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, 2015, Intel Corporation.
26 * lustre/lfsck/lfsck_namespace.c
28 * Author: Fan, Yong <fan.yong@intel.com>
31 #define DEBUG_SUBSYSTEM S_LFSCK
33 #include <lustre/lustre_idl.h>
34 #include <lu_object.h>
35 #include <dt_object.h>
36 #include <md_object.h>
37 #include <lustre_fid.h>
38 #include <lustre_lib.h>
39 #include <lustre_net.h>
40 #include <lustre/lustre_user.h>
42 #include "lfsck_internal.h"
44 #define LFSCK_NAMESPACE_MAGIC_V1 0xA0629D03
45 #define LFSCK_NAMESPACE_MAGIC_V2 0xA0621A0B
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_V2
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_of0(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 = le32_to_cpu(src->ln_run_time_phase1);
108 dst->ln_run_time_phase2 = le32_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);
169 static void lfsck_namespace_cpu_to_le(struct lfsck_namespace *dst,
170 struct lfsck_namespace *src)
172 dst->ln_magic = cpu_to_le32(src->ln_magic);
173 dst->ln_status = cpu_to_le32(src->ln_status);
174 dst->ln_flags = cpu_to_le32(src->ln_flags);
175 dst->ln_success_count = cpu_to_le32(src->ln_success_count);
176 dst->ln_run_time_phase1 = cpu_to_le32(src->ln_run_time_phase1);
177 dst->ln_run_time_phase2 = cpu_to_le32(src->ln_run_time_phase2);
178 dst->ln_time_last_complete = cpu_to_le64(src->ln_time_last_complete);
179 dst->ln_time_latest_start = cpu_to_le64(src->ln_time_latest_start);
180 dst->ln_time_last_checkpoint =
181 cpu_to_le64(src->ln_time_last_checkpoint);
182 lfsck_position_cpu_to_le(&dst->ln_pos_latest_start,
183 &src->ln_pos_latest_start);
184 lfsck_position_cpu_to_le(&dst->ln_pos_last_checkpoint,
185 &src->ln_pos_last_checkpoint);
186 lfsck_position_cpu_to_le(&dst->ln_pos_first_inconsistent,
187 &src->ln_pos_first_inconsistent);
188 dst->ln_items_checked = cpu_to_le64(src->ln_items_checked);
189 dst->ln_items_repaired = cpu_to_le64(src->ln_items_repaired);
190 dst->ln_items_failed = cpu_to_le64(src->ln_items_failed);
191 dst->ln_dirs_checked = cpu_to_le64(src->ln_dirs_checked);
192 dst->ln_objs_checked_phase2 = cpu_to_le64(src->ln_objs_checked_phase2);
193 dst->ln_objs_repaired_phase2 =
194 cpu_to_le64(src->ln_objs_repaired_phase2);
195 dst->ln_objs_failed_phase2 = cpu_to_le64(src->ln_objs_failed_phase2);
196 dst->ln_objs_nlink_repaired = cpu_to_le64(src->ln_objs_nlink_repaired);
197 fid_cpu_to_le(&dst->ln_fid_latest_scanned_phase2,
198 &src->ln_fid_latest_scanned_phase2);
199 dst->ln_dirent_repaired = cpu_to_le64(src->ln_dirent_repaired);
200 dst->ln_linkea_repaired = cpu_to_le64(src->ln_linkea_repaired);
201 dst->ln_mul_linked_checked = cpu_to_le64(src->ln_mul_linked_checked);
202 dst->ln_mul_linked_repaired = cpu_to_le64(src->ln_mul_linked_repaired);
203 dst->ln_unknown_inconsistency =
204 cpu_to_le64(src->ln_unknown_inconsistency);
205 dst->ln_unmatched_pairs_repaired =
206 cpu_to_le64(src->ln_unmatched_pairs_repaired);
207 dst->ln_dangling_repaired = cpu_to_le64(src->ln_dangling_repaired);
208 dst->ln_mul_ref_repaired = cpu_to_le64(src->ln_mul_ref_repaired);
209 dst->ln_bad_type_repaired = cpu_to_le64(src->ln_bad_type_repaired);
210 dst->ln_lost_dirent_repaired =
211 cpu_to_le64(src->ln_lost_dirent_repaired);
212 dst->ln_striped_dirs_scanned =
213 cpu_to_le64(src->ln_striped_dirs_scanned);
214 dst->ln_striped_dirs_repaired =
215 cpu_to_le64(src->ln_striped_dirs_repaired);
216 dst->ln_striped_dirs_failed =
217 cpu_to_le64(src->ln_striped_dirs_failed);
218 dst->ln_striped_dirs_disabled =
219 cpu_to_le64(src->ln_striped_dirs_disabled);
220 dst->ln_striped_dirs_skipped =
221 cpu_to_le64(src->ln_striped_dirs_skipped);
222 dst->ln_striped_shards_scanned =
223 cpu_to_le64(src->ln_striped_shards_scanned);
224 dst->ln_striped_shards_repaired =
225 cpu_to_le64(src->ln_striped_shards_repaired);
226 dst->ln_striped_shards_failed =
227 cpu_to_le64(src->ln_striped_shards_failed);
228 dst->ln_striped_shards_skipped =
229 cpu_to_le64(src->ln_striped_shards_skipped);
230 dst->ln_name_hash_repaired = cpu_to_le64(src->ln_name_hash_repaired);
231 dst->ln_local_lpf_scanned = cpu_to_le64(src->ln_local_lpf_scanned);
232 dst->ln_local_lpf_moved = cpu_to_le64(src->ln_local_lpf_moved);
233 dst->ln_local_lpf_skipped = cpu_to_le64(src->ln_local_lpf_skipped);
234 dst->ln_local_lpf_failed = cpu_to_le64(src->ln_local_lpf_failed);
235 dst->ln_bitmap_size = cpu_to_le32(src->ln_bitmap_size);
238 static void lfsck_namespace_record_failure(const struct lu_env *env,
239 struct lfsck_instance *lfsck,
240 struct lfsck_namespace *ns)
242 struct lfsck_position pos;
244 ns->ln_items_failed++;
245 lfsck_pos_fill(env, lfsck, &pos, false);
246 if (lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent) ||
247 lfsck_pos_is_eq(&pos, &ns->ln_pos_first_inconsistent) < 0) {
248 ns->ln_pos_first_inconsistent = pos;
250 CDEBUG(D_LFSCK, "%s: namespace LFSCK hit first non-repaired "
251 "inconsistency at the pos ["LPU64", "DFID", "LPX64"]\n",
252 lfsck_lfsck2name(lfsck),
253 ns->ln_pos_first_inconsistent.lp_oit_cookie,
254 PFID(&ns->ln_pos_first_inconsistent.lp_dir_parent),
255 ns->ln_pos_first_inconsistent.lp_dir_cookie);
260 * Load the MDT bitmap from the lfsck_namespace trace file.
