scan_dir(env, com, child, flags);
3710 if (!rc && flags & LNTF_CHECK_AGENT_ENTRY)
3711 rc = lfsck_namespace_check_agent_entry(env, com, child);
3716 rc = lfsck_links_read(env, child, &ldata);
3717 dt_read_unlock(env, child);
3719 if (rc == -EINVAL) {
3720 struct lustre_handle lh = { 0 };
3722 rc = lfsck_ibits_lock(env, com->lc_lfsck, child, &lh,
3723 MDS_INODELOCK_UPDATE |
3724 MDS_INODELOCK_XATTR, LCK_EX);
3726 rc = lfsck_namespace_links_remove(env, com, child);
3727 lfsck_ibits_unlock(&lh, LCK_EX);
3736 if (!(ns->ln_flags & LF_INCOMPLETE) &&
3737 unlikely(lfsck_namespace_linkea_stale_overflow(&ldata, ns))) {
3738 rc = lfsck_namespace_linkea_clear_overflow(env, com, &ldata,
3744 ns->ln_linkea_overflow_cleared++;
3747 linkea_first_entry(&ldata);
3748 while (ldata.ld_lee != NULL) {
3749 rc = lfsck_namespace_unpack_linkea_entry(&ldata, cname, pfid,
3751 sizeof(info->lti_key));
3752 /* Invalid PFID in the linkEA entry. */
3754 rc = lfsck_namespace_shrink_linkea(env, com, child,
3755 &ldata, cname, pfid, true);
3765 rc = lfsck_namespace_filter_linkea_entry(&ldata, cname, pfid,
3767 /* Found repeated linkEA entries */
3769 rc = lfsck_namespace_shrink_linkea(env, com, child,
3770 &ldata, cname, pfid, false);
3782 parent = lfsck_object_find_bottom(env, lfsck, pfid);
3783 if (IS_ERR(parent)) {
3784 rc = PTR_ERR(parent);
3785 /* if @pfid doesn't have a valid OI mapping, it will
3786 * trigger OI scrub, and -ENONET is is returned if it's
3787 * remote, -EINPROGRESS if local.
3789 if ((rc == -ENOENT || rc == -EINPROGRESS) &&
3790 ldata.ld_leh->leh_reccount > 1)
3791 rc = lfsck_namespace_shrink_linkea(env, com,
3792 child, &ldata, cname, pfid, true);
3796 if (!dt_object_exists(parent)) {
3799 if (ldata.ld_leh->leh_reccount > 1) {
3800 /* If it is NOT the last linkEA entry, then
3801 * there is still other chance to make the
3802 * child to be visible via other parent, then
3803 * remove this linkEA entry. */
3804 rc = lfsck_namespace_shrink_linkea(env, com,
3805 child, &ldata, cname, pfid, true);
3807 /* If the LFSCK is marked as LF_INCOMPLETE,
3808 * then means some MDT has ever tried to
3809 * verify some remote MDT-object that resides
3810 * on this MDT, but this MDT failed to respond
3811 * such request. So means there may be some
3812 * remote name entry on other MDT that
3813 * references this object with another name,
3814 * so we cannot know whether this linkEA is
3815 * valid or not. So keep it there and maybe
3816 * resolved when next LFSCK run. */
3817 if (ns->ln_flags & LF_INCOMPLETE) {
3818 lfsck_object_put(env, parent);
3823 /* Create the lost parent as an orphan. */
3824 rc = lfsck_namespace_create_orphan_dir(env, com,
3827 lfsck_object_put(env, parent);
3835 /* Add the missing name entry to the parent. */
3836 rc = lfsck_namespace_insert_normal(env, com,
3837 parent, child, cname);
3838 if (unlikely(rc == -EEXIST))
3839 /* Unfortunately, someone reused the
3840 * name under the parent by race. So we
3841 * have to remove the linkEA entry from
3842 * current child object. It means that
3843 * the LFSCK cannot recover the system
3844 * totally back to its original status,
3845 * but it is necessary to make the
3846 * current system to be consistent. */
3847 rc = lfsck_namespace_shrink_linkea(env,
3851 linkea_next_entry(&ldata);
3854 lfsck_object_put(env, parent);
3862 } /* !dt_object_exists(parent) */
3864 /* The linkEA entry with bad parent will be removed. */
3865 if (unlikely(!dt_try_as_dir(env, parent, true))) {
3866 lfsck_object_put(env, parent);
3867 rc = lfsck_namespace_shrink_linkea(env, com, child,
3868 &ldata, cname, pfid, true);
3878 rc = dt_lookup_dir(env, parent, cname->ln_name, cfid);
3879 if (rc != 0 && rc != -ENOENT) {
3880 lfsck_object_put(env, parent);
3886 if (lu_fid_eq(cfid, lfsck_dto2fid(child))) {
3887 /* It is the most common case that we
3888 * find the name entry corresponding
3889 * to the linkEA entry. */
3890 lfsck_object_put(env, parent);
3891 linkea_next_entry(&ldata);
3893 /* The name entry references another
3894 * MDT-object that may be created by
3895 * the LFSCK for repairing dangling
3896 * name entry. Try to replace it. */
3897 rc = lfsck_namespace_replace_cond(env, com,
3898 parent, child, cfid, cname);
3899 lfsck_object_put(env, parent);
3905 linkea_next_entry(&ldata);
3907 rc = lfsck_namespace_shrink_linkea(env,
3921 /* The following handles -ENOENT case */
3923 rc = dt_attr_get(env, child, la);
3927 /* If there is no name entry in the parent dir and the object
3928 * link count is fewer than the linkea entries count, then the
3929 * linkea entry should be removed. */
3930 if (ldata.ld_leh->leh_reccount > la->la_nlink) {
3931 rc = lfsck_namespace_shrink_linkea_cond(env, com,
3932 parent, child, &ldata, cname, pfid);
3933 lfsck_object_put(env, parent);
3943 /* If the LFSCK is marked as LF_INCOMPLETE, then means some
3944 * MDT has ever tried to verify some remote MDT-object that
3945 * resides on this MDT, but this MDT failed to respond such
3946 * request. So means there may be some remote name entry on
3947 * other MDT that references this object with another name,
3948 * so we cannot know whether this linkEA is valid or not.
3949 * So keep it there and maybe resolved when next LFSCK run. */
3950 if (ns->ln_flags & LF_INCOMPLETE) {
3951 lfsck_object_put(env, parent);
3956 rc = lfsck_namespace_check_name(env, lfsck, parent, child,
3962 lfsck_object_put(env, parent);
3967 /* It is an invalid name entry, drop it. */
3968 if (unlikely(rc > 0)) {
3969 lfsck_object_put(env, parent);
3970 rc = lfsck_namespace_shrink_linkea(env, com, child,
3971 &ldata, cname, pfid, true);
3981 /* Add the missing name entry back to the namespace. */
3982 rc = lfsck_namespace_insert_normal(env, com, parent, child,
3984 if (unlikely(rc == -ESTALE))
3985 /* It may happen when the remote object has been
3986 * removed, but the local MDT is not aware of that. */
3989 if (unlikely(rc == -EEXIST))
3990 /* Unfortunately, someone reused the name under the
3991 * parent by race. So we have to remove the linkEA
3992 * entry from current child object. It means that the
3993 * LFSCK cannot recover the system totally back to
3994 * its original status, but it is necessary to make
3995 * the current system to be consistent.
3997 * It also may be because of the LFSCK found some
3998 * internal status of create operation. Under such
3999 * case, nothing to be done. */
4000 rc = lfsck_namespace_shrink_linkea_cond(env, com,
4001 parent, child, &ldata, cname, pfid);
4003 linkea_next_entry(&ldata);
4005 lfsck_object_put(env, parent);
4016 if (rc < 0 && rc != -ENODATA)
4019 if (rc == 0 && ldata.ld_leh != NULL)
4020 count = ldata.ld_leh->leh_reccount;
4023 /* If the LFSCK is marked as LF_INCOMPLETE, then means some
4024 * MDT has ever tried to verify some remote MDT-object that
4025 * resides on this MDT, but this MDT failed to respond such
4026 * request. So means there may be some remote name entry on
4027 * other MDT that references this object with another name,
4028 * so we cannot know whether this linkEA is valid or not.
4029 * So keep it there and maybe resolved when next LFSCK run. */
4030 if (!(ns->ln_flags & LF_INCOMPLETE) &&
4031 (ldata.ld_leh == NULL ||
4032 !ldata.ld_leh->leh_overflow_time)) {
4033 /* If the child becomes orphan, then insert it into
4034 * the global .lustre/lost+found/MDTxxxx directory. */
4035 rc = lfsck_namespace_insert_orphan(env, com, child,
4041 ns->ln_mul_ref_repaired++;
4046 rc = dt_attr_get(env, child, la);
4050 if (la->la_nlink != 0 && la->la_nlink != count) {
4051 if (unlikely(!S_ISREG(lfsck_object_type(child)) &&
4052 !S_ISLNK(lfsck_object_type(child)))) {
4053 CDEBUG(D_LFSCK, "%s: namespace LFSCK finds "
4054 "the object "DFID"'s nlink count %d "
4055 "does not match linkEA count %d, "
4056 "type %o, skip it.\n",
4057 lfsck_lfsck2name(lfsck),
4058 PFID(lfsck_dto2fid(child)),
4059 la->la_nlink, count,
4060 lfsck_object_type(child));
4061 } else if (la->la_nlink < count &&
4062 likely(!ldata.ld_leh->leh_overflow_time)) {
4063 rc = lfsck_namespace_repair_nlink(env, com,
4066 ns->ln_objs_nlink_repaired++;
4074 if (la->la_nlink > 1)
4075 ns->ln_mul_linked_repaired++;
4081 if (!rc && flags & LNTF_CHECK_AGENT_ENTRY)
4082 rc = lfsck_namespace_check_agent_entry(env, com, child);
4087 static void lfsck_namespace_dump_statistics(struct seq_file *m,
4088 struct lfsck_namespace *ns,
4089 __u64 checked_phase1,
4090 __u64 checked_phase2,
4091 time64_t time_phase1,
4092 time64_t time_phase2, bool dryrun)
4094 const char *postfix = dryrun ? "inconsistent" : "repaired";
4096 seq_printf(m, "checked_phase1: %llu\n"
4097 "checked_phase2: %llu\n"
4100 "failed_phase1: %llu\n"
4101 "failed_phase2: %llu\n"
4102 "directories: %llu\n"
4106 "multiple_linked_checked: %llu\n"
4107 "multiple_linked_%s: %llu\n"
4108 "unknown_inconsistency: %llu\n"
4109 "unmatched_pairs_%s: %llu\n"
4110 "dangling_%s: %llu\n"
4111 "multiple_referenced_%s: %llu\n"
4112 "bad_file_type_%s: %llu\n"
4113 "lost_dirent_%s: %llu\n"
4114 "local_lost_found_scanned: %llu\n"
4115 "local_lost_found_moved: %llu\n"
4116 "local_lost_found_skipped: %llu\n"
4117 "local_lost_found_failed: %llu\n"
4118 "striped_dirs_scanned: %llu\n"
4119 "striped_dirs_%s: %llu\n"
4120 "striped_dirs_failed: %llu\n"
4121 "striped_dirs_disabled: %llu\n"
4122 "striped_dirs_skipped: %llu\n"
4123 "striped_shards_scanned: %llu\n"
4124 "striped_shards_%s: %llu\n"
4125 "striped_shards_failed: %llu\n"
4126 "striped_shards_skipped: %llu\n"
4127 "name_hash_%s: %llu\n"
4128 "linkea_overflow_%s: %llu\n"
4129 "agent_entries_%s: %llu\n"
4130 "success_count: %u\n"
4131 "run_time_phase1: %lld seconds\n"
4132 "run_time_phase2: %lld seconds\n",
4135 dryrun ? "inconsistent" : "updated",
4136 ns->ln_items_repaired,
4137 dryrun ? "inconsistent" : "updated",
4138 ns->ln_objs_repaired_phase2,
4139 ns->ln_items_failed,
4140 ns->ln_objs_failed_phase2,
4141 ns->ln_dirs_checked,
4142 postfix, ns->ln_dirent_repaired,
4143 postfix, ns->ln_linkea_repaired,
4144 postfix, ns->ln_objs_nlink_repaired,
4145 ns->ln_mul_linked_checked,
4146 postfix, ns->ln_mul_linked_repaired,
4147 ns->ln_unknown_inconsistency,
4148 postfix, ns->ln_unmatched_pairs_repaired,
4149 postfix, ns->ln_dangling_repaired,
4150 postfix, ns->ln_mul_ref_repaired,
4151 postfix, ns->ln_bad_type_repaired,
4152 postfix, ns->ln_lost_dirent_repaired,
4153 ns->ln_local_lpf_scanned,
4154 ns->ln_local_lpf_moved,
4155 ns->ln_local_lpf_skipped,
4156 ns->ln_local_lpf_failed,
4157 ns->ln_striped_dirs_scanned,
4158 postfix, ns->ln_striped_dirs_repaired,
4159 ns->ln_striped_dirs_failed,
4160 ns->ln_striped_dirs_disabled,
4161 ns->ln_striped_dirs_skipped,
4162 ns->ln_striped_shards_scanned,
4163 postfix, ns->ln_striped_shards_repaired,
4164 ns->ln_striped_shards_failed,
4165 ns->ln_striped_shards_skipped,
4166 postfix, ns->ln_name_hash_repaired,
4167 dryrun ? "inconsistent" : "cleared",
4168 ns->ln_linkea_overflow_cleared,
4169 postfix, ns->ln_agent_entries_repaired,
4170 ns->ln_success_count,
4175 static void lfsck_namespace_release_lmv(const struct lu_env *env,
4176 struct lfsck_component *com)
4178 struct lfsck_instance *lfsck = com->lc_lfsck;
4179 struct lfsck_namespace *ns = com->lc_file_ram;
4181 while (!list_empty(&lfsck->li_list_lmv)) {
4182 struct lfsck_lmv_unit *llu;
4183 struct lfsck_lmv *llmv;
4185 llu = list_first_entry(&lfsck->li_list_lmv,
4186 struct lfsck_lmv_unit, llu_link);
4187 llmv = &llu->llu_lmv;
4189 LASSERTF(atomic_read(&llmv->ll_ref) == 1,
4190 "still in using: %u\n",
4191 atomic_read(&llmv->ll_ref));
4193 ns->ln_striped_dirs_skipped++;
4194 lfsck_lmv_put(env, llmv);
4198 static int lfsck_namespace_check_for_double_scan(const struct lu_env *env,
4199 struct lfsck_component *com,
4200 struct dt_object *obj)
4202 struct lu_attr *la = &lfsck_env_info(env)->lti_la;
4205 rc = dt_attr_get(env, obj, la);
4209 /* zero-linkEA object may be orphan, but it also maybe because
4210 * of upgrading. Currently, we cannot record it for double scan.
4211 * Because it may cause the LFSCK trace file to be too large. */
4213 /* "la_ctime" == 1 means that it has ever been removed from
4214 * backend /lost+found directory but not been added back to
4215 * the normal namespace yet. */
4217 if ((S_ISREG(lfsck_object_type(obj)) && la->la_nlink > 1) ||
4218 unlikely(la->la_ctime == 1))
4219 rc = lfsck_namespace_trace_update(env, com, lfsck_dto2fid(obj),
4220 LNTF_CHECK_LINKEA, true);
4225 /* namespace APIs */
4227 static int lfsck_namespace_reset(const struct lu_env *env,
4228 struct lfsck_component *com, bool init)
4230 struct lfsck_instance *lfsck = com->lc_lfsck;
4231 struct lfsck_namespace *ns = com->lc_file_ram;
4232 struct lfsck_assistant_data *lad = com->lc_data;
4233 struct dt_object *root;
4237 root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
4239 GOTO(log, rc = PTR_ERR(root));
4241 if (unlikely(!dt_try_as_dir(env, root, true)))
4242 GOTO(put, rc = -ENOTDIR);
4244 down_write(&com->lc_sem);
4246 memset(ns, 0, sizeof(*ns));
4248 __u32 count = ns->ln_success_count;
4249 time64_t last_time = ns->ln_time_last_complete;
4251 memset(ns, 0, sizeof(*ns));
4252 ns->ln_success_count = count;
4253 ns->ln_time_last_complete = last_time;
4255 ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
4256 ns->ln_status = LS_INIT;
4257 ns->ln_time_latest_reset = ktime_get_real_seconds();
4259 rc = lfsck_load_one_trace_file(env, com, root, &com->lc_obj,
4260 &dt_lfsck_namespace_features,
4261 LFSCK_NAMESPACE, true);
4265 rc = lfsck_load_sub_trace_files(env, com, &dt_lfsck_namespace_features,
4266 LFSCK_NAMESPACE, true);
4270 clear_bit(LAD_INCOMPLETE, &lad->lad_flags);
4271 bitmap_zero(lad->lad_bitmap, lad->lad_bitmap_count);
4273 rc = lfsck_namespace_store(env, com);
4278 up_write(&com->lc_sem);
4281 lfsck_object_put(env, root);
4283 CDEBUG(D_LFSCK, "%s: namespace LFSCK reset: rc = %d\n",
4284 lfsck_lfsck2name(lfsck), rc);
4289 lfsck_namespace_fail(const struct lu_env *env, struct lfsck_component *com,
4292 struct lfsck_namespace *ns = com->lc_file_ram;
4294 down_write(&com->lc_sem);
4296 com->lc_new_checked++;
4297 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
4298 up_write(&com->lc_sem);
4301 static void lfsck_namespace_close_dir(const struct lu_env *env,
4302 struct lfsck_component *com)
4304 struct lfsck_namespace *ns = com->lc_file_ram;
4305 struct lfsck_assistant_data *lad = com->lc_data;
4306 struct lfsck_assistant_object *lso = NULL;
4307 struct lfsck_instance *lfsck = com->lc_lfsck;
4308 struct lfsck_lmv *llmv = lfsck->li_lmv;
4309 struct lfsck_namespace_req *lnr;
4310 struct lu_attr *la = &lfsck_env_info(env)->lti_la2;
4311 __u32 size = sizeof(*lnr) + LFSCK_TMPBUF_LEN;
4313 bool wakeup = false;
4319 rc = dt_attr_get(env, lfsck->li_obj_dir, la);
4323 OBD_ALLOC(lnr, size);
4325 ns->ln_striped_dirs_skipped++;
4330 lso = lfsck_assistant_object_init(env, lfsck_dto2fid(lfsck->li_obj_dir),
4331 la, lfsck->li_pos_current.lp_oit_cookie, true);
4333 OBD_FREE(lnr, size);
4334 ns->ln_striped_dirs_skipped++;
4339 /* Generate a dummy request to indicate that all shards' name entry
4340 * in this striped directory has been scanned for the first time. */
4341 INIT_LIST_HEAD(&lnr->lnr_lar.lar_list);
4342 lnr->lnr_lar.lar_parent = lso;
4343 lnr->lnr_lmv = lfsck_lmv_get(llmv);
4344 lnr->lnr_fid = *lfsck_dto2fid(lfsck->li_obj_dir);
4345 lnr->lnr_dir_cookie = MDS_DIR_END_OFF;
4346 lnr->lnr_size = size;
4347 lnr->lnr_type = lso->lso_attr.la_mode;
4349 spin_lock(&lad->lad_lock);
4350 if (lad->lad_assistant_status < 0 ||
4351 unlikely(!thread_is_running(&lfsck->li_thread) ||
4352 !thread_is_running(&lad->lad_thread))) {
4353 spin_unlock(&lad->lad_lock);
4354 lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar);
4355 ns->ln_striped_dirs_skipped++;
4360 list_add_tail(&lnr->lnr_lar.lar_list, &lad->lad_req_list);
4361 if (lad->lad_prefetched == 0)
4364 lad->lad_prefetched++;
4365 spin_unlock(&lad->lad_lock);
4367 wake_up(&lad->lad_thread.t_ctl_waitq);
4372 static int lfsck_namespace_open_dir(const struct lu_env *env,
4373 struct lfsck_component *com)
4375 struct lfsck_instance *lfsck = com->lc_lfsck;
4376 struct lfsck_namespace *ns = com->lc_file_ram;
4377 struct lfsck_lmv *llmv = lfsck->li_lmv;
4384 if (llmv->ll_lmv_master) {
4385 struct lmv_mds_md_v1 *lmv = &llmv->ll_lmv;
4387 if (lmv->lmv_master_mdt_index != lfsck_dev_idx(lfsck)) {
4388 lmv->lmv_master_mdt_index =
4389 lfsck_dev_idx(lfsck);
4390 ns->ln_flags |= LF_INCONSISTENT;
4391 llmv->ll_lmv_updated = 1;
4394 rc = lfsck_namespace_verify_stripe_slave(env, com,
4395 lfsck->li_obj_dir, llmv);
4398 RETURN(rc > 0 ? 0 : rc);
4401 static int lfsck_namespace_checkpoint(const struct lu_env *env,
4402 struct lfsck_component *com, bool init)
4404 struct lfsck_instance *lfsck = com->lc_lfsck;
4405 struct lfsck_namespace *ns = com->lc_file_ram;
4409 rc = lfsck_checkpoint_generic(env, com);
4414 down_write(&com->lc_sem);
4416 ns->ln_pos_latest_start = lfsck->li_pos_checkpoint;
4418 ns->ln_pos_last_checkpoint = lfsck->li_pos_checkpoint;
4419 ns->ln_run_time_phase1 += ktime_get_seconds() -
4420 lfsck->li_time_last_checkpoint;
4421 ns->ln_time_last_checkpoint = ktime_get_real_seconds();
4422 ns->ln_items_checked += com->lc_new_checked;
4423 com->lc_new_checked = 0;
4426 rc = lfsck_namespace_store(env, com);
4427 up_write(&com->lc_sem);
4430 CDEBUG(D_LFSCK, "%s: namespace LFSCK checkpoint at the pos [%llu"
4431 ", "DFID", %#llx], status = %d: rc = %d\n",
4432 lfsck_lfsck2name(lfsck), lfsck->li_pos_current.lp_oit_cookie,
4433 PFID(&lfsck->li_pos_current.lp_dir_parent),
4434 lfsck->li_pos_current.lp_dir_cookie, ns->ln_status, rc);
4436 return rc > 0 ? 0 : rc;
4439 static int lfsck_namespace_prep(const struct lu_env *env,
4440 struct lfsck_component *com,
4441 struct lfsck_start_param *lsp)
4443 struct lfsck_instance *lfsck = com->lc_lfsck;
4444 struct lfsck_namespace *ns = com->lc_file_ram;
4445 struct lfsck_position *pos = &com->lc_pos_start;
4448 rc = lfsck_namespace_load_bitmap(env, com);
4449 if (rc != 0 || ns->ln_status == LS_COMPLETED) {
4450 rc = lfsck_namespace_reset(env, com, false);
4452 rc = lfsck_set_param(env, lfsck, lsp->lsp_start, true);
4455 CDEBUG(D_LFSCK, "%s: namespace LFSCK prep failed: "
4456 "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
4462 down_write(&com->lc_sem);
4463 ns->ln_time_latest_start = ktime_get_real_seconds();
4464 spin_lock(&lfsck->li_lock);
4466 if (ns->ln_flags & LF_SCANNED_ONCE) {
4467 if (!lfsck->li_drop_dryrun ||
4468 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent)) {
4469 ns->ln_status = LS_SCANNING_PHASE2;
4470 list_move_tail(&com->lc_link,
4471 &lfsck->li_list_double_scan);
4472 if (!list_empty(&com->lc_link_dir))
4473 list_del_init(&com->lc_link_dir);
4474 lfsck_pos_set_zero(pos);
4476 ns->ln_status = LS_SCANNING_PHASE1;
4477 ns->ln_run_time_phase1 = 0;
4478 ns->ln_run_time_phase2 = 0;
4479 ns->ln_items_checked = 0;
4480 ns->ln_items_repaired = 0;
4481 ns->ln_items_failed = 0;
4482 ns->ln_dirs_checked = 0;
4483 ns->ln_objs_checked_phase2 = 0;
4484 ns->ln_objs_repaired_phase2 = 0;
4485 ns->ln_objs_failed_phase2 = 0;
4486 ns->ln_objs_nlink_repaired = 0;
4487 ns->ln_dirent_repaired = 0;
4488 ns->ln_linkea_repaired = 0;
4489 ns->ln_mul_linked_checked = 0;
4490 ns->ln_mul_linked_repaired = 0;
4491 ns->ln_unknown_inconsistency = 0;
4492 ns->ln_unmatched_pairs_repaired = 0;
4493 ns->ln_dangling_repaired = 0;
4494 ns->ln_mul_ref_repaired = 0;
4495 ns->ln_bad_type_repaired = 0;
4496 ns->ln_lost_dirent_repaired = 0;
4497 ns->ln_striped_dirs_scanned = 0;
4498 ns->ln_striped_dirs_repaired = 0;
4499 ns->ln_striped_dirs_failed = 0;
4500 ns->ln_striped_dirs_disabled = 0;
4501 ns->ln_striped_dirs_skipped = 0;
4502 ns->ln_striped_shards_scanned = 0;
4503 ns->ln_striped_shards_repaired = 0;
4504 ns->ln_striped_shards_failed = 0;
4505 ns->ln_striped_shards_skipped = 0;
4506 ns->ln_name_hash_repaired = 0;
4507 fid_zero(&ns->ln_fid_latest_scanned_phase2);
4508 if (list_empty(&com->lc_link_dir))
4509 list_add_tail(&com->lc_link_dir,
4510 &lfsck->li_list_dir);
4511 *pos = ns->ln_pos_first_inconsistent;
4514 ns->ln_status = LS_SCANNING_PHASE1;
4515 if (list_empty(&com->lc_link_dir))
4516 list_add_tail(&com->lc_link_dir,
4517 &lfsck->li_list_dir);
4518 if (!lfsck->li_drop_dryrun ||
4519 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent)) {
4520 *pos = ns->ln_pos_last_checkpoint;
4521 pos->lp_oit_cookie++;
4523 *pos = ns->ln_pos_first_inconsistent;
4527 spin_unlock(&lfsck->li_lock);
4528 up_write(&com->lc_sem);
4530 rc = lfsck_start_assistant(env, com, lsp);
4532 CDEBUG(D_LFSCK, "%s: namespace LFSCK prep done, start pos [%llu, "
4533 DFID", %#llx]: rc = %d\n",
4534 lfsck_lfsck2name(lfsck), pos->lp_oit_cookie,
4535 PFID(&pos->lp_dir_parent), pos->lp_dir_cookie, rc);
4540 static int lfsck_namespace_exec_oit(const struct lu_env *env,
4541 struct lfsck_component *com,
4542 struct dt_object *obj)
4544 struct lfsck_thread_info *info = lfsck_env_info(env);
4545 struct lfsck_namespace *ns = com->lc_file_ram;
4546 struct lfsck_instance *lfsck = com->lc_lfsck;
4547 const struct lu_fid *fid = lfsck_dto2fid(obj);
4548 struct lu_fid *pfid = &info->lti_fid2;
4549 struct lu_name *cname = &info->lti_name;
4550 struct lu_seq_range *range = &info->lti_range;
4551 struct seq_server_site *ss = lfsck_dev_site(lfsck);
4552 struct linkea_data ldata = { NULL };
4553 __u32 idx = lfsck_dev_idx(lfsck);
4554 struct lu_attr la = { .la_valid = 0 };
4555 bool remote = false;
4559 rc = dt_attr_get(env, obj, &la);
4560 if (unlikely(rc || (la.la_valid & LA_FLAGS &&
4561 la.la_flags & LUSTRE_ORPHAN_FL))) {
4563 "%s: skip orphan "DFID", %llx/%x: rc = %d\n",
4564 lfsck_lfsck2name(lfsck), PFID(fid),
4565 la.la_valid, la.la_flags, rc);
4570 rc = lfsck_links_read(env, obj, &ldata);
4574 /* -EINVAL means crashed linkEA, should be verified. */
4575 if (rc == -EINVAL) {
4576 rc = lfsck_namespace_trace_update(env, com, fid,
4577 LNTF_CHECK_LINKEA, true);
4579 struct lustre_handle lh = { 0 };
4581 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
4582 MDS_INODELOCK_UPDATE |
4583 MDS_INODELOCK_XATTR, LCK_EX);
4585 rc = lfsck_namespace_links_remove(env, com,
4587 lfsck_ibits_unlock(&lh, LCK_EX);
4591 GOTO(out, rc = (rc == -ENOENT ? 0 : rc));
4594 if (rc && rc != -ENODATA)
4597 if (rc == -ENODATA || unlikely(!ldata.ld_leh->leh_reccount)) {
4598 rc = lfsck_namespace_check_for_double_scan(env, com, obj);
4603 linkea_first_entry(&ldata);
4604 while (ldata.ld_lee != NULL) {
4605 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
4607 if (!fid_is_sane(pfid)) {
4608 rc = lfsck_namespace_trace_update(env, com, fid,
4609 LNTF_CHECK_PARENT, true);