262 * \param[in] env pointer to the thread context
263 * \param[in] com pointer to the lfsck component
265 * \retval 0 for success
266 * \retval negative error number on failure or data corruption
268 static int lfsck_namespace_load_bitmap(const struct lu_env *env,
269 struct lfsck_component *com)
271 struct dt_object *obj = com->lc_obj;
272 struct lfsck_assistant_data *lad = com->lc_data;
273 struct lfsck_namespace *ns = com->lc_file_ram;
274 cfs_bitmap_t *bitmap = lad->lad_bitmap;
280 if (com->lc_lfsck->li_mdt_descs.ltd_tgts_bitmap->size >
282 nbits = com->lc_lfsck->li_mdt_descs.ltd_tgts_bitmap->size;
284 nbits = ns->ln_bitmap_size;
286 if (unlikely(nbits < BITS_PER_LONG))
287 nbits = BITS_PER_LONG;
289 if (nbits > bitmap->size) {
290 __u32 new_bits = bitmap->size;
291 cfs_bitmap_t *new_bitmap;
293 while (new_bits < nbits)
296 new_bitmap = CFS_ALLOCATE_BITMAP(new_bits);
297 if (new_bitmap == NULL)
300 lad->lad_bitmap = new_bitmap;
301 CFS_FREE_BITMAP(bitmap);
305 if (ns->ln_bitmap_size == 0) {
306 lad->lad_incomplete = 0;
307 CFS_RESET_BITMAP(bitmap);
312 size = (ns->ln_bitmap_size + 7) >> 3;
313 rc = dt_xattr_get(env, obj,
314 lfsck_buf_get(env, bitmap->data, size),
315 XATTR_NAME_LFSCK_BITMAP);
317 RETURN(rc >= 0 ? -EINVAL : rc);
319 if (cfs_bitmap_check_empty(bitmap))
320 lad->lad_incomplete = 0;
322 lad->lad_incomplete = 1;
328 * Load namespace LFSCK statistics information from the trace file.
330 * For old release (Lustre-2.6 or older), the statistics information was
331 * stored as XATTR_NAME_LFSCK_NAMESPACE_OLD EA. But in Lustre-2.7, we need
332 * more statistics information. To avoid confusing old MDT when downgrade,
333 * Lustre-2.7 stores the namespace LFSCK statistics information as new
334 * XATTR_NAME_LFSCK_NAMESPACE EA.
336 * \param[in] env pointer to the thread context
337 * \param[in] com pointer to the lfsck component
339 * \retval 0 for success
340 * \retval negative error number on failure
342 static int lfsck_namespace_load(const struct lu_env *env,
343 struct lfsck_component *com)
345 int len = com->lc_file_size;
348 rc = dt_xattr_get(env, com->lc_obj,
349 lfsck_buf_get(env, com->lc_file_disk, len),
350 XATTR_NAME_LFSCK_NAMESPACE);
352 struct lfsck_namespace *ns = com->lc_file_ram;
354 lfsck_namespace_le_to_cpu(ns,
355 (struct lfsck_namespace *)com->lc_file_disk);
356 if (ns->ln_magic != LFSCK_NAMESPACE_MAGIC) {
357 CDEBUG(D_LFSCK, "%s: invalid lfsck_namespace magic "
358 "%#x != %#x\n", lfsck_lfsck2name(com->lc_lfsck),
359 ns->ln_magic, LFSCK_NAMESPACE_MAGIC);
364 } else if (rc != -ENODATA) {
365 CDEBUG(D_LFSCK, "%s: fail to load lfsck_namespace, "
366 "expected = %d: rc = %d\n",
367 lfsck_lfsck2name(com->lc_lfsck), len, rc);
371 /* Check whether it is old trace file or not.
372 * If yes, it should be reset via returning -ESTALE. */
373 rc = dt_xattr_get(env, com->lc_obj,
374 lfsck_buf_get(env, com->lc_file_disk, len),
375 XATTR_NAME_LFSCK_NAMESPACE_OLD);
383 static int lfsck_namespace_store(const struct lu_env *env,
384 struct lfsck_component *com, bool init)
386 struct dt_object *obj = com->lc_obj;
387 struct lfsck_instance *lfsck = com->lc_lfsck;
388 struct lfsck_namespace *ns = com->lc_file_ram;
389 struct lfsck_assistant_data *lad = com->lc_data;
390 struct dt_device *dev = lfsck_obj2dev(obj);
391 cfs_bitmap_t *bitmap = NULL;
392 struct thandle *handle;
394 int len = com->lc_file_size;
396 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 8, 53, 0)
397 struct lu_buf tbuf = { &len, sizeof(len) };
402 bitmap = lad->lad_bitmap;
403 nbits = bitmap->size;
406 LASSERTF((nbits & 7) == 0, "Invalid nbits %u\n", nbits);
409 ns->ln_bitmap_size = nbits;
410 lfsck_namespace_cpu_to_le((struct lfsck_namespace *)com->lc_file_disk,
412 handle = dt_trans_create(env, dev);
414 GOTO(log, rc = PTR_ERR(handle));
416 rc = dt_declare_xattr_set(env, obj,
417 lfsck_buf_get(env, com->lc_file_disk, len),
418 XATTR_NAME_LFSCK_NAMESPACE, 0, handle);
422 if (bitmap != NULL) {
423 rc = dt_declare_xattr_set(env, obj,
424 lfsck_buf_get(env, bitmap->data, nbits >> 3),
425 XATTR_NAME_LFSCK_BITMAP, 0, handle);
430 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 8, 53, 0)
431 /* To be compatible with old Lustre-2.x MDT (x <= 6), generate dummy
432 * XATTR_NAME_LFSCK_NAMESPACE_OLD EA, then when downgrade to Lustre-2.x,
433 * the old LFSCK will find "invalid" XATTR_NAME_LFSCK_NAMESPACE_OLD EA,
434 * then reset the namespace LFSCK trace file. */
436 rc = dt_declare_xattr_set(env, obj, &tbuf,
437 XATTR_NAME_LFSCK_NAMESPACE_OLD,
438 LU_XATTR_CREATE, handle);
444 rc = dt_trans_start_local(env, dev, handle);
448 rc = dt_xattr_set(env, obj,
449 lfsck_buf_get(env, com->lc_file_disk, len),
450 XATTR_NAME_LFSCK_NAMESPACE, 0, handle);
451 if (rc == 0 && bitmap != NULL)
452 rc = dt_xattr_set(env, obj,
453 lfsck_buf_get(env, bitmap->data, nbits >> 3),
454 XATTR_NAME_LFSCK_BITMAP, 0, handle);
456 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 8, 53, 0)
458 rc = dt_xattr_set(env, obj, &tbuf,
459 XATTR_NAME_LFSCK_NAMESPACE_OLD,
460 LU_XATTR_CREATE, handle);
466 dt_trans_stop(env, dev, handle);
470 CDEBUG(D_LFSCK, "%s: fail to store lfsck_namespace: rc = %d\n",