4610 } else if (!linkea_entry_is_valid(&ldata, cname, pfid)) {
4613 fld_range_set_mdt(range);
4614 rc = fld_server_lookup(env, ss->ss_server_fld,
4615 fid_seq(pfid), range);
4616 if ((rc == -ENOENT) ||
4617 (!rc && range->lsr_index != idx)) {
4625 linkea_next_entry(&ldata);
4628 if ((lu_object_has_agent_entry(&obj->do_lu) && !remote) ||
4629 (!lu_object_has_agent_entry(&obj->do_lu) && remote)) {
4630 rc = lfsck_namespace_trace_update(env, com, fid,
4631 LNTF_CHECK_AGENT_ENTRY, true);
4636 /* Record multiple-linked object. */
4637 if (ldata.ld_leh->leh_reccount > 1) {
4638 rc = lfsck_namespace_trace_update(env, com, fid,
4639 LNTF_CHECK_LINKEA, true);
4645 rc = lfsck_namespace_trace_update(env, com, fid,
4646 LNTF_CHECK_LINKEA, true);
4648 rc = lfsck_namespace_check_for_double_scan(env, com, obj);
4653 down_write(&com->lc_sem);
4654 if (S_ISDIR(lfsck_object_type(obj)))
4655 ns->ln_dirs_checked++;
4657 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
4658 up_write(&com->lc_sem);
4663 static int lfsck_namespace_exec_dir(const struct lu_env *env,
4664 struct lfsck_component *com,
4665 struct lfsck_assistant_object *lso,
4666 struct lu_dirent *ent, __u16 type)
4668 struct lfsck_assistant_data *lad = com->lc_data;
4669 struct lfsck_instance *lfsck = com->lc_lfsck;
4670 struct lfsck_namespace_req *lnr;
4671 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4672 struct ptlrpc_thread *mthread = &lfsck->li_thread;
4673 struct ptlrpc_thread *athread = &lad->lad_thread;
4674 bool wakeup = false;
4676 wait_event_idle(mthread->t_ctl_waitq,
4677 lad->lad_prefetched < bk->lb_async_windows ||
4678 !thread_is_running(mthread) ||
4679 !thread_is_running(athread));
4681 if (unlikely(!thread_is_running(mthread) ||
4682 !thread_is_running(athread)))
4685 if (unlikely(lfsck_is_dead_obj(lfsck->li_obj_dir)))
4688 lnr = lfsck_namespace_assistant_req_init(com->lc_lfsck, lso, ent, type);
4690 struct lfsck_namespace *ns = com->lc_file_ram;
4692 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
4693 return PTR_ERR(lnr);
4696 spin_lock(&lad->lad_lock);
4697 if (lad->lad_assistant_status < 0 ||
4698 unlikely(!thread_is_running(mthread) ||
4699 !thread_is_running(athread))) {
4700 spin_unlock(&lad->lad_lock);
4701 lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar);
4702 return lad->lad_assistant_status;
4705 list_add_tail(&lnr->lnr_lar.lar_list, &lad->lad_req_list);
4706 if (lad->lad_prefetched == 0)
4709 lad->lad_prefetched++;
4710 spin_unlock(&lad->lad_lock);
4712 wake_up(&lad->lad_thread.t_ctl_waitq);
4714 down_write(&com->lc_sem);
4715 com->lc_new_checked++;
4716 up_write(&com->lc_sem);
4721 static int lfsck_namespace_post(const struct lu_env *env,
4722 struct lfsck_component *com,
4723 int result, bool init)
4725 struct lfsck_instance *lfsck = com->lc_lfsck;
4726 struct lfsck_namespace *ns = com->lc_file_ram;
4730 lfsck_post_generic(env, com, &result);
4732 down_write(&com->lc_sem);
4733 lfsck_namespace_release_lmv(env, com);
4735 spin_lock(&lfsck->li_lock);
4737 ns->ln_pos_last_checkpoint = lfsck->li_pos_checkpoint;
4739 ns->ln_status = LS_SCANNING_PHASE2;
4740 ns->ln_flags |= LF_SCANNED_ONCE;
4741 ns->ln_flags &= ~LF_UPGRADE;
4742 list_del_init(&com->lc_link_dir);
4743 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
4744 } else if (result == 0) {
4745 if (lfsck->li_status != 0)
4746 ns->ln_status = lfsck->li_status;
4748 ns->ln_status = LS_STOPPED;
4749 if (ns->ln_status != LS_PAUSED) {
4750 list_del_init(&com->lc_link_dir);
4751 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4754 ns->ln_status = LS_FAILED;
4755 list_del_init(&com->lc_link_dir);
4756 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4758 spin_unlock(&lfsck->li_lock);
4761 ns->ln_run_time_phase1 += ktime_get_seconds() -
4762 lfsck->li_time_last_checkpoint;
4763 ns->ln_time_last_checkpoint = ktime_get_real_seconds();
4764 ns->ln_items_checked += com->lc_new_checked;
4765 com->lc_new_checked = 0;
4768 rc = lfsck_namespace_store(env, com);
4769 up_write(&com->lc_sem);
4771 CDEBUG(D_LFSCK, "%s: namespace LFSCK post done: rc = %d\n",
4772 lfsck_lfsck2name(lfsck), rc);
4778 lfsck_namespace_dump(const struct lu_env *env, struct lfsck_component *com,
4781 struct lfsck_instance *lfsck = com->lc_lfsck;
4782 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4783 struct lfsck_namespace *ns = com->lc_file_ram;
4785 down_read(&com->lc_sem);
4786 seq_printf(m, "name: lfsck_namespace\n"
4792 lfsck_status2name(ns->ln_status));
4794 lfsck_bits_dump(m, ns->ln_flags, lfsck_flags_names, "flags");
4796 lfsck_bits_dump(m, bk->lb_param, lfsck_param_names, "param");
4798 lfsck_time_dump(m, ns->ln_time_last_complete, "last_completed");
4800 lfsck_time_dump(m, ns->ln_time_latest_start, "latest_start");
4802 lfsck_time_dump(m, ns->ln_time_last_checkpoint, "last_checkpoint");
4804 lfsck_pos_dump(m, &ns->ln_pos_latest_start, "latest_start_position");
4806 lfsck_pos_dump(m, &ns->ln_pos_last_checkpoint,
4807 "last_checkpoint_position");
4809 lfsck_pos_dump(m, &ns->ln_pos_first_inconsistent,
4810 "first_failure_position");
4812 if (ns->ln_status == LS_SCANNING_PHASE1) {
4813 struct lfsck_position pos;
4814 time64_t duration = ktime_get_seconds() -
4815 lfsck->li_time_last_checkpoint;
4816 u64 checked = ns->ln_items_checked + com->lc_new_checked;
4817 u64 speed = checked;
4818 u64 new_checked = com->lc_new_checked;
4819 time64_t rtime = ns->ln_run_time_phase1 + duration;
4822 new_checked = div64_s64(new_checked, duration);
4825 speed = div64_s64(speed, rtime);
4827 lfsck_namespace_dump_statistics(m, ns, checked, 0, rtime, 0,
4828 bk->lb_param & LPF_DRYRUN);
4829 seq_printf(m, "average_speed_phase1: %llu items/sec\n"
4830 "average_speed_phase2: N/A\n"
4831 "average_speed_total: %llu items/sec\n"
4832 "real_time_speed_phase1: %llu items/sec\n"
4833 "real_time_speed_phase2: N/A\n",
4838 if (likely(lfsck->li_di_oit)) {
4839 const struct dt_it_ops *iops =
4840 &lfsck->li_obj_oit->do_index_ops->dio_it;
4842 /* The low layer otable-based iteration position may NOT
4843 * exactly match the namespace-based directory traversal
4844 * cookie. Generally, it is not a serious issue. But the
4845 * caller should NOT make assumption on that. */
4846 pos.lp_oit_cookie = iops->store(env, lfsck->li_di_oit);
4847 if (!lfsck->li_current_oit_processed)
4848 pos.lp_oit_cookie--;
4850 spin_lock(&lfsck->li_lock);
4851 if (lfsck->li_di_dir) {
4852 pos.lp_dir_cookie = lfsck->li_cookie_dir;
4853 if (pos.lp_dir_cookie >= MDS_DIR_END_OFF) {
4854 fid_zero(&pos.lp_dir_parent);
4855 pos.lp_dir_cookie = 0;
4858 *lfsck_dto2fid(lfsck->li_obj_dir);
4861 fid_zero(&pos.lp_dir_parent);
4862 pos.lp_dir_cookie = 0;
4864 spin_unlock(&lfsck->li_lock);
4866 pos = ns->ln_pos_last_checkpoint;
4869 lfsck_pos_dump(m, &pos, "current_position");
4870 } else if (ns->ln_status == LS_SCANNING_PHASE2) {
4871 time64_t duration = ktime_get_seconds() -
4872 com->lc_time_last_checkpoint;
4873 __u64 checked = ns->ln_objs_checked_phase2 +
4874 com->lc_new_checked;
4875 __u64 speed1 = ns->ln_items_checked;
4876 __u64 speed2 = checked;
4877 __u64 speed0 = speed1 + speed2;
4878 __u64 new_checked = com->lc_new_checked;
4879 time64_t rtime = ns->ln_run_time_phase2 + duration;
4880 time64_t time0 = ns->ln_run_time_phase1 + rtime;
4883 new_checked = div64_s64(new_checked, duration);
4885 if (ns->ln_run_time_phase1 != 0)
4886 speed1 = div64_s64(speed1, ns->ln_run_time_phase1);
4887 else if (ns->ln_items_checked != 0)
4891 speed2 = div64_s64(speed2, rtime);
4892 else if (checked != 0)
4896 speed0 = div64_s64(speed0, time0);
4898 lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked,
4900 ns->ln_run_time_phase1, rtime,
4901 bk->lb_param & LPF_DRYRUN);
4902 seq_printf(m, "average_speed_phase1: %llu items/sec\n"
4903 "average_speed_phase2: %llu objs/sec\n"
4904 "average_speed_total: %llu items/sec\n"
4905 "real_time_speed_phase1: N/A\n"
4906 "real_time_speed_phase2: %llu objs/sec\n"
4907 "current_position: "DFID"\n",
4912 PFID(&ns->ln_fid_latest_scanned_phase2));
4914 __u64 speed1 = ns->ln_items_checked;
4915 __u64 speed2 = ns->ln_objs_checked_phase2;
4916 __u64 speed0 = speed1 + speed2;
4917 time64_t time0 = ns->ln_run_time_phase1 + ns->ln_run_time_phase2;
4919 if (ns->ln_run_time_phase1 != 0)
4920 speed1 = div64_s64(speed1, ns->ln_run_time_phase1);
4921 else if (ns->ln_items_checked != 0)
4924 if (ns->ln_run_time_phase2 != 0)
4925 speed2 = div64_s64(speed2, ns->ln_run_time_phase2);
4926 else if (ns->ln_objs_checked_phase2 != 0)
4930 speed0 = div64_s64(speed0, time0);
4932 lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked,
4933 ns->ln_objs_checked_phase2,
4934 ns->ln_run_time_phase1,
4935 ns->ln_run_time_phase2,
4936 bk->lb_param & LPF_DRYRUN);
4937 seq_printf(m, "average_speed_phase1: %llu items/sec\n"
4938 "average_speed_phase2: %llu objs/sec\n"
4939 "average_speed_total: %llu items/sec\n"
4940 "real_time_speed_phase1: N/A\n"
4941 "real_time_speed_phase2: N/A\n"
4942 "current_position: N/A\n",
4948 up_read(&com->lc_sem);
4951 static int lfsck_namespace_double_scan(const struct lu_env *env,
4952 struct lfsck_component *com)
4954 struct lfsck_namespace *ns = com->lc_file_ram;
4955 struct lfsck_assistant_data *lad = com->lc_data;
4956 struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
4957 struct lfsck_tgt_desc *ltd;
4958 struct lfsck_tgt_desc *next;
4961 rc = lfsck_double_scan_generic(env, com, ns->ln_status);
4962 if (thread_is_stopped(&lad->lad_thread)) {
4963 LASSERT(list_empty(&lad->lad_req_list));
4964 LASSERT(list_empty(&lad->lad_mdt_phase1_list));
4966 spin_lock(<ds->ltd_lock);
4967 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4968 ltd_namespace_phase_list) {
4969 list_del_init(<d->ltd_namespace_phase_list);
4971 spin_unlock(<ds->ltd_lock);
4977 static void lfsck_namespace_data_release(const struct lu_env *env,
4978 struct lfsck_component *com)
4980 struct lfsck_assistant_data *lad = com->lc_data;
4981 struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
4982 struct lfsck_tgt_desc *ltd;
4983 struct lfsck_tgt_desc *next;
4985 LASSERT(lad != NULL);
4986 LASSERT(thread_is_init(&lad->lad_thread) ||
4987 thread_is_stopped(&lad->lad_thread));
4988 LASSERT(list_empty(&lad->lad_req_list));
4990 com->lc_data = NULL;
4991 lfsck_namespace_release_lmv(env, com);
4993 spin_lock(<ds->ltd_lock);
4994 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
4995 ltd_namespace_phase_list) {
4996 list_del_init(<d->ltd_namespace_phase_list);
4998 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4999 ltd_namespace_phase_list) {
5000 list_del_init(<d->ltd_namespace_phase_list);
5002 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_list,
5003 ltd_namespace_list) {
5004 list_del_init(<d->ltd_namespace_list);
5006 spin_unlock(<ds->ltd_lock);
5008 if (likely(lad->lad_bitmap != NULL))
5009 bitmap_free(lad->lad_bitmap);
5014 static void lfsck_namespace_quit(const struct lu_env *env,
5015 struct lfsck_component *com)
5017 struct lfsck_assistant_data *lad = com->lc_data;
5018 struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
5019 struct lfsck_tgt_desc *ltd;
5020 struct lfsck_tgt_desc *next;
5022 LASSERT(lad != NULL);
5024 lfsck_quit_generic(env, com);
5026 LASSERT(thread_is_init(&lad->lad_thread) ||
5027 thread_is_stopped(&lad->lad_thread));
5028 LASSERT(list_empty(&lad->lad_req_list));
5030 lfsck_namespace_release_lmv(env, com);
5032 spin_lock(<ds->ltd_lock);
5033 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
5034 ltd_namespace_phase_list) {
5035 list_del_init(<d->ltd_namespace_phase_list);
5037 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
5038 ltd_namespace_phase_list) {
5039 list_del_init(<d->ltd_namespace_phase_list);
5041 spin_unlock(<ds->ltd_lock);
5044 static int lfsck_namespace_in_notify(const struct lu_env *env,
5045 struct lfsck_component *com,
5046 struct lfsck_request *lr)
5048 struct lfsck_instance *lfsck = com->lc_lfsck;
5049 struct lfsck_namespace *ns = com->lc_file_ram;
5050 struct lfsck_assistant_data *lad = com->lc_data;
5051 struct lfsck_tgt_descs *ltds = &lfsck->li_mdt_descs;
5052 struct lfsck_tgt_desc *ltd;
5057 switch (lr->lr_event) {
5058 case LE_SET_LMV_MASTER: {
5059 struct dt_object *obj;
5061 obj = lfsck_object_find_bottom(env, lfsck, &lr->lr_fid);
5063 RETURN(PTR_ERR(obj));
5065 if (likely(dt_object_exists(obj)))
5066 rc = lfsck_namespace_notify_lmv_master_local(env, com,
5069 lfsck_object_put(env, obj);
5071 RETURN(rc > 0 ? 0 : rc);
5073 case LE_SET_LMV_SLAVE: {
5074 if (!(lr->lr_flags & LEF_RECHECK_NAME_HASH))
5075 ns->ln_striped_shards_repaired++;
5077 rc = lfsck_namespace_trace_update(env, com, &lr->lr_fid,
5078 LNTF_RECHECK_NAME_HASH, true);
5080 RETURN(rc > 0 ? 0 : rc);
5082 case LE_PHASE1_DONE:
5083 case LE_PHASE2_DONE:
5090 CDEBUG(D_LFSCK, "%s: namespace LFSCK handles notify %u from MDT %x, "
5091 "status %d, flags %x\n", lfsck_lfsck2name(lfsck), lr->lr_event,
5092 lr->lr_index, lr->lr_status, lr->lr_flags2);
5094 spin_lock(<ds->ltd_lock);
5095 ltd = lfsck_ltd2tgt(ltds, lr->lr_index);
5097 spin_unlock(<ds->ltd_lock);
5102 list_del_init(<d->ltd_namespace_phase_list);
5103 switch (lr->lr_event) {
5104 case LE_PHASE1_DONE:
5105 if (lr->lr_status <= 0) {
5106 ltd->ltd_namespace_done = 1;
5107 list_del_init(<d->ltd_namespace_list);
5108 CDEBUG(D_LFSCK, "%s: MDT %x failed/stopped at "
5109 "phase1 for namespace LFSCK: rc = %d.\n",
5110 lfsck_lfsck2name(lfsck),
5111 ltd->ltd_index, lr->lr_status);
5112 ns->ln_flags |= LF_INCOMPLETE;
5117 if (lr->lr_flags2 & LF_INCOMPLETE)
5118 ns->ln_flags |= LF_INCOMPLETE;
5120 if (list_empty(<d->ltd_namespace_list))
5121 list_add_tail(<d->ltd_namespace_list,
5122 &lad->lad_mdt_list);
5123 list_add_tail(<d->ltd_namespace_phase_list,
5124 &lad->lad_mdt_phase2_list);
5126 case LE_PHASE2_DONE:
5127 ltd->ltd_namespace_done = 1;
5128 list_del_init(<d->ltd_namespace_list);
5132 ltd->ltd_namespace_done = 1;
5133 list_del_init(<d->ltd_namespace_list);
5134 if (!(lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT)) {
5136 "%s: the peer MDT %x exit namespace LFSCK\n",
5137 lfsck_lfsck2name(lfsck), ltd->ltd_index);
5138 ns->ln_flags |= LF_INCOMPLETE;
5144 spin_unlock(<ds->ltd_lock);
5146 if (fail && lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) {
5147 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
5149 memset(stop, 0, sizeof(*stop));
5150 stop->ls_status = lr->lr_status;
5151 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
5152 lfsck_stop(env, lfsck->li_bottom, stop);
5153 } else if (lfsck_phase2_next_ready(lad)) {
5154 wake_up(&lad->lad_thread.t_ctl_waitq);
5160 static void lfsck_namespace_repaired(struct lfsck_namespace *ns, __u64 *count)
5162 *count += ns->ln_objs_nlink_repaired;
5163 *count += ns->ln_dirent_repaired;
5164 *count += ns->ln_linkea_repaired;
5165 *count += ns->ln_mul_linked_repaired;
5166 *count += ns->ln_unmatched_pairs_repaired;
5167 *count += ns->ln_dangling_repaired;
5168 *count += ns->ln_mul_ref_repaired;
5169 *count += ns->ln_bad_type_repaired;
5170 *count += ns->ln_lost_dirent_repaired;
5171 *count += ns->ln_striped_dirs_disabled;
5172 *count += ns->ln_striped_dirs_repaired;
5173 *count += ns->ln_striped_shards_repaired;
5174 *count += ns->ln_name_hash_repaired;
5175 *count += ns->ln_local_lpf_moved;
5178 static int lfsck_namespace_query_all(const struct lu_env *env,
5179 struct lfsck_component *com,
5180 __u32 *mdts_count, __u64 *repaired)
5182 struct lfsck_namespace *ns = com->lc_file_ram;
5183 struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
5184 struct lfsck_tgt_desc *ltd;
5189 rc = lfsck_query_all(env, com);
5193 down_read(<ds->ltd_rw_sem);
5194 for_each_set_bit(idx, ltds->ltd_tgts_bitmap, ltds->ltd_tgts_mask_len) {
5195 ltd = lfsck_ltd2tgt(ltds, idx);
5196 LASSERT(ltd != NULL);
5198 mdts_count[ltd->ltd_namespace_status]++;
5199 *repaired += ltd->ltd_namespace_repaired;
5201 up_read(<ds->ltd_rw_sem);
5203 down_read(&com->lc_sem);
5204 mdts_count[ns->ln_status]++;
5205 lfsck_namespace_repaired(ns, repaired);
5206 up_read(&com->lc_sem);
5211 static int lfsck_namespace_query(const struct lu_env *env,
5212 struct lfsck_component *com,
5213 struct lfsck_request *req,
5214 struct lfsck_reply *rep,
5215 struct lfsck_query *que, int idx)
5217 struct lfsck_namespace *ns = com->lc_file_ram;
5221 LASSERT(com->lc_lfsck->li_master);
5223 rc = lfsck_namespace_query_all(env, com,
5224 que->lu_mdts_count[idx],
5225 &que->lu_repaired[idx]);
5227 down_read(&com->lc_sem);
5228 rep->lr_status = ns->ln_status;
5229 if (req->lr_flags & LEF_QUERY_ALL)
5230 lfsck_namespace_repaired(ns, &rep->lr_repaired);
5231 up_read(&com->lc_sem);
5237 static const struct lfsck_operations lfsck_namespace_ops = {
5238 .lfsck_reset = lfsck_namespace_reset,
5239 .lfsck_fail = lfsck_namespace_fail,
5240 .lfsck_close_dir = lfsck_namespace_close_dir,
5241 .lfsck_open_dir = lfsck_namespace_open_dir,
5242 .lfsck_checkpoint = lfsck_namespace_checkpoint,
5243 .lfsck_prep = lfsck_namespace_prep,
5244 .lfsck_exec_oit = lfsck_namespace_exec_oit,
5245 .lfsck_exec_dir = lfsck_namespace_exec_dir,
5246 .lfsck_post = lfsck_namespace_post,
5247 .lfsck_dump = lfsck_namespace_dump,
5248 .lfsck_double_scan = lfsck_namespace_double_scan,
5249 .lfsck_data_release = lfsck_namespace_data_release,
5250 .lfsck_quit = lfsck_namespace_quit,
5251 .lfsck_in_notify = lfsck_namespace_in_notify,
5252 .lfsck_query = lfsck_namespace_query,
5256 * Repair dangling name entry.
5258 * For the name entry with dangling reference, we need to repare the
5259 * inconsistency according to the LFSCK sponsor's requirement:
5261 * 1) Keep the inconsistency there and report the inconsistency case,
5262 * then give the chance to the application to find related issues,
5263 * and the users can make the decision about how to handle it with
5264 * more human knownledge. (by default)
5266 * 2) Re-create the missing MDT-object with the FID information.
5268 * \param[in] env pointer to the thread context
5269 * \param[in] com pointer to the lfsck component
5270 * \param[in] parent pointer to the dir object that contains the dangling
5272 * \param[in] child pointer to the object corresponding to the dangling
5274 * \param[in] lnr pointer to the namespace request that contains the
5275 * name's name, parent object, parent's LMV, and ect.
5277 * \retval positive number if no need to repair
5278 * \retval zero for repaired successfully
5279 * \retval negative error number on failure
5281 int lfsck_namespace_repair_dangling(const struct lu_env *env,
5282 struct lfsck_component *com,
5283 struct dt_object *parent,
5284 struct dt_object *child,
5285 struct lfsck_namespace_req *lnr)
5287 struct lfsck_thread_info *info = lfsck_env_info(env);
5288 struct lu_attr *la = &info->lti_la;
5289 struct dt_allocation_hint *hint = &info->lti_hint;
5290 struct dt_object_format *dof = &info->lti_dof;
5291 struct dt_insert_rec *rec = &info->lti_dt_rec;
5292 struct lmv_mds_md_v1 *lmv2 = &info->lti_lmv2;
5293 const struct lu_name *cname;
5294 const struct lu_fid *pfid = lfsck_dto2fid(parent);
5295 const struct lu_fid *cfid = lfsck_dto2fid(child);
5296 struct linkea_data ldata = { NULL };
5297 struct lfsck_lock_handle *llh = &info->lti_llh;
5298 struct lustre_handle rlh = { 0 };
5299 struct lustre_handle clh = { 0 };
5300 struct lu_buf linkea_buf;
5301 struct lu_buf lmv_buf;
5302 struct lfsck_instance *lfsck = com->lc_lfsck;
5303 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5304 struct dt_device *dev = lfsck->li_next;
5305 struct thandle *th = NULL;
5307 __u16 type = lnr->lnr_type;
5311 cname = lfsck_name_get_const(env, lnr->lnr_name, lnr->lnr_namelen);
5312 if (bk->lb_param & LPF_CREATE_MDTOBJ)
5317 if (!create || bk->lb_param & LPF_DRYRUN)
5320 /* We may need to create the sub-objects of the @child via LOD,
5321 * so make the modification based on lfsck->li_next. */
5323 parent = lfsck_object_locate(dev, parent);
5325 GOTO(log, rc = PTR_ERR(parent));
5327 if (unlikely(!dt_try_as_dir(env, parent, true)))
5328 GOTO(log, rc = -ENOTDIR);
5330 child = lfsck_object_locate(dev, child);
5332 GOTO(log, rc = PTR_ERR(child));
5334 rc = linkea_links_new(&ldata, &info->lti_linkea_buf2,
5339 rc = lfsck_lock(env, lfsck, parent, lnr->lnr_name, llh,
5340 MDS_INODELOCK_UPDATE, LCK_PW);
5344 rc = lfsck_namespace_check_exist(env, parent, child, lnr->lnr_name);
5348 if (dt_object_remote(child)) {
5349 rc = lfsck_remote_lookup_lock(env, lfsck, parent, child, &rlh,
5355 rc = lfsck_ibits_lock(env, lfsck, child, &clh,
5356 MDS_INODELOCK_UPDATE | MDS_INODELOCK_LOOKUP |
5357 MDS_INODELOCK_XATTR, LCK_EX);
5359 GOTO(unlock_remote_lookup, rc);
5361 /* Set the ctime as zero, then others can know it is created for
5362 * repairing dangling name entry by LFSCK. And if the LFSCK made
5363 * wrong decision and the real MDT-object has been found later,
5364 * then the LFSCK has chance to fix the incosistency properly. */
5365 memset(la, 0, sizeof(*la));
5367 la->la_mode = (type & S_IFMT) | 0700;
5369 la->la_mode = (type & S_IFMT) | 0600;
5370 la->la_valid = LA_TYPE | LA_MODE | LA_CTIME;
5373 * if it's directory, skip do_ah_init() to create a plain directory
5374 * because it may have shards already, which will be inserted back
5375 * later, besides, it may be remote, and creating stripe directory
5376 * remotely is not supported.