471 lfsck_lfsck2name(lfsck), rc);
475 static struct dt_object *
476 lfsck_namespace_load_one_trace_file(const struct lu_env *env,
477 struct lfsck_component *com,
478 struct dt_object *parent,
479 const char *name, bool reset)
481 struct lfsck_instance *lfsck = com->lc_lfsck;
482 struct dt_object *obj;
486 rc = local_object_unlink(env, lfsck->li_bottom, parent, name);
487 if (rc != 0 && rc != -ENOENT)
491 obj = local_index_find_or_create(env, lfsck->li_los, parent, name,
492 S_IFREG | S_IRUGO | S_IWUSR,
498 static int lfsck_namespace_load_sub_trace_files(const struct lu_env *env,
499 struct lfsck_component *com,
502 char *name = lfsck_env_info(env)->lti_key;
503 struct lfsck_sub_trace_obj *lsto;
504 struct dt_object *obj;
508 for (i = 0, lsto = &com->lc_sub_trace_objs[0];
509 i < LFSCK_STF_COUNT; i++, lsto++) {
510 snprintf(name, NAME_MAX, "%s_%02d", LFSCK_NAMESPACE, i);
511 if (lsto->lsto_obj != NULL) {
515 lfsck_object_put(env, lsto->lsto_obj);
516 lsto->lsto_obj = NULL;
519 obj = lfsck_namespace_load_one_trace_file(env, com,
520 com->lc_lfsck->li_lfsck_dir, name, reset);
524 lsto->lsto_obj = obj;
525 rc = obj->do_ops->do_index_try(env, obj, &dt_lfsck_features);
533 static int lfsck_namespace_init(const struct lu_env *env,
534 struct lfsck_component *com)
536 struct lfsck_namespace *ns = com->lc_file_ram;
539 memset(ns, 0, sizeof(*ns));
540 ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
541 ns->ln_status = LS_INIT;
542 down_write(&com->lc_sem);
543 rc = lfsck_namespace_store(env, com, true);
544 up_write(&com->lc_sem);
546 rc = lfsck_namespace_load_sub_trace_files(env, com, true);
552 * Update the namespace LFSCK trace file for the given @fid
554 * \param[in] env pointer to the thread context
555 * \param[in] com pointer to the lfsck component
556 * \param[in] fid the fid which flags to be updated in the lfsck
558 * \param[in] add true if add new flags, otherwise remove flags
560 * \retval 0 for success or nothing to be done
561 * \retval negative error number on failure
563 int lfsck_namespace_trace_update(const struct lu_env *env,
564 struct lfsck_component *com,
565 const struct lu_fid *fid,
566 const __u8 flags, bool add)
568 struct lfsck_instance *lfsck = com->lc_lfsck;
569 struct dt_object *obj;
570 struct lu_fid *key = &lfsck_env_info(env)->lti_fid3;
571 struct dt_device *dev;
572 struct thandle *th = NULL;
581 if (unlikely(!fid_is_sane(fid)))
584 idx = lfsck_sub_trace_file_fid2idx(fid);
585 obj = com->lc_sub_trace_objs[idx].lsto_obj;
586 dev = lfsck_obj2dev(obj);
587 mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
588 fid_cpu_to_be(key, fid);
589 rc = dt_lookup(env, obj, (struct dt_rec *)&old,
590 (const struct dt_key *)key);
593 GOTO(unlock, rc = 0);
597 } else if (rc == 0) {
599 if ((old & flags) == flags)
600 GOTO(unlock, rc = 0);
604 if ((old & flags) == 0)
605 GOTO(unlock, rc = 0);
613 th = dt_trans_create(env, dev);
615 GOTO(log, rc = PTR_ERR(th));
618 rc = dt_declare_delete(env, obj,
619 (const struct dt_key *)key, th);
625 rc = dt_declare_insert(env, obj,
626 (const struct dt_rec *)&new,
627 (const struct dt_key *)key, th);
632 rc = dt_trans_start_local(env, dev, th);
637 rc = dt_delete(env, obj, (const struct dt_key *)key, th);
643 rc = dt_insert(env, obj, (const struct dt_rec *)&new,
644 (const struct dt_key *)key, th, 1);
652 if (th != NULL && !IS_ERR(th))
653 dt_trans_stop(env, dev, th);
655 CDEBUG(D_LFSCK, "%s: namespace LFSCK %s flags for "DFID" in the "
656 "trace file, flags %x, old %x, new %x: rc = %d\n",
657 lfsck_lfsck2name(lfsck), add ? "add" : "del", PFID(fid),
658 (__u32)flags, (__u32)old, (__u32)new, rc);
661 mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
666 int lfsck_namespace_check_exist(const struct lu_env *env,
667 struct dt_object *dir,
668 struct dt_object *obj, const char *name)
670 struct lu_fid *fid = &lfsck_env_info(env)->lti_fid;
674 if (unlikely(lfsck_is_dead_obj(obj)))
675 RETURN(LFSCK_NAMEENTRY_DEAD);
677 rc = dt_lookup(env, dir, (struct dt_rec *)fid,
678 (const struct dt_key *)name);
680 RETURN(LFSCK_NAMEENTRY_REMOVED);
685 if (!lu_fid_eq(fid, lfsck_dto2fid(obj)))
686 RETURN(LFSCK_NAMEENTRY_RECREATED);
691 static int lfsck_declare_namespace_exec_dir(const struct lu_env *env,
692 struct dt_object *obj,
693 struct thandle *handle)
697 /* For destroying all invalid linkEA entries. */
698 rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, handle);
702 /* For insert new linkEA entry. */
703 rc = dt_declare_xattr_set(env, obj,
704 lfsck_buf_get_const(env, NULL, DEFAULT_LINKEA_SIZE),
705 XATTR_NAME_LINK, 0, handle);
709 int __lfsck_links_read(const struct lu_env *env, struct dt_object *obj,
710 struct linkea_data *ldata)
714 if (ldata->ld_buf->lb_buf == NULL)
717 if (!dt_object_exists(obj))
720 rc = dt_xattr_get(env, obj, ldata->ld_buf, XATTR_NAME_LINK);
722 /* Buf was too small, figure out what we need. */
723 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LINK);
727 lu_buf_realloc(ldata->ld_buf, rc);
728 if (ldata->ld_buf->lb_buf == NULL)
731 rc = dt_xattr_get(env, obj, ldata->ld_buf, XATTR_NAME_LINK);
735 rc = linkea_init(ldata);
741 * Remove linkEA for the given object.
743 * The caller should take the ldlm lock before the calling.
745 * \param[in] env pointer to the thread context
746 * \param[in] com pointer to the lfsck component
747 * \param[in] obj pointer to the dt_object to be handled
749 * \retval 0 for repaired cases
750 * \retval negative error number on failure
752 static int lfsck_namespace_links_remove(const struct lu_env *env,
753 struct lfsck_component *com,
754 struct dt_object *obj)
756 struct lfsck_instance *lfsck = com->lc_lfsck;
757 struct dt_device *dev = lfsck_obj2dev(obj);
758 struct thandle *th = NULL;
762 LASSERT(dt_object_remote(obj) == 0);
764 th = dt_trans_create(env, dev);
766 GOTO(log, rc = PTR_ERR(th));
768 rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, th);
772 rc = dt_trans_start_local(env, dev, th);
776 dt_write_lock(env, obj, 0);
777 if (unlikely(lfsck_is_dead_obj(obj)))
778 GOTO(unlock, rc = -ENOENT);
780 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
781 GOTO(unlock, rc = 0);
783 rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, th);
788 dt_write_unlock(env, obj);
791 dt_trans_stop(env, dev, th);
794 CDEBUG(D_LFSCK, "%s: namespace LFSCK remove invalid linkEA "
795 "for the object "DFID": rc = %d\n",
796 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
799 struct lfsck_namespace *ns = com->lc_file_ram;
801 ns->ln_flags |= LF_INCONSISTENT;
807 static int lfsck_links_write(const struct lu_env *env, struct dt_object *obj,
808 struct linkea_data *ldata, struct thandle *handle)
810 const struct lu_buf *buf = lfsck_buf_get_const(env,
811 ldata->ld_buf->lb_buf,
812 ldata->ld_leh->leh_len);
814 return dt_xattr_set(env, obj, buf, XATTR_NAME_LINK, 0, handle);
817 static void lfsck_namespace_unpack_linkea_entry(struct linkea_data *ldata,
818 struct lu_name *cname,
822 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, cname, pfid);
823 /* To guarantee the 'name' is terminated with '0'. */
824 memcpy(buf, cname->ln_name, cname->ln_namelen);
825 buf[cname->ln_namelen] = 0;
826 cname->ln_name = buf;
829 static int lfsck_namespace_filter_linkea_entry(struct linkea_data *ldata,
830 struct lu_name *cname,
834 struct link_ea_entry *oldlee;
838 oldlee = ldata->ld_lee;
839 oldlen = ldata->ld_reclen;
840 linkea_next_entry(ldata);
841 while (ldata->ld_lee != NULL) {
842 ldata->ld_reclen = (ldata->ld_lee->lee_reclen[0] << 8) |
843 ldata->ld_lee->lee_reclen[1];
844 if (unlikely(ldata->ld_reclen == oldlen &&
845 memcmp(ldata->ld_lee, oldlee, oldlen) == 0)) {
850 linkea_del_buf(ldata, cname);
852 linkea_next_entry(ldata);
855 ldata->ld_lee = oldlee;
856 ldata->ld_reclen = oldlen;
862 * Insert orphan into .lustre/lost+found/MDTxxxx/ locally.
864 * Add the specified orphan MDT-object to the .lustre/lost+found/MDTxxxx/
865 * with the given type to generate the name, the detailed rules for name
866 * have been described as following.
868 * The function also generates the linkEA corresponding to the name entry
869 * under the .lustre/lost+found/MDTxxxx/ for the orphan MDT-object.
871 * \param[in] env pointer to the thread context
872 * \param[in] com pointer to the lfsck component
873 * \param[in] orphan pointer to the orphan MDT-object
874 * \param[in] infix additional information for the orphan name, such as
875 * the FID for original
876 * \param[in] type the type for describing why the orphan MDT-object is
877 * created. The rules are as following:
879 * type "D": The MDT-object is a directory, it may knows its parent
880 * but because there is no valid linkEA, the LFSCK cannot
881 * know where to put it back to the namespace.
882 * type "O": The MDT-object has no linkEA, and there is no name
883 * entry that references the MDT-object.
885 * type "S": The orphan MDT-object is a shard of a striped directory
887 * \see lfsck_layout_recreate_parent() for more types.
889 * The orphan name will be like:
890 * ${FID}-${infix}-${type}-${conflict_version}
892 * \param[out] count if some others inserted some linkEA entries by race,
893 * then return the linkEA entries count.