5379 child->do_ops->do_ah_init(env, hint, parent, child, type);
5380 else if (S_ISDIR(type))
5381 child->do_ops->do_ah_init(env, hint, NULL, child, type);
5383 memset(dof, 0, sizeof(*dof));
5384 dof->dof_type = dt_mode_to_dft(type);
5385 /* If the target is a regular file, then the LFSCK will only create
5386 * the MDT-object without stripes (dof->dof_reg.striped = 0). related
5387 * OST-objects will be created when write open. */
5389 th = lfsck_trans_create(env, dev, lfsck);
5391 GOTO(unlock_child, rc = PTR_ERR(th));
5393 /* 1a. create child. */
5394 rc = dt_declare_create(env, child, la, hint, dof, th);
5398 if (S_ISDIR(type)) {
5399 if (unlikely(!dt_try_as_dir(env, child, false)))
5400 GOTO(stop, rc = -ENOTDIR);
5402 /* 2a. increase child nlink */
5403 rc = dt_declare_ref_add(env, child, th);
5407 /* 3a. insert dot into child dir */
5408 rec->rec_type = S_IFDIR;
5409 rec->rec_fid = cfid;
5410 rc = dt_declare_insert(env, child,
5411 (const struct dt_rec *)rec,
5412 (const struct dt_key *)dot, th);
5416 /* 4a. insert dotdot into child dir */
5417 rec->rec_fid = pfid;
5418 rc = dt_declare_insert(env, child,
5419 (const struct dt_rec *)rec,
5420 (const struct dt_key *)dotdot, th);
5424 /* 5a. generate slave LMV EA. */
5425 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
5428 idx = lfsck_shard_name_to_index(env,
5429 lnr->lnr_name, lnr->lnr_namelen,
5431 if (unlikely(idx < 0))
5432 GOTO(stop, rc = idx);
5434 *lmv2 = lnr->lnr_lmv->ll_lmv;
5435 lmv2->lmv_magic = LMV_MAGIC_STRIPE;
5436 lmv2->lmv_master_mdt_index = idx;
5438 lfsck_lmv_header_cpu_to_le(lmv2, lmv2);
5439 lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
5440 rc = dt_declare_xattr_set(env, child, &lmv_buf,
5441 XATTR_NAME_LMV, 0, th);
5447 /* 6a. insert linkEA for child */
5448 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
5449 ldata.ld_leh->leh_len);
5450 rc = dt_declare_xattr_set(env, child, &linkea_buf,
5451 XATTR_NAME_LINK, 0, th);
5455 /* 7a. if child is remote, delete and insert to generate local agent */
5456 if (dt_object_remote(child)) {
5457 rc = dt_declare_delete(env, parent,
5458 (const struct dt_key *)lnr->lnr_name,
5463 rec->rec_fid = cfid;
5464 rec->rec_type = type;
5465 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
5466 (const struct dt_key *)lnr->lnr_name,
5472 rc = dt_trans_start_local(env, dev, th);
5474 GOTO(stop, rc = (rc == -EEXIST ? 1 : rc));
5476 dt_write_lock(env, child, 0);
5477 /* 1b. create child */
5478 rc = dt_create(env, child, la, hint, dof, th);
5480 GOTO(unlock, rc = (rc == -EEXIST ? 1 : rc));
5482 if (S_ISDIR(type)) {
5483 /* 2b. increase child nlink */
5484 rc = dt_ref_add(env, child, th);
5488 /* 3b. insert dot into child dir */
5489 rec->rec_type = S_IFDIR;
5490 rec->rec_fid = cfid;
5491 rc = dt_insert(env, child, (const struct dt_rec *)rec,
5492 (const struct dt_key *)dot, th);
5496 /* 4b. insert dotdot into child dir */
5497 rec->rec_fid = pfid;
5498 rc = dt_insert(env, child, (const struct dt_rec *)rec,
5499 (const struct dt_key *)dotdot, th);
5503 /* 5b. generate slave LMV EA. */
5504 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
5505 rc = dt_xattr_set(env, child, &lmv_buf, XATTR_NAME_LMV,
5512 /* 6b. insert linkEA for child. */
5513 rc = dt_xattr_set(env, child, &linkea_buf,
5514 XATTR_NAME_LINK, 0, th);
5518 /* 7b. if child is remote, delete and insert to generate local agent */
5519 if (dt_object_remote(child)) {
5520 rc = dt_delete(env, parent,
5521 (const struct dt_key *)lnr->lnr_name, th);
5525 rec->rec_type = type;
5526 rec->rec_fid = cfid;
5527 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
5528 (const struct dt_key *)lnr->lnr_name, th);
5536 dt_write_unlock(env, child);
5539 dt_trans_stop(env, dev, th);
5542 lfsck_ibits_unlock(&clh, LCK_EX);
5543 unlock_remote_lookup:
5544 if (dt_object_remote(child))
5545 lfsck_ibits_unlock(&rlh, LCK_EX);
5548 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found dangling "
5549 "reference for: parent "DFID", child "DFID", type %u, "
5550 "name %s. %s: rc = %d\n", lfsck_lfsck2name(lfsck),
5551 PFID(pfid), PFID(cfid), type, cname->ln_name,
5552 create ? "Create the lost MDT-object as required" :
5553 "Keep the MDT-object there by default", rc);
5556 struct lfsck_namespace *ns = com->lc_file_ram;
5558 ns->ln_flags |= LF_INCONSISTENT;
5564 static int lfsck_namespace_assistant_handler_p1(const struct lu_env *env,
5565 struct lfsck_component *com,
5566 struct lfsck_assistant_req *lar)
5568 struct lfsck_thread_info *info = lfsck_env_info(env);
5569 struct lu_attr *la = &info->lti_la;
5570 struct lfsck_instance *lfsck = com->lc_lfsck;
5571 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5572 struct lfsck_namespace *ns = com->lc_file_ram;
5573 struct lfsck_assistant_data *lad = com->lc_data;
5574 struct linkea_data ldata = { NULL };
5575 const struct lu_name *cname;
5576 struct thandle *handle = NULL;
5577 struct lfsck_namespace_req *lnr =
5578 container_of(lar, struct lfsck_namespace_req, lnr_lar);
5579 struct dt_object *dir = NULL;
5580 struct dt_object *obj = NULL;
5581 struct lfsck_assistant_object *lso = lar->lar_parent;
5582 const struct lu_fid *pfid = &lso->lso_fid;
5583 struct dt_device *dev = NULL;
5584 struct lustre_handle lh = { 0 };
5585 bool repaired = false;
5586 bool dtlocked = false;
5587 bool remove = false;
5588 bool newdata = false;
5590 bool bad_hash = false;
5591 bool bad_linkea = false;
5595 enum lfsck_namespace_inconsistency_type type = LNIT_NONE;
5602 if (lnr->lnr_attr & LUDA_UPGRADE) {
5603 ns->ln_flags |= LF_UPGRADE;
5604 ns->ln_dirent_repaired++;
5606 } else if (lnr->lnr_attr & LUDA_REPAIR) {
5607 ns->ln_flags |= LF_INCONSISTENT;
5608 ns->ln_dirent_repaired++;
5612 if (unlikely(fid_is_zero(&lnr->lnr_fid) &&
5613 strcmp(lnr->lnr_name, dotdot) == 0)) {
5614 rc = lfsck_namespace_trace_update(env, com, pfid,
5615 LNTF_CHECK_PARENT, true);
5620 if (unlikely(!fid_is_sane(&lnr->lnr_fid))) {
5621 CDEBUG(D_LFSCK, "%s: dir scan find invalid FID "DFID
5622 " for the name entry %.*s under "DFID"\n",
5623 lfsck_lfsck2name(lfsck), PFID(&lnr->lnr_fid),
5624 lnr->lnr_namelen, lnr->lnr_name, PFID(pfid));
5626 if (strcmp(lnr->lnr_name, dotdot) != 0)
5627 /* invalid FID means bad name entry, remove it. */
5628 type = LNIT_BAD_DIRENT;
5630 /* If the parent FID is invalid, we cannot remove
5631 * the ".." entry directly. */
5632 rc = lfsck_namespace_trace_update(env, com, pfid,
5633 LNTF_CHECK_PARENT, true);
5638 if (unlikely(lnr->lnr_dir_cookie == MDS_DIR_END_OFF)) {
5639 rc = lfsck_namespace_striped_dir_rescan(env, com, lnr);
5644 if (fid_seq_is_dot(fid_seq(&lnr->lnr_fid)))
5647 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
5648 rc = lfsck_namespace_handle_striped_master(env, com, lnr);
5653 idx = lfsck_find_mdt_idx_by_fid(env, lfsck, &lnr->lnr_fid);
5655 GOTO(out, rc = idx);
5657 if (idx == lfsck_dev_idx(lfsck)) {
5658 if (unlikely(strcmp(lnr->lnr_name, dotdot) == 0))
5661 dev = lfsck->li_bottom;
5663 struct lfsck_tgt_desc *ltd;
5665 /* Usually, some local filesystem consistency verification
5666 * tools can guarantee the local namespace tree consistenct.
5667 * So the LFSCK will only verify the remote directory. */
5668 if (unlikely(strcmp(lnr->lnr_name, dotdot) == 0)) {
5669 rc = lfsck_namespace_trace_update(env, com, pfid,
5670 LNTF_CHECK_PARENT, true);
5675 ltd = lfsck_ltd2tgt(&lfsck->li_mdt_descs, idx);
5676 if (unlikely(ltd == NULL)) {
5677 CDEBUG(D_LFSCK, "%s: cannot talk with MDT %x which "
5678 "did not join the namespace LFSCK\n",
5679 lfsck_lfsck2name(lfsck), idx);
5680 lfsck_lad_set_bitmap(env, com, idx);
5682 GOTO(out, rc = -ENODEV);
5688 obj = lfsck_object_find_by_dev(env, dev, &lnr->lnr_fid);
5690 GOTO(out, rc = PTR_ERR(obj));
5692 cname = lfsck_name_get_const(env, lnr->lnr_name, lnr->lnr_namelen);
5693 if (dt_object_exists(obj) == 0) {
5697 dir = lfsck_assistant_object_load(env, lfsck, lso);
5701 GOTO(trace, rc == -ENOENT ? 0 : rc);
5705 rc = lfsck_namespace_check_exist(env, dir, obj, lnr->lnr_name);
5707 if (!lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5708 lnr->lnr_name, lnr->lnr_namelen)) {
5709 type = LNIT_BAD_DIRENT;
5714 type = LNIT_DANGLING;
5715 rc = lfsck_namespace_repair_dangling(env, com, dir,
5724 if (!(bk->lb_param & LPF_DRYRUN) && lad->lad_advance_lock) {
5727 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
5728 MDS_INODELOCK_UPDATE |
5729 MDS_INODELOCK_XATTR, LCK_EX);
5733 handle = lfsck_trans_create(env, dev, lfsck);
5735 GOTO(out, rc = PTR_ERR(handle));
5737 rc = lfsck_declare_namespace_exec_dir(env, obj, handle);
5741 rc = dt_trans_start_local(env, dev, handle);
5745 dt_write_lock(env, obj, 0);
5749 rc = lfsck_links_read(env, obj, &ldata);
5750 if (unlikely(rc == -ENOENT)) {
5751 if (handle != NULL) {
5752 dt_write_unlock(env, obj);
5755 dt_trans_stop(env, dev, handle);
5758 lfsck_ibits_unlock(&lh, LCK_EX);
5761 /* It may happen when the remote object has been removed,
5762 * but the local MDT is not aware of that. */
5764 } else if (rc == 0) {
5765 count = ldata.ld_leh->leh_reccount;
5766 rc = linkea_links_find(&ldata, cname, pfid);
5768 (count == 1 || !S_ISDIR(lfsck_object_type(obj)))) {
5769 if ((lfsck_object_type(obj) & S_IFMT) !=
5771 ns->ln_flags |= LF_INCONSISTENT;
5772 type = LNIT_BAD_TYPE;
5778 /* If the name entry hash does not match the slave striped
5779 * directory, and the name entry does not match also, then
5780 * it is quite possible that name entry is corrupted. */
5781 if (!lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5782 lnr->lnr_name, lnr->lnr_namelen)) {
5783 ns->ln_flags |= LF_INCONSISTENT;
5784 type = LNIT_BAD_DIRENT;
5789 /* If the file type stored in the name entry does not match
5790 * the file type claimed by the object, and the object does
5791 * not recognize the name entry, then it is quite possible
5792 * that the name entry is corrupted. */
5793 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type) {
5794 ns->ln_flags |= LF_INCONSISTENT;
5795 type = LNIT_BAD_DIRENT;
5800 /* For sub-dir object, we cannot make sure whether the sub-dir
5801 * back references the parent via ".." name entry correctly or
5802 * not in the LFSCK first-stage scanning. It may be that the
5803 * (remote) sub-dir ".." name entry has no parent FID after
5804 * file-level backup/restore and its linkEA may be wrong.