895 * \retval positive number for repaired cases
896 * \retval 0 if needs to repair nothing
897 * \retval negative error number on failure
899 static int lfsck_namespace_insert_orphan(const struct lu_env *env,
900 struct lfsck_component *com,
901 struct dt_object *orphan,
902 const char *infix, const char *type,
905 struct lfsck_thread_info *info = lfsck_env_info(env);
906 struct lu_name *cname = &info->lti_name;
907 struct dt_insert_rec *rec = &info->lti_dt_rec;
908 struct lu_attr *la = &info->lti_la2;
909 const struct lu_fid *cfid = lfsck_dto2fid(orphan);
910 const struct lu_fid *pfid;
912 struct lfsck_instance *lfsck = com->lc_lfsck;
913 struct dt_device *dev = lfsck_obj2dev(orphan);
914 struct dt_object *parent;
915 struct thandle *th = NULL;
916 struct lfsck_lock_handle *pllh = &info->lti_llh;
917 struct lustre_handle clh = { 0 };
918 struct linkea_data ldata = { NULL };
919 struct lu_buf linkea_buf;
926 cname->ln_name = NULL;
927 if (unlikely(lfsck->li_lpf_obj == NULL))
928 GOTO(log, rc = -ENXIO);
930 parent = lfsck->li_lpf_obj;
931 pfid = lfsck_dto2fid(parent);
935 namelen = snprintf(info->lti_key, NAME_MAX, DFID"%s-%s-%d",
936 PFID(cfid), infix, type, idx++);
937 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
938 (const struct dt_key *)info->lti_key);
939 if (rc != 0 && rc != -ENOENT)
942 if (unlikely(rc == 0 && lu_fid_eq(cfid, &tfid)))
944 } while (rc == 0 && !exist);
946 rc = lfsck_lock(env, lfsck, parent, info->lti_key, pllh,
947 MDS_INODELOCK_UPDATE, LCK_PW);
951 /* Re-check whether the name conflict with othrs after taken
953 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
954 (const struct dt_key *)info->lti_key);
956 if (!lu_fid_eq(cfid, &tfid)) {
963 } else if (rc != -ENOENT) {
969 cname->ln_name = info->lti_key;
970 cname->ln_namelen = namelen;
971 rc = linkea_data_new(&ldata, &info->lti_linkea_buf2);
975 rc = linkea_add_buf(&ldata, cname, pfid);
979 rc = lfsck_ibits_lock(env, lfsck, orphan, &clh,
980 MDS_INODELOCK_UPDATE | MDS_INODELOCK_LOOKUP |
981 MDS_INODELOCK_XATTR, LCK_EX);
985 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
986 ldata.ld_leh->leh_len);
987 th = dt_trans_create(env, dev);
989 GOTO(log, rc = PTR_ERR(th));
991 if (S_ISDIR(lfsck_object_type(orphan))) {
992 rc = dt_declare_delete(env, orphan,
993 (const struct dt_key *)dotdot, th);
997 rec->rec_type = S_IFDIR;
999 rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
1000 (const struct dt_key *)dotdot, th);
1005 rc = dt_declare_xattr_set(env, orphan, &linkea_buf,
1006 XATTR_NAME_LINK, 0, th);
1011 rec->rec_type = lfsck_object_type(orphan) & S_IFMT;
1012 rec->rec_fid = cfid;
1013 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1014 (const struct dt_key *)cname->ln_name,
1019 if (S_ISDIR(rec->rec_type)) {
1020 rc = dt_declare_ref_add(env, parent, th);
1026 memset(la, 0, sizeof(*la));
1027 la->la_ctime = cfs_time_current_sec();
1028 la->la_valid = LA_CTIME;
1029 rc = dt_declare_attr_set(env, orphan, la, th);
1033 rc = dt_trans_start_local(env, dev, th);
1037 dt_write_lock(env, orphan, 0);
1038 rc = lfsck_links_read(env, orphan, &ldata);
1039 if (likely((rc == -ENODATA) || (rc == -EINVAL) ||
1040 (rc == 0 && ldata.ld_leh->leh_reccount == 0))) {
1041 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1042 GOTO(unlock, rc = 1);
1044 if (S_ISDIR(lfsck_object_type(orphan))) {
1045 rc = dt_delete(env, orphan,
1046 (const struct dt_key *)dotdot, th);
1050 rec->rec_type = S_IFDIR;
1051 rec->rec_fid = pfid;
1052 rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1053 (const struct dt_key *)dotdot, th, 1);
1058 rc = dt_xattr_set(env, orphan, &linkea_buf, XATTR_NAME_LINK, 0,
1061 if (rc == 0 && count != NULL)
1062 *count = ldata.ld_leh->leh_reccount;
1066 dt_write_unlock(env, orphan);
1068 if (rc == 0 && !exist) {
1069 rec->rec_type = lfsck_object_type(orphan) & S_IFMT;
1070 rec->rec_fid = cfid;
1071 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1072 (const struct dt_key *)cname->ln_name, th, 1);
1073 if (rc == 0 && S_ISDIR(rec->rec_type)) {
1074 dt_write_lock(env, parent, 0);
1075 rc = dt_ref_add(env, parent, th);
1076 dt_write_unlock(env, parent);
1081 rc = dt_attr_set(env, orphan, la, th);
1083 GOTO(stop, rc = (rc == 0 ? 1 : rc));
1086 dt_write_unlock(env, orphan);
1089 dt_trans_stop(env, dev, th);
1092 lfsck_ibits_unlock(&clh, LCK_EX);
1094 CDEBUG(D_LFSCK, "%s: namespace LFSCK insert orphan for the "
1095 "object "DFID", name = %s: rc = %d\n",
1096 lfsck_lfsck2name(lfsck), PFID(cfid),
1097 cname->ln_name != NULL ? cname->ln_name : "<NULL>", rc);
1100 struct lfsck_namespace *ns = com->lc_file_ram;
1102 ns->ln_flags |= LF_INCONSISTENT;
1109 * Add the specified name entry back to namespace.
1111 * If there is a linkEA entry that back references a name entry under
1112 * some parent directory, but such parent directory does not have the
1113 * claimed name entry. On the other hand, the linkEA entries count is
1114 * not larger than the MDT-object's hard link count. Under such case,
1115 * it is quite possible that the name entry is lost. Then the LFSCK
1116 * should add the name entry back to the namespace.
1118 * \param[in] env pointer to the thread context
1119 * \param[in] com pointer to the lfsck component
1120 * \param[in] parent pointer to the directory under which the name entry
1121 * will be inserted into
1122 * \param[in] child pointer to the object referenced by the name entry
1123 * that to be inserted into the parent
1124 * \param[in] name the name for the child in the parent directory
1126 * \retval positive number for repaired cases
1127 * \retval 0 if nothing to be repaired
1128 * \retval negative error number on failure
1130 static int lfsck_namespace_insert_normal(const struct lu_env *env,
1131 struct lfsck_component *com,
1132 struct dt_object *parent,
1133 struct dt_object *child,
1136 struct lfsck_thread_info *info = lfsck_env_info(env);
1137 struct lu_attr *la = &info->lti_la;
1138 struct dt_insert_rec *rec = &info->lti_dt_rec;
1139 struct lfsck_instance *lfsck = com->lc_lfsck;
1140 /* The child and its name may be on different MDTs. */
1141 const struct lu_fid *pfid = lfsck_dto2fid(parent);
1142 const struct lu_fid *cfid = lfsck_dto2fid(child);
1143 struct dt_device *dev = lfsck->li_next;
1144 struct thandle *th = NULL;
1145 struct lfsck_lock_handle *llh = &info->lti_llh;
1149 /* @parent/@child may be based on lfsck->li_bottom,
1150 * but here we need the object based on the lfsck->li_next. */
1152 parent = lfsck_object_locate(dev, parent);
1154 GOTO(log, rc = PTR_ERR(parent));
1156 if (unlikely(!dt_try_as_dir(env, parent)))
1157 GOTO(log, rc = -ENOTDIR);
1159 child = lfsck_object_locate(dev, child);
1161 GOTO(log, rc = PTR_ERR(child));
1163 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1166 rc = lfsck_lock(env, lfsck, parent, name, llh,
1167 MDS_INODELOCK_UPDATE, LCK_PW);
1171 th = dt_trans_create(env, dev);
1173 GOTO(unlock, rc = PTR_ERR(th));
1175 rec->rec_type = lfsck_object_type(child) & S_IFMT;
1176 rec->rec_fid = cfid;
1177 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1178 (const struct dt_key *)name, th);
1182 if (S_ISDIR(rec->rec_type)) {
1183 rc = dt_declare_ref_add(env, parent, th);
1188 memset(la, 0, sizeof(*la));
1189 la->la_ctime = cfs_time_current_sec();
1190 la->la_valid = LA_CTIME;
1191 rc = dt_declare_attr_set(env, parent, la, th);
1195 rc = dt_declare_attr_set(env, child, la, th);
1199 rc = dt_trans_start_local(env, dev, th);
1203 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1204 (const struct dt_key *)name, th, 1);
1208 if (S_ISDIR(rec->rec_type)) {
1209 dt_write_lock(env, parent, 0);
1210 rc = dt_ref_add(env, parent, th);
1211 dt_write_unlock(env, parent);
1216 la->la_ctime = cfs_time_current_sec();
1217 rc = dt_attr_set(env, parent, la, th);
1221 rc = dt_attr_set(env, child, la, th);
1223 GOTO(stop, rc = (rc == 0 ? 1 : rc));
1226 dt_trans_stop(env, dev, th);
1232 CDEBUG(D_LFSCK, "%s: namespace LFSCK insert object "DFID" with "
1233 "the name %s and type %o to the parent "DFID": rc = %d\n",
1234 lfsck_lfsck2name(lfsck), PFID(cfid), name,
1235 lfsck_object_type(child) & S_IFMT, PFID(pfid), rc);
1238 struct lfsck_namespace *ns = com->lc_file_ram;
1240 ns->ln_flags |= LF_INCONSISTENT;
1242 ns->ln_lost_dirent_repaired++;
1249 * Create the specified orphan directory.
1251 * For the case that the parent MDT-object stored in some MDT-object's
1252 * linkEA entry is lost, the LFSCK will re-create the parent object as
1253 * an orphan and insert it into .lustre/lost+found/MDTxxxx/ directory
1254 * with the name ${FID}-P-${conflict_version}.
1256 * \param[in] env pointer to the thread context
1257 * \param[in] com pointer to the lfsck component
1258 * \param[in] orphan pointer to the orphan MDT-object to be created
1259 * \param[in] lmv pointer to master LMV EA that will be set to the orphan
1261 * \retval positive number for repaired cases
1262 * \retval negative error number on failure
1264 static int lfsck_namespace_create_orphan_dir(const struct lu_env *env,
1265 struct lfsck_component *com,
1266 struct dt_object *orphan,
1267 struct lmv_mds_md_v1 *lmv)
1269 struct lfsck_thread_info *info = lfsck_env_info(env);
1270 struct lu_attr *la = &info->lti_la;
1271 struct dt_allocation_hint *hint = &info->lti_hint;
1272 struct dt_object_format *dof = &info->lti_dof;
1273 struct lu_name *cname = &info->lti_name2;
1274 struct dt_insert_rec *rec = &info->lti_dt_rec;
1275 struct lmv_mds_md_v1 *lmv2 = &info->lti_lmv2;
1276 const struct lu_fid *cfid = lfsck_dto2fid(orphan);
1278 struct lfsck_instance *lfsck = com->lc_lfsck;
1279 struct lfsck_namespace *ns = com->lc_file_ram;
1280 struct dt_device *dev = lfsck_obj2dev(orphan);
1281 struct dt_object *parent = NULL;
1282 struct thandle *th = NULL;
1283 struct lfsck_lock_handle *llh = &info->lti_llh;
1284 struct linkea_data ldata = { NULL };
1285 struct lu_buf linkea_buf;
1286 struct lu_buf lmv_buf;
1294 LASSERT(!dt_object_exists(orphan));
1296 cname->ln_name = NULL;
1297 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1300 if (dt_object_remote(orphan)) {
1301 LASSERT(lfsck->li_lpf_root_obj != NULL);
1303 idx = lfsck_find_mdt_idx_by_fid(env, lfsck, cfid);
1305 GOTO(log, rc = idx);
1307 snprintf(name, 8, "MDT%04x", idx);
1308 rc = dt_lookup(env, lfsck->li_lpf_root_obj,
1309 (struct dt_rec *)&tfid,
1310 (const struct dt_key *)name);
1312 GOTO(log, rc = (rc == -ENOENT ? -ENXIO : rc));
1314 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
1316 GOTO(log, rc = PTR_ERR(parent));
1318 if (unlikely(!dt_try_as_dir(env, parent)))
1319 GOTO(log, rc = -ENOTDIR);
1321 if (unlikely(lfsck->li_lpf_obj == NULL))
1322 GOTO(log, rc = -ENXIO);
1324 parent = lfsck->li_lpf_obj;
1327 dev = lfsck_find_dev_by_fid(env, lfsck, cfid);
1329 GOTO(log, rc = PTR_ERR(dev));
1335 namelen = snprintf(name, 31, DFID"-P-%d",
1337 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
1338 (const struct dt_key *)name);
1339 if (rc != 0 && rc != -ENOENT)
1343 rc = lfsck_lock(env, lfsck, parent, name, llh,
1344 MDS_INODELOCK_UPDATE, LCK_PW);
1348 /* Re-check whether the name conflict with othrs after taken
1350 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
1351 (const struct dt_key *)name);
1352 if (unlikely(rc == 0)) {
1360 cname->ln_name = name;
1361 cname->ln_namelen = namelen;
1363 memset(la, 0, sizeof(*la));
1364 la->la_mode = S_IFDIR | 0700;
1365 la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
1366 LA_ATIME | LA_MTIME | LA_CTIME;