5805 * So under such case, we should replace the linkEA according
5806 * to current name entry. But this needs to be done during the
5807 * LFSCK second-stage scanning. The LFSCK will record the name
5808 * entry for further possible using. */
5812 } else if (unlikely(rc == -EINVAL)) {
5813 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type)
5814 type = LNIT_BAD_TYPE;
5817 /* The magic crashed, we are not sure whether there are more
5818 * corrupt data in the linkea, so remove all linkea entries. */
5822 } else if (rc == -ENODATA) {
5823 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type)
5824 type = LNIT_BAD_TYPE;
5831 if (rc == -ENOENT &&
5832 linkea_will_overflow(&ldata, cname)) {
5833 CDEBUG(D_INODE, "No enough space to hold linkea entry '"
5834 DFID": %.*s' at %u\n", PFID(pfid),
5835 cname->ln_namelen, cname->ln_name,
5836 ldata.ld_leh->leh_overflow_time);
5842 if (bk->lb_param & LPF_DRYRUN) {
5844 ns->ln_flags |= LF_UPGRADE;
5846 ns->ln_flags |= LF_INCONSISTENT;
5847 ns->ln_linkea_repaired++;
5853 if (!lustre_handle_is_used(&lh)) {
5861 LASSERT(handle != NULL);
5864 dir = lfsck_assistant_object_load(env, lfsck, lso);
5868 GOTO(stop, rc == -ENOENT ? 0 : rc);
5872 rc = lfsck_namespace_check_exist(env, dir, obj, lnr->lnr_name);
5877 if (!remove && newdata)
5878 ns->ln_flags |= LF_UPGRADE;
5879 else if (remove || !(ns->ln_flags & LF_UPGRADE))
5880 ns->ln_flags |= LF_INCONSISTENT;
5885 rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, handle);
5886 if (rc != 0 && rc != -ENOENT && rc != -ENODATA)
5891 rc = linkea_data_new(&ldata,
5892 &lfsck_env_info(env)->lti_linkea_buf);
5897 rc = linkea_add_buf(&ldata, cname, pfid, false);
5899 rc = lfsck_links_write(env, obj, &ldata, handle);
5903 count = ldata.ld_leh->leh_reccount;
5904 if (!S_ISDIR(lfsck_object_type(obj)) ||
5905 !dt_object_remote(obj)) {
5906 ns->ln_linkea_repaired++;
5916 dt_write_unlock(env, obj);
5918 if (handle != NULL && !IS_ERR(handle))
5919 dt_trans_stop(env, dev, handle);
5922 lfsck_ibits_unlock(&lh, LCK_EX);
5924 if (!name_is_dot_or_dotdot(lnr->lnr_name, lnr->lnr_namelen) &&
5925 !lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5926 lnr->lnr_name, lnr->lnr_namelen) &&
5927 type != LNIT_BAD_DIRENT) {
5928 ns->ln_flags |= LF_INCONSISTENT;
5932 dir = lfsck_assistant_object_load(env, lfsck, lso);
5936 GOTO(trace, rc == -ENOENT ? 0 : rc);
5940 rc = lfsck_namespace_repair_bad_name_hash(env, com, dir,
5941 lnr->lnr_lmv, lnr->lnr_name);
5947 if (type != LNIT_NONE && dir == NULL) {
5948 dir = lfsck_assistant_object_load(env, lfsck, lso);
5952 GOTO(trace, rc == -ENOENT ? 0 : rc);
5959 rc = lfsck_namespace_repair_dirent(env, com, dir,
5960 obj, lnr->lnr_name, lnr->lnr_name,
5961 lnr->lnr_type, true, false);
5965 case LNIT_BAD_DIRENT:
5967 /* XXX: This is a bad dirent, we do not know whether
5968 * the original name entry reference a regular
5969 * file or a directory, then keep the parent's
5970 * nlink count unchanged here. */
5971 rc = lfsck_namespace_repair_dirent(env, com, dir,
5972 obj, lnr->lnr_name, lnr->lnr_name,
5973 lnr->lnr_type, false, false);
5981 if (obj != NULL && count == 1 &&
5982 S_ISREG(lfsck_object_type(obj)))
5983 dt_attr_get(env, obj, la);
5985 /* if new linkea entry is added, the old entry may be stale,
5986 * check it in phase 2. Sigh, linkea check can only be done
5989 if (bad_linkea && !remove && !newdata &&
5990 !dt_object_remote(obj) && count > 1)
5991 rc = lfsck_namespace_trace_update(env, com,
5998 down_write(&com->lc_sem);
6000 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail to handle "
6001 "the entry: "DFID", parent "DFID", name %.*s: rc = %d\n",
6002 lfsck_lfsck2name(lfsck), PFID(&lnr->lnr_fid), PFID(pfid),
6003 lnr->lnr_namelen, lnr->lnr_name, rc);
6005 lfsck_namespace_record_failure(env, lfsck, ns);
6006 if ((rc == -ENOTCONN || rc == -ESHUTDOWN || rc == -EREMCHG ||
6007 rc == -ETIMEDOUT || rc == -EHOSTDOWN ||
6008 rc == -EHOSTUNREACH || rc == -EINPROGRESS) &&
6009 dev != NULL && dev != lfsck->li_bottom)
6010 lfsck_lad_set_bitmap(env, com, idx);
6012 if (!(bk->lb_param & LPF_FAILOUT))
6016 ns->ln_items_repaired++;
6018 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant "
6019 "repaired the entry: "DFID", parent "DFID
6020 ", name %.*s, type %d\n",
6021 lfsck_lfsck2name(lfsck),
6022 PFID(&lnr->lnr_fid), PFID(pfid),
6023 lnr->lnr_namelen, lnr->lnr_name, type);
6027 ns->ln_dangling_repaired++;
6030 ns->ln_bad_type_repaired++;
6032 case LNIT_BAD_DIRENT:
6033 ns->ln_dirent_repaired++;
6039 if (bk->lb_param & LPF_DRYRUN &&
6040 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent))
6041 lfsck_pos_fill(env, lfsck,
6042 &ns->ln_pos_first_inconsistent,
6047 ns->ln_name_hash_repaired++;
6049 /* Not count repeatedly. */
6051 ns->ln_items_repaired++;
6053 CDEBUG(D_LFSCK, "%s: namespace LFSCK "
6054 "assistant repaired the entry: "
6057 lfsck_lfsck2name(lfsck),
6058 PFID(&lnr->lnr_fid), PFID(pfid),
6059 lnr->lnr_namelen, lnr->lnr_name);
6062 if (bk->lb_param & LPF_DRYRUN &&
6063 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent))
6064 lfsck_pos_fill(env, lfsck,
6065 &ns->ln_pos_first_inconsistent,
6072 if (count > 1 || la->la_nlink > 1)
6073 ns->ln_mul_linked_checked++;
6075 up_write(&com->lc_sem);
6077 if (obj != NULL && !IS_ERR(obj))
6078 lfsck_object_put(env, obj);
6080 if (dir != NULL && !IS_ERR(dir))
6081 lfsck_object_put(env, dir);
6083 lad->lad_advance_lock = bad_linkea;
6089 * Handle one orphan under the backend /lost+found directory
6091 * Insert the orphan FID into the namespace LFSCK trace file for further
6092 * processing (via the subsequent namespace LFSCK second-stage scanning).
6093 * At the same time, remove the orphan name entry from backend /lost+found
6094 * directory. There is an interval between the orphan name entry removed
6095 * from the backend /lost+found directory and the orphan FID in the LFSCK
6096 * trace file handled. In such interval, the LFSCK can be reset, then
6097 * all the FIDs recorded in the namespace LFSCK trace file will be dropped.
6098 * To guarantee that the orphans can be found when LFSCK run next time
6099 * without e2fsck again, when remove the orphan name entry, the LFSCK
6100 * will set the orphan's ctime attribute as 1. Since normal applications
6101 * cannot change the object's ctime attribute as 1. Then when LFSCK run
6102 * next time, it can record the object (that ctime is 1) in the namespace
6103 * LFSCK trace file during the first-stage scanning.
6105 * \param[in] env pointer to the thread context
6106 * \param[in] com pointer to the lfsck component
6107 * \param[in] parent pointer to the object for the backend /lost+found
6108 * \param[in] ent pointer to the name entry for the target under the
6109 * backend /lost+found
6111 * \retval positive for repaired
6112 * \retval 0 if needs to repair nothing
6113 * \retval negative error number on failure
6115 static int lfsck_namespace_scan_local_lpf_one(const struct lu_env *env,
6116 struct lfsck_component *com,
6117 struct dt_object *parent,
6118 struct lu_dirent *ent)
6120 struct lfsck_thread_info *info = lfsck_env_info(env);
6121 struct lu_fid *key = &info->lti_fid;
6122 struct lu_attr *la = &info->lti_la;
6123 struct lfsck_instance *lfsck = com->lc_lfsck;
6124 struct dt_object *obj;
6125 struct dt_device *dev = lfsck->li_bottom;
6126 struct dt_object *child = NULL;
6127 struct thandle *th = NULL;
6135 child = lfsck_object_find_by_dev(env, dev, &ent->lde_fid);
6137 RETURN(PTR_ERR(child));
6139 if (!dt_object_exists(child)) {
6140 CDEBUG(D_LFSCK, "%s: lost+found/%s doesn't exist\n",
6141 lfsck_lfsck2name(lfsck), ent->lde_name);
6142 GOTO(out, rc = -ENOENT);
6145 LASSERT(!dt_object_remote(child));
6147 idx = lfsck_sub_trace_file_fid2idx(&ent->lde_fid);
6148 obj = com->lc_sub_trace_objs[idx].lsto_obj;
6149 fid_cpu_to_be(key, &ent->lde_fid);
6150 rc = dt_lookup(env, obj, (struct dt_rec *)&flags,
6151 (const struct dt_key *)key);
6154 flags |= LNTF_CHECK_ORPHAN;
6155 } else if (rc == -ENOENT) {
6156 flags = LNTF_CHECK_ORPHAN;
6161 th = lfsck_trans_create(env, dev, lfsck);
6163 GOTO(out, rc = PTR_ERR(th));
6165 /* a1. remove name entry from backend /lost+found */
6166 rc = dt_declare_delete(env, parent,
6167 (const struct dt_key *)ent->lde_name, th);
6171 if (S_ISDIR(lfsck_object_type(child))) {
6172 /* a2. decrease parent's nlink */
6173 rc = dt_declare_ref_del(env, parent, th);
6179 /* a3. remove child's FID from the LFSCK trace file. */
6180 rc = dt_declare_delete(env, obj,
6181 (const struct dt_key *)key, th);
6185 /* a4. set child's ctime as 1 */
6186 memset(la, 0, sizeof(*la));
6188 la->la_valid = LA_CTIME;
6189 rc = dt_declare_attr_set(env, child, la, th);
6194 /* a5. insert child's FID into the LFSCK trace file. */
6195 rc = dt_declare_insert(env, obj, (const struct dt_rec *)&flags,
6196 (const struct dt_key *)key, th);
6200 rc = dt_trans_start_local(env, dev, th);
6204 /* b1. remove name entry from backend /lost+found */
6205 rc = dt_delete(env, parent, (const struct dt_key *)ent->lde_name, th);
6209 if (S_ISDIR(lfsck_object_type(child))) {
6210 /* b2. decrease parent's nlink */
6211 dt_write_lock(env, parent, 0);
6212 rc = dt_ref_del(env, parent, th);
6213 dt_write_unlock(env, parent);
6219 /* a3. remove child's FID from the LFSCK trace file. */
6220 rc = dt_delete(env, obj, (const struct dt_key *)key, th);
6224 /* b4. set child's ctime as 1 */
6225 rc = dt_attr_set(env, child, la, th);
6230 /* b5. insert child's FID into the LFSCK trace file. */
6231 rc = dt_insert(env, obj, (const struct dt_rec *)&flags,
6232 (const struct dt_key *)key, th);
6234 GOTO(stop, rc = (rc == 0 ? 1 : rc));
6237 dt_trans_stop(env, dev, th);
6240 lfsck_object_put(env, child);
6246 * Handle orphans under the backend /lost+found directory
6248 * Some backend checker, such as e2fsck for ldiskfs may find some orphans
6249 * and put them under the backend /lost+found directory that is invisible
6250 * to client. The LFSCK will scan such directory, for the original client
6251 * visible orphans, add their fids into the namespace LFSCK trace file,
6252 * then the subsenquent namespace LFSCK second-stage scanning can handle
6253 * them as other objects to be double scanned: either move back to normal
6254 * namespace, or to the global visible orphan directory:
6255 * /ROOT/.lustre/lost+found/MDTxxxx/
6257 * \param[in] env pointer to the thread context
6258 * \param[in] com pointer to the lfsck component
6260 static void lfsck_namespace_scan_local_lpf(const struct lu_env *env,
6261 struct lfsck_component *com)
6263 struct lfsck_thread_info *info = lfsck_env_info(env);
6264 struct lu_dirent *ent =
6265 (struct lu_dirent *)info->lti_key;
6266 struct lu_seq_range *range = &info->lti_range;
6267 struct lfsck_instance *lfsck = com->lc_lfsck;
6268 struct ptlrpc_thread *thread = &lfsck->li_thread;
6269 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
6270 struct lfsck_namespace *ns = com->lc_file_ram;
6271 struct dt_object *parent;
6272 const struct dt_it_ops *iops;
6274 struct seq_server_site *ss = lfsck_dev_site(lfsck);
6276 __u32 idx = lfsck_dev_idx(lfsck);
6281 parent = lfsck_object_find_by_dev(env, lfsck->li_bottom,
6282 &LU_BACKEND_LPF_FID);
6283 if (IS_ERR(parent)) {
6284 CERROR("%s: fail to find backend /lost+found: rc = %ld\n",
6285 lfsck_lfsck2name(lfsck), PTR_ERR(parent));
6289 /* It is normal that the /lost+found does not exist for ZFS backend. */
6290 if (!dt_object_exists(parent))
6293 if (unlikely(!dt_try_as_dir(env, parent, true)))
6294 GOTO(out, rc = -ENOTDIR);
6296 CDEBUG(D_LFSCK, "%s: start to scan backend /lost+found\n",
6297 lfsck_lfsck2name(lfsck));
6299 com->lc_new_scanned = 0;
6300 iops = &parent->do_index_ops->dio_it;
6301 di = iops->init(env, parent, LUDA_64BITHASH | LUDA_TYPE);
6303 GOTO(out, rc = PTR_ERR(di));
6305 rc = iops->load(env, di, 0);
6307 rc = iops->next(env, di);
6312 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
6313 unlikely(!thread_is_running(thread)))
6316 rc = iops->rec(env, di, (struct dt_rec *)ent,
6317 LUDA_64BITHASH | LUDA_TYPE);
6319 rc = lfsck_unpack_ent(ent, &cookie, &type);
6321 if (unlikely(rc != 0)) {
6322 CDEBUG(D_LFSCK, "%s: fail to iterate backend "
6323 "/lost+found: rc = %d\n",
6324 lfsck_lfsck2name(lfsck), rc);
6329 /* skip dot and dotdot entries */
6330 if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen))
6333 if (!fid_seq_in_fldb(fid_seq(&ent->lde_fid)))
6336 if (fid_is_norm(&ent->lde_fid)) {
6337 fld_range_set_mdt(range);
6338 rc = fld_local_lookup(env, ss->ss_server_fld,
6339 fid_seq(&ent->lde_fid), range);
6342 } else if (idx != 0) {
6343 /* If the returned FID is IGIF, then there are three
6346 * 1) The object is upgraded from old Lustre-1.8 with
6347 * IGIF assigned to such object.