1368 orphan->do_ops->do_ah_init(env, hint, parent, orphan,
1369 la->la_mode & S_IFMT);
1371 memset(dof, 0, sizeof(*dof));
1372 dof->dof_type = dt_mode_to_dft(S_IFDIR);
1374 rc = linkea_data_new(&ldata, &info->lti_linkea_buf2);
1378 rc = linkea_add_buf(&ldata, cname, lfsck_dto2fid(parent));
1382 th = dt_trans_create(env, dev);
1384 GOTO(unlock1, rc = PTR_ERR(th));
1386 /* Sync the remote transaction to guarantee that the subsequent
1387 * lock against the @orphan can find the @orphan in time. */
1388 if (dt_object_remote(orphan))
1391 rc = dt_declare_create(env, orphan, la, hint, dof, th);
1395 if (unlikely(!dt_try_as_dir(env, orphan)))
1396 GOTO(stop, rc = -ENOTDIR);
1398 rec->rec_type = S_IFDIR;
1399 rec->rec_fid = cfid;
1400 rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
1401 (const struct dt_key *)dot, th);
1405 rec->rec_fid = lfsck_dto2fid(parent);
1406 rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
1407 (const struct dt_key *)dotdot, th);
1409 rc = dt_declare_ref_add(env, orphan, th);
1414 rc = dt_declare_ref_add(env, orphan, th);
1419 lmv->lmv_magic = LMV_MAGIC;
1420 lmv->lmv_master_mdt_index = lfsck_dev_idx(lfsck);
1421 lfsck_lmv_header_cpu_to_le(lmv2, lmv);
1422 lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
1423 rc = dt_declare_xattr_set(env, orphan, &lmv_buf,
1424 XATTR_NAME_LMV, 0, th);
1429 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
1430 ldata.ld_leh->leh_len);
1431 rc = dt_declare_xattr_set(env, orphan, &linkea_buf,
1432 XATTR_NAME_LINK, 0, th);
1436 rec->rec_fid = cfid;
1437 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1438 (const struct dt_key *)name, th);
1440 rc = dt_declare_ref_add(env, parent, th);
1445 rc = dt_trans_start_local(env, dev, th);
1449 dt_write_lock(env, orphan, 0);
1450 rc = dt_create(env, orphan, la, hint, dof, th);
1454 rec->rec_fid = cfid;
1455 rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1456 (const struct dt_key *)dot, th, 1);
1460 rec->rec_fid = lfsck_dto2fid(parent);
1461 rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1462 (const struct dt_key *)dotdot, th, 1);
1466 rc = dt_ref_add(env, orphan, th);
1471 rc = dt_xattr_set(env, orphan, &lmv_buf, XATTR_NAME_LMV, 0, th);
1476 rc = dt_xattr_set(env, orphan, &linkea_buf,
1477 XATTR_NAME_LINK, 0, th);
1478 dt_write_unlock(env, orphan);
1482 rec->rec_fid = cfid;
1483 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1484 (const struct dt_key *)name, th, 1);
1486 dt_write_lock(env, parent, 0);
1487 rc = dt_ref_add(env, parent, th);
1488 dt_write_unlock(env, parent);
1491 GOTO(stop, rc = (rc == 0 ? 1 : rc));
1494 dt_write_unlock(env, orphan);
1497 rc1 = dt_trans_stop(env, dev, th);
1498 if (rc1 != 0 && rc > 0)
1505 CDEBUG(D_LFSCK, "%s: namespace LFSCK create orphan dir for "
1506 "the object "DFID", name = %s: rc = %d\n",
1507 lfsck_lfsck2name(lfsck), PFID(cfid),
1508 cname->ln_name != NULL ? cname->ln_name : "<NULL>", rc);
1510 if (parent != NULL && !IS_ERR(parent) && parent != lfsck->li_lpf_obj)
1511 lfsck_object_put(env, parent);
1514 ns->ln_flags |= LF_INCONSISTENT;
1520 * Remove the specified entry from the linkEA.
1522 * Locate the linkEA entry with the given @cname and @pfid, then
1523 * remove this entry or the other entries those are repeated with
1526 * \param[in] env pointer to the thread context
1527 * \param[in] com pointer to the lfsck component
1528 * \param[in] obj pointer to the dt_object to be handled
1529 * \param[in,out]ldata pointer to the buffer that holds the linkEA
1530 * \param[in] cname the name for the child in the parent directory
1531 * \param[in] pfid the parent directory's FID for the linkEA
1532 * \param[in] next if true, then remove the first found linkEA
1533 * entry, and move the ldata->ld_lee to next entry
1535 * \retval positive number for repaired cases
1536 * \retval 0 if nothing to be repaired
1537 * \retval negative error number on failure
1539 static int lfsck_namespace_shrink_linkea(const struct lu_env *env,
1540 struct lfsck_component *com,
1541 struct dt_object *obj,
1542 struct linkea_data *ldata,
1543 struct lu_name *cname,
1544 struct lu_fid *pfid,
1547 struct lfsck_instance *lfsck = com->lc_lfsck;
1548 struct dt_device *dev = lfsck_obj2dev(obj);
1549 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
1550 struct thandle *th = NULL;
1551 struct lustre_handle lh = { 0 };
1552 struct linkea_data ldata_new = { NULL };
1553 struct lu_buf linkea_buf;
1557 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
1558 MDS_INODELOCK_UPDATE | MDS_INODELOCK_XATTR,
1564 linkea_del_buf(ldata, cname);
1566 lfsck_namespace_filter_linkea_entry(ldata, cname, pfid,
1568 lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
1569 ldata->ld_leh->leh_len);
1572 th = dt_trans_create(env, dev);
1574 GOTO(unlock1, rc = PTR_ERR(th));
1576 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
1577 XATTR_NAME_LINK, 0, th);
1581 rc = dt_trans_start_local(env, dev, th);
1585 dt_write_lock(env, obj, 0);
1586 if (unlikely(lfsck_is_dead_obj(obj)))
1587 GOTO(unlock2, rc = -ENOENT);
1589 rc = lfsck_links_read2(env, obj, &ldata_new);
1593 /* The specified linkEA entry has been removed by race. */
1594 rc = linkea_links_find(&ldata_new, cname, pfid);
1596 GOTO(unlock2, rc = 0);
1598 if (bk->lb_param & LPF_DRYRUN)
1599 GOTO(unlock2, rc = 1);
1602 linkea_del_buf(&ldata_new, cname);
1604 lfsck_namespace_filter_linkea_entry(&ldata_new, cname, pfid,
1607 if (linkea_buf.lb_len < ldata_new.ld_leh->leh_len) {
1608 dt_write_unlock(env, obj);
1609 dt_trans_stop(env, dev, th);
1610 lfsck_buf_init(&linkea_buf, ldata_new.ld_buf->lb_buf,
1611 ldata_new.ld_leh->leh_len);
1615 lfsck_buf_init(&linkea_buf, ldata_new.ld_buf->lb_buf,
1616 ldata_new.ld_leh->leh_len);
1617 rc = dt_xattr_set(env, obj, &linkea_buf, XATTR_NAME_LINK, 0, th);
1619 GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
1622 dt_write_unlock(env, obj);
1625 dt_trans_stop(env, dev, th);
1628 lfsck_ibits_unlock(&lh, LCK_EX);
1631 CDEBUG(D_LFSCK, "%s: namespace LFSCK remove %s linkEA entry "
1632 "for the object: "DFID", parent "DFID", name %.*s\n",
1633 lfsck_lfsck2name(lfsck), next ? "invalid" : "redundant",
1634 PFID(lfsck_dto2fid(obj)), PFID(pfid), cname->ln_namelen,
1638 struct lfsck_namespace *ns = com->lc_file_ram;
1640 ns->ln_flags |= LF_INCONSISTENT;
1647 * Conditionally remove the specified entry from the linkEA.