6348 * 2) The object is a backend local object and is
6349 * invisible to client.
6350 * 3) The object lost its LMV EA, and since there is
6351 * no FID-in-dirent for the orphan in the backend
6352 * /lost+found directory, then the low layer will
6353 * return IGIF for such object.
6355 * For MDTx (x != 0), it is either case 2) or case 3),
6356 * but from the LFSCK view, they are indistinguishable.
6357 * To be safe, the LFSCK will keep it there and report
6358 * some message, then the adminstrator can handle that
6361 * For MDT0, it is more possible the case 1). The LFSCK
6362 * will handle the orphan as an upgraded object. */
6363 CDEBUG(D_LFSCK, "%s: the orphan %.*s with IGIF "DFID
6364 "in the backend /lost+found on the MDT %04x, "
6365 "to be safe, skip it.\n",
6366 lfsck_lfsck2name(lfsck), ent->lde_namelen,
6367 ent->lde_name, PFID(&ent->lde_fid), idx);
6371 rc = lfsck_namespace_scan_local_lpf_one(env, com, parent, ent);
6374 down_write(&com->lc_sem);
6375 com->lc_new_scanned++;
6376 ns->ln_local_lpf_scanned++;
6378 ns->ln_local_lpf_moved++;
6380 ns->ln_local_lpf_skipped++;
6382 ns->ln_local_lpf_failed++;
6383 up_write(&com->lc_sem);
6385 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
6389 lfsck_control_speed_by_self(com);
6390 if (unlikely(!thread_is_running(thread))) {
6395 rc = iops->next(env, di);
6399 iops->fini(env, di);
6404 CDEBUG(D_LFSCK, "%s: stop to scan backend /lost+found: rc = %d\n",
6405 lfsck_lfsck2name(lfsck), rc);
6407 lfsck_object_put(env, parent);
6411 * Rescan the striped directory after the master LMV EA reset.
6413 * Sometimes, the master LMV EA of the striped directory maybe lost, so when
6414 * the namespace LFSCK engine scan the striped directory for the first time,
6415 * it will be regarded as a normal directory. As the LFSCK processing, some
6416 * other LFSCK instance on other MDT will find the shard of this striped dir,
6417 * and find that the master MDT-object of the striped directory lost its LMV
6418 * EA, then such remote LFSCK instance will regenerate the master LMV EA and
6419 * notify the LFSCK instance on this MDT to rescan the striped directory.
6421 * \param[in] env pointer to the thread context
6422 * \param[in] com pointer to the lfsck component
6423 * \param[in] llu the lfsck_lmv_unit that contains the striped directory
6426 * \retval positive number for success
6427 * \retval 0 for LFSCK stopped/paused
6428 * \retval negative error number on failure
6430 static int lfsck_namespace_rescan_striped_dir(const struct lu_env *env,
6431 struct lfsck_component *com,
6432 struct lfsck_lmv_unit *llu)
6434 struct lfsck_thread_info *info = lfsck_env_info(env);
6435 struct lfsck_instance *lfsck = com->lc_lfsck;
6436 struct lfsck_assistant_data *lad = com->lc_data;
6437 struct dt_object *dir;
6438 const struct dt_it_ops *iops;
6440 struct lu_dirent *ent =
6441 (struct lu_dirent *)info->lti_key;
6442 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
6443 struct ptlrpc_thread *thread = &lfsck->li_thread;
6444 struct lfsck_assistant_object *lso = NULL;
6445 struct lfsck_namespace_req *lnr;
6446 struct lfsck_assistant_req *lar;
6451 LASSERT(list_empty(&lad->lad_req_list));
6453 lfsck->li_lmv = &llu->llu_lmv;
6454 lfsck->li_obj_dir = lfsck_object_get(llu->llu_obj);
6455 rc = lfsck_open_dir(env, lfsck, 0);
6459 dir = lfsck->li_obj_dir;
6460 di = lfsck->li_di_dir;
6461 iops = &dir->do_index_ops->dio_it;
6463 rc = iops->rec(env, di, (struct dt_rec *)ent,
6464 lfsck->li_args_dir);
6466 rc = lfsck_unpack_ent(ent, &lfsck->li_cookie_dir,
6470 if (bk->lb_param & LPF_FAILOUT)
6476 if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen))
6480 lso = lfsck_assistant_object_init(env,
6481 lfsck_dto2fid(dir), NULL,
6482 lfsck->li_pos_current.lp_oit_cookie, true);
6484 if (bk->lb_param & LPF_FAILOUT)
6485 GOTO(out, rc = PTR_ERR(lso));
6492 lnr = lfsck_namespace_assistant_req_init(lfsck, lso, ent, type);
6494 if (bk->lb_param & LPF_FAILOUT)
6495 GOTO(out, rc = PTR_ERR(lnr));
6500 lar = &lnr->lnr_lar;
6501 rc = lfsck_namespace_assistant_handler_p1(env, com, lar);
6502 lfsck_namespace_assistant_req_fini(env, lar);
6503 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
6506 if (unlikely(!thread_is_running(thread)))
6510 rc = iops->next(env, di);
6514 if (lso != NULL && !IS_ERR(lso))
6515 lfsck_assistant_object_put(env, lso);
6517 lfsck_close_dir(env, lfsck, rc);
6521 /* The close_dir() may insert a dummy lnr in the lad->lad_req_list. */
6522 if (list_empty(&lad->lad_req_list))
6525 spin_lock(&lad->lad_lock);
6526 lar = list_first_entry(&lad->lad_req_list, struct lfsck_assistant_req,
6528 list_del_init(&lar->lar_list);
6529 spin_unlock(&lad->lad_lock);
6531 rc = lfsck_namespace_assistant_handler_p1(env, com, lar);
6532 lfsck_namespace_assistant_req_fini(env, lar);
6534 RETURN(rc == 0 ? 1 : rc);
6538 lfsck_namespace_double_scan_one_trace_file(const struct lu_env *env,
6539 struct lfsck_component *com,
6540 struct dt_object *obj, bool first)
6542 struct lfsck_instance *lfsck = com->lc_lfsck;
6543 struct ptlrpc_thread *thread = &lfsck->li_thread;
6544 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
6545 struct lfsck_namespace *ns = com->lc_file_ram;
6546 const struct dt_it_ops *iops = &obj->do_index_ops->dio_it;
6547 struct dt_object *target;
6555 di = iops->init(env, obj, 0);
6557 RETURN(PTR_ERR(di));
6560 fid_cpu_to_be(&fid, &ns->ln_fid_latest_scanned_phase2);
6563 rc = iops->get(env, di, (const struct dt_key *)&fid);
6568 /* The start one either has been processed or does not exist,
6570 rc = iops->next(env, di);
6576 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
6577 unlikely(!thread_is_running(thread)))
6580 key = iops->key(env, di);
6589 fid_be_to_cpu(&fid, (const struct lu_fid *)key);
6590 if (!fid_is_sane(&fid)) {
6595 target = lfsck_object_find_bottom(env, lfsck, &fid);
6596 if (IS_ERR(target)) {
6597 rc = PTR_ERR(target);
6601 if (dt_object_exists(target)) {
6602 rc = iops->rec(env, di, (struct dt_rec *)&flags, 0);
6604 rc = lfsck_namespace_double_scan_one(env, com,
6611 lfsck_object_put(env, target);
6614 down_write(&com->lc_sem);
6615 com->lc_new_checked++;
6616 com->lc_new_scanned++;
6618 ns->ln_fid_latest_scanned_phase2 = fid;
6621 ns->ln_objs_repaired_phase2++;
6623 ns->ln_objs_failed_phase2++;
6624 up_write(&com->lc_sem);
6626 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
6629 if (unlikely(com->lc_time_next_checkpoint <=
6630 ktime_get_seconds()) &&
6631 com->lc_new_checked != 0) {
6632 down_write(&com->lc_sem);
6633 ns->ln_run_time_phase2 += ktime_get_seconds() -
6634 com->lc_time_last_checkpoint;
6635 ns->ln_time_last_checkpoint = ktime_get_real_seconds();
6636 ns->ln_objs_checked_phase2 += com->lc_new_checked;
6637 com->lc_new_checked = 0;
6638 lfsck_namespace_store(env, com);
6639 up_write(&com->lc_sem);
6641 com->lc_time_last_checkpoint = ktime_get_seconds();
6642 com->lc_time_next_checkpoint =
6643 com->lc_time_last_checkpoint +
6644 LFSCK_CHECKPOINT_INTERVAL;
6647 lfsck_control_speed_by_self(com);
6648 if (unlikely(!thread_is_running(thread)))
6651 rc = iops->next(env, di);
6660 iops->fini(env, di);
6665 static int lfsck_namespace_assistant_handler_p2(const struct lu_env *env,
6666 struct lfsck_component *com)
6668 struct lfsck_instance *lfsck = com->lc_lfsck;
6669 struct lfsck_namespace *ns = com->lc_file_ram;
6674 while (!list_empty(&lfsck->li_list_lmv)) {
6675 struct lfsck_lmv_unit *llu;
6677 spin_lock(&lfsck->li_lock);
6678 llu = list_first_entry(&lfsck->li_list_lmv,
6679 struct lfsck_lmv_unit, llu_link);
6680 list_del_init(&llu->llu_link);
6681 spin_unlock(&lfsck->li_lock);
6683 rc = lfsck_namespace_rescan_striped_dir(env, com, llu);
6688 CDEBUG(D_LFSCK, "%s: namespace LFSCK phase2 scan start\n",
6689 lfsck_lfsck2name(lfsck));
6691 lfsck_namespace_scan_local_lpf(env, com);
6693 com->lc_new_checked = 0;
6694 com->lc_new_scanned = 0;
6695 com->lc_time_last_checkpoint = ktime_get_seconds();
6696 com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
6697 LFSCK_CHECKPOINT_INTERVAL;
6699 i = lfsck_sub_trace_file_fid2idx(&ns->ln_fid_latest_scanned_phase2);
6700 rc = lfsck_namespace_double_scan_one_trace_file(env, com,
6701 com->lc_sub_trace_objs[i].lsto_obj, true);
6702 while (rc > 0 && ++i < LFSCK_STF_COUNT)
6703 rc = lfsck_namespace_double_scan_one_trace_file(env, com,
6704 com->lc_sub_trace_objs[i].lsto_obj, false);
6706 CDEBUG(D_LFSCK, "%s: namespace LFSCK phase2 scan stop at the No. %d "
6707 "trace file: rc = %d\n", lfsck_lfsck2name(lfsck), i, rc);
6712 static void lfsck_namespace_assistant_fill_pos(const struct lu_env *env,
6713 struct lfsck_component *com,
6714 struct lfsck_position *pos)
6716 struct lfsck_assistant_data *lad = com->lc_data;
6717 struct lfsck_namespace_req *lnr;
6719 if (((struct lfsck_namespace *)(com->lc_file_ram))->ln_status !=
6723 if (list_empty(&lad->lad_req_list))
6726 lnr = list_first_entry(&lad->lad_req_list,
6727 struct lfsck_namespace_req,
6729 pos->lp_oit_cookie = lnr->lnr_lar.lar_parent->lso_oit_cookie;
6730 pos->lp_dir_cookie = lnr->lnr_dir_cookie - 1;
6731 pos->lp_dir_parent = lnr->lnr_lar.lar_parent->lso_fid;
6734 static int lfsck_namespace_double_scan_result(const struct lu_env *env,
6735 struct lfsck_component *com,
6738 struct lfsck_instance *lfsck = com->lc_lfsck;
6739 struct lfsck_namespace *ns = com->lc_file_ram;
6741 down_write(&com->lc_sem);
6742 ns->ln_run_time_phase2 += ktime_get_seconds() -
6743 com->lc_time_last_checkpoint;
6744 ns->ln_time_last_checkpoint = ktime_get_real_seconds();
6745 ns->ln_objs_checked_phase2 += com->lc_new_checked;
6746 com->lc_new_checked = 0;
6749 if (ns->ln_flags & LF_INCOMPLETE)
6750 ns->ln_status = LS_PARTIAL;
6752 ns->ln_status = LS_COMPLETED;
6753 ns->ln_flags &= ~LF_SCANNED_ONCE;
6754 if (!(lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN))
6755 ns->ln_flags &= ~LF_INCONSISTENT;
6756 ns->ln_time_last_complete = ns->ln_time_last_checkpoint;
6757 ns->ln_success_count++;
6758 } else if (rc == 0) {
6759 if (lfsck->li_status != 0)
6760 ns->ln_status = lfsck->li_status;
6762 ns->ln_status = LS_STOPPED;
6764 ns->ln_status = LS_FAILED;
6767 rc = lfsck_namespace_store(env, com);
6768 up_write(&com->lc_sem);
6774 lfsck_namespace_assistant_sync_failures_interpret(const struct lu_env *env,
6775 struct ptlrpc_request *req,
6779 struct lfsck_async_interpret_args *laia = args;
6780 struct lfsck_tgt_desc *ltd = laia->laia_ltd;
6782 ltd->ltd_synced_failures = 1;
6789 * Notify remote LFSCK instances about former failures.
6791 * The local LFSCK instance has recorded which MDTs have ever failed to respond
6792 * some LFSCK verification requests (maybe because of network issues or the MDT
6793 * itself trouble). During the respond gap the MDT may missed some name entries
6794 * verification, then the MDT cannot know whether related MDT-objects have been
6795 * referenced by related name entries or not, then in the second-stage scanning,
6796 * these MDT-objects will be regarded as orphan, if the MDT-object contains bad
6797 * linkEA for back reference, then it will misguide the LFSCK to generate wrong
6798 * name entry for repairing the orphan.
6800 * To avoid above trouble, when layout LFSCK finishes the first-stage scanning,
6801 * it will scan the bitmap for the ever failed MDTs, and notify them that they
6802 * have ever missed some name entries verification and should skip the handling
6803 * for orphan MDT-objects.
6805 * \param[in] env pointer to the thread context
6806 * \param[in] com pointer to the lfsck component
6807 * \param[in] lr pointer to the lfsck request
6809 static void lfsck_namespace_assistant_sync_failures(const struct lu_env *env,
6810 struct lfsck_component *com,
6811 struct lfsck_request *lr)
6813 struct lfsck_async_interpret_args *laia =
6814 &lfsck_env_info(env)->lti_laia2;
6815 struct lfsck_assistant_data *lad = com->lc_data;
6816 struct lfsck_namespace *ns = com->lc_file_ram;
6817 struct lfsck_instance *lfsck = com->lc_lfsck;
6818 struct lfsck_tgt_descs *ltds = &lfsck->li_mdt_descs;
6819 struct lfsck_tgt_desc *ltd;
6820 struct ptlrpc_request_set *set;
6825 if (!test_bit(LAD_INCOMPLETE, &lad->lad_flags))
6828 set = ptlrpc_prep_set();
6830 GOTO(out, rc = -ENOMEM);
6832 lr->lr_flags2 = ns->ln_flags | LF_INCOMPLETE;
6833 memset(laia, 0, sizeof(*laia));
6834 lad->lad_touch_gen++;
6836 down_read(<ds->ltd_rw_sem);
6837 for_each_set_bit(idx, lad->lad_bitmap, lad->lad_bitmap_count) {
6838 ltd = lfsck_ltd2tgt(ltds, idx);
6842 laia->laia_ltd = ltd;
6843 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
6844 lfsck_namespace_assistant_sync_failures_interpret,
6845 laia, LFSCK_NOTIFY);
6847 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail "
6848 "to sync failure with MDT %x: rc = %d\n",
6849 lfsck_lfsck2name(lfsck), ltd->ltd_index, rc);
6851 up_read(<ds->ltd_rw_sem);
6853 rc = ptlrpc_set_wait(env, set);
6854 ptlrpc_set_destroy(set);
6860 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail "
6861 "to sync failure with MDTs, and related MDTs "
6862 "may handle orphan improperly: rc = %d\n",
6863 lfsck_lfsck2name(lfsck), rc);
6868 const struct lfsck_assistant_operations lfsck_namespace_assistant_ops = {
6869 .la_handler_p1 = lfsck_namespace_assistant_handler_p1,
6870 .la_handler_p2 = lfsck_namespace_assistant_handler_p2,
6871 .la_fill_pos = lfsck_namespace_assistant_fill_pos,
6872 .la_double_scan_result = lfsck_namespace_double_scan_result,
6873 .la_req_fini = lfsck_namespace_assistant_req_fini,
6874 .la_sync_failures = lfsck_namespace_assistant_sync_failures,
6878 * Verify the specified linkEA entry for the given directory object.
6879 * If the object has no such linkEA entry or it has more other linkEA
6880 * entries, then re-generate the linkEA with the given information.
6882 * \param[in] env pointer to the thread context
6883 * \param[in] obj pointer to the dt_object to be handled
6884 * \param[in] cname the name for the child in the parent directory
6885 * \param[in] pfid the parent directory's FID for the linkEA
6887 * \retval 0 for success
6888 * \retval negative error number on failure
6890 int lfsck_verify_linkea(const struct lu_env *env, struct lfsck_instance *lfsck,
6891 struct dt_object *obj, const struct lu_name *cname,
6892 const struct lu_fid *pfid)
6894 struct dt_device *dev = lfsck_obj2dev(obj);
6895 struct linkea_data ldata = { NULL };
6896 struct lu_buf linkea_buf;
6899 int fl = LU_XATTR_CREATE;
6904 if (!dt_object_exists(obj))
6907 if (!S_ISDIR(lfsck_object_type(obj)))
6910 rc = lfsck_links_read_with_rec(env, obj, &ldata);
6911 if (rc == -ENODATA) {
6913 } else if (rc == 0) {
6914 fl = LU_XATTR_REPLACE;
6915 if (ldata.ld_leh->leh_reccount != 1) {
6918 rc = linkea_links_find(&ldata, cname, pfid);
6927 rc = linkea_links_new(&ldata, &lfsck_env_info(env)->lti_linkea_buf,
6932 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
6933 ldata.ld_leh->leh_len);
6934 th = lfsck_trans_create(env, dev, lfsck);
6936 RETURN(PTR_ERR(th));
6938 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
6939 XATTR_NAME_LINK, fl, th);
6943 rc = dt_trans_start_local(env, dev, th);
6947 dt_write_lock(env, obj, 0);
6948 rc = dt_xattr_set(env, obj, &linkea_buf,
6949 XATTR_NAME_LINK, fl, th);
6950 dt_write_unlock(env, obj);
6955 dt_trans_stop(env, dev, th);
6960 * Get the name and parent directory's FID from the first linkEA entry.
6962 * \param[in] env pointer to the thread context
6963 * \param[in] obj pointer to the object which get linkEA from
6964 * \param[out] name pointer to the buffer to hold the name
6965 * in the first linkEA entry
6966 * \param[out] pfid pointer to the buffer to hold the parent
6967 * directory's FID in the first linkEA entry
6969 * \retval 0 for success
6970 * \retval negative error number on failure
6972 int lfsck_links_get_first(const struct lu_env *env, struct dt_object *obj,
6973 char *name, struct lu_fid *pfid)
6975 struct lu_name *cname = &lfsck_env_info(env)->lti_name;
6976 struct linkea_data ldata = { NULL };
6979 rc = lfsck_links_read_with_rec(env, obj, &ldata);
6983 linkea_first_entry(&ldata);
6984 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, cname, pfid);
6985 if (!linkea_entry_is_valid(&ldata, cname, pfid))
6988 /* To guarantee the 'name' is terminated with '0'. */
6989 memcpy(name, cname->ln_name, cname->ln_namelen);
6990 name[cname->ln_namelen] = 0;
6996 * Update the object's name entry with the given FID.
6998 * \param[in] env pointer to the thread context
6999 * \param[in] lfsck pointer to the lfsck instance
7000 * \param[in] dir pointer to the directory that holds
7002 * \param[in] name the name for the entry to be updated
7003 * \param[in] fid the new FID for the name entry referenced
7004 * \param[in] type the type for the name entry to be updated
7006 * \retval 0 for success
7007 * \retval negative error number on failure
7009 int lfsck_update_name_entry(const struct lu_env *env,
7010 struct lfsck_instance *lfsck,
7011 struct dt_object *dir, const char *name,
7012 const struct lu_fid *fid, __u32 type)
7014 struct lfsck_thread_info *info = lfsck_env_info(env);
7015 struct dt_insert_rec *rec = &info->lti_dt_rec;
7016 struct lfsck_lock_handle *llh = &info->lti_llh;
7017 struct dt_device *dev = lfsck_obj2dev(dir);
7023 rc = lfsck_lock(env, lfsck, dir, name, llh,
7024 MDS_INODELOCK_UPDATE, LCK_PW);
7028 th = lfsck_trans_create(env, dev, lfsck);
7030 GOTO(unlock, rc = PTR_ERR(th));
7032 rc = dt_declare_delete(env, dir, (const struct dt_key *)name, th);
7036 rec->rec_type = type;
7038 rc = dt_declare_insert(env, dir, (const struct dt_rec *)rec,
7039 (const struct dt_key *)name, th);
7043 rc = dt_declare_ref_add(env, dir, th);
7047 rc = dt_trans_start_local(env, dev, th);
7051 rc = dt_delete(env, dir, (const struct dt_key *)name, th);
7052 if (rc == -ENOENT) {
7060 rc = dt_insert(env, dir, (const struct dt_rec *)rec,
7061 (const struct dt_key *)name, th);
7062 if (rc == 0 && S_ISDIR(type) && !exists) {
7063 dt_write_lock(env, dir, 0);
7064 rc = dt_ref_add(env, dir, th);
7065 dt_write_unlock(env, dir);
7071 dt_trans_stop(env, dev, th);
7075 CDEBUG(D_LFSCK, "%s: update name entry "DFID"/%s with the FID "DFID
7076 " and the type %o: rc = %d\n", lfsck_lfsck2name(lfsck),
7077 PFID(lfsck_dto2fid(dir)), name, PFID(fid), type, rc);
7082 int lfsck_namespace_setup(const struct lu_env *env,
7083 struct lfsck_instance *lfsck)
7085 struct lfsck_component *com;
7086 struct lfsck_namespace *ns;
7087 struct dt_object *root = NULL;
7088 struct dt_object *obj;
7093 LASSERT(lfsck->li_master);
7099 INIT_LIST_HEAD(&com->lc_link);
7100 INIT_LIST_HEAD(&com->lc_link_dir);
7101 init_rwsem(&com->lc_sem);
7102 atomic_set(&com->lc_ref, 1);
7103 com->lc_lfsck = lfsck;
7104 com->lc_type = LFSCK_TYPE_NAMESPACE;
7105 com->lc_ops = &lfsck_namespace_ops;
7106 com->lc_data = lfsck_assistant_data_init(
7107 &lfsck_namespace_assistant_ops,
7109 if (com->lc_data == NULL)
7110 GOTO(out, rc = -ENOMEM);
7112 com->lc_file_size = sizeof(struct lfsck_namespace);
7113 OBD_ALLOC(com->lc_file_ram, com->lc_file_size);
7114 if (com->lc_file_ram == NULL)
7115 GOTO(out, rc = -ENOMEM);
7117 OBD_ALLOC(com->lc_file_disk, com->lc_file_size);
7118 if (com->lc_file_disk == NULL)
7119 GOTO(out, rc = -ENOMEM);
7121 for (i = 0; i < LFSCK_STF_COUNT; i++)
7122 mutex_init(&com->lc_sub_trace_objs[i].lsto_mutex);
7124 root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
7126 GOTO(out, rc = PTR_ERR(root));
7128 if (unlikely(!dt_try_as_dir(env, root, true)))
7129 GOTO(out, rc = -ENOTDIR);
7131 obj = local_index_find_or_create(env, lfsck->li_los, root,
7133 S_IFREG | S_IRUGO | S_IWUSR,
7134 &dt_lfsck_namespace_features);
7136 GOTO(out, rc = PTR_ERR(obj));
7139 rc = lfsck_namespace_load(env, com);
7140 if (rc == -ENODATA) {
7141 rc = lfsck_namespace_init(env, com);
7142 } else if (rc < 0) {
7143 rc = lfsck_namespace_reset(env, com, true);
7145 rc = lfsck_load_sub_trace_files(env, com,
7146 &dt_lfsck_namespace_features, LFSCK_NAMESPACE, false);
7148 rc = lfsck_namespace_reset(env, com, true);
7153 ns = com->lc_file_ram;
7154 switch (ns->ln_status) {
7159 spin_lock(&lfsck->li_lock);
7160 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
7161 spin_unlock(&lfsck->li_lock);
7164 CERROR("%s: unknown lfsck_namespace status %d\n",
7165 lfsck_lfsck2name(lfsck), ns->ln_status);
7167 case LS_SCANNING_PHASE1:
7168 case LS_SCANNING_PHASE2:
7169 /* No need to store the status to disk right now.
7170 * If the system crashed before the status stored,
7171 * it will be loaded back when next time. */
7172 ns->ln_status = LS_CRASHED;
7176 spin_lock(&lfsck->li_lock);
7177 list_add_tail(&com->lc_link, &lfsck->li_list_scan);
7178 list_add_tail(&com->lc_link_dir, &lfsck->li_list_dir);
7179 spin_unlock(&lfsck->li_lock);
7186 if (root != NULL && !IS_ERR(root))
7187 lfsck_object_put(env, root);
7189 lfsck_component_cleanup(env, com);
7190 CERROR("%s: fail to init namespace LFSCK component: rc = %d\n",
7191 lfsck_lfsck2name(lfsck), rc);