1649 * Take the parent lock firstly, then check whether the specified
1650 * name entry exists or not: if yes, do nothing; otherwise, call
1651 * lfsck_namespace_shrink_linkea() to remove the linkea entry.
1653 * \param[in] env pointer to the thread context
1654 * \param[in] com pointer to the lfsck component
1655 * \param[in] parent pointer to the parent directory
1656 * \param[in] child pointer to the child object that holds the linkEA
1657 * \param[in,out]ldata pointer to the buffer that holds the linkEA
1658 * \param[in] cname the name for the child in the parent directory
1659 * \param[in] pfid the parent directory's FID for the linkEA
1661 * \retval positive number for repaired cases
1662 * \retval 0 if nothing to be repaired
1663 * \retval negative error number on failure
1665 static int lfsck_namespace_shrink_linkea_cond(const struct lu_env *env,
1666 struct lfsck_component *com,
1667 struct dt_object *parent,
1668 struct dt_object *child,
1669 struct linkea_data *ldata,
1670 struct lu_name *cname,
1671 struct lu_fid *pfid)
1673 struct lfsck_thread_info *info = lfsck_env_info(env);
1674 struct lu_fid *cfid = &info->lti_fid3;
1675 struct lfsck_lock_handle *llh = &info->lti_llh;
1679 rc = lfsck_lock(env, com->lc_lfsck, parent, cname->ln_name, llh,
1680 MDS_INODELOCK_UPDATE, LCK_PR);
1684 dt_read_lock(env, parent, 0);
1685 if (unlikely(lfsck_is_dead_obj(parent))) {
1686 dt_read_unlock(env, parent);
1688 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
1694 rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
1695 (const struct dt_key *)cname->ln_name);
1696 dt_read_unlock(env, parent);
1698 /* It is safe to release the ldlm lock, because when the logic come
1699 * here, we have got all the needed information above whether the
1700 * linkEA entry is valid or not. It is not important that others
1701 * may add new linkEA entry after the ldlm lock released. If other
1702 * has removed the specified linkEA entry by race, then it is OK,
1703 * because the subsequent lfsck_namespace_shrink_linkea() can handle
1706 if (rc == -ENOENT) {
1707 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
1716 /* The LFSCK just found some internal status of cross-MDTs
1717 * create operation. That is normal. */
1718 if (lu_fid_eq(cfid, lfsck_dto2fid(child))) {
1719 linkea_next_entry(ldata);
1724 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata, cname,
1731 * Conditionally replace name entry in the parent.
1733 * As required, the LFSCK may re-create the lost MDT-object for dangling
1734 * name entry, but such repairing may be wrong because of bad FID in the
1735 * name entry. As the LFSCK processing, the real MDT-object may be found,
1736 * then the LFSCK should check whether the former re-created MDT-object
1737 * has been modified or not, if not, then destroy it and update the name
1738 * entry in the parent to reference the real MDT-object.
1740 * \param[in] env pointer to the thread context
1741 * \param[in] com pointer to the lfsck component
1742 * \param[in] parent pointer to the parent directory
1743 * \param[in] child pointer to the MDT-object that may be the real
1744 * MDT-object corresponding to the name entry in parent
1745 * \param[in] cfid the current FID in the name entry
1746 * \param[in] cname contains the name of the child in the parent directory
1748 * \retval positive number for repaired cases
1749 * \retval 0 if nothing to be repaired
1750 * \retval negative error number on failure
1752 static int lfsck_namespace_replace_cond(const struct lu_env *env,
1753 struct lfsck_component *com,
1754 struct dt_object *parent,
1755 struct dt_object *child,
1756 const struct lu_fid *cfid,
1757 const struct lu_name *cname)
1759 struct lfsck_thread_info *info = lfsck_env_info(env);
1760 struct lu_attr *la = &info->lti_la;
1761 struct dt_insert_rec *rec = &info->lti_dt_rec;
1763 struct lfsck_instance *lfsck = com->lc_lfsck;
1764 /* The child and its name may be on different MDTs. */
1765 struct dt_device *dev = lfsck->li_next;
1766 const char *name = cname->ln_name;
1767 const struct lu_fid *pfid = lfsck_dto2fid(parent);
1768 struct dt_object *cobj = NULL;
1769 struct lfsck_lock_handle *pllh = &info->lti_llh;
1770 struct lustre_handle clh = { 0 };
1771 struct linkea_data ldata = { NULL };
1772 struct thandle *th = NULL;
1777 /* @parent/@child may be based on lfsck->li_bottom,
1778 * but here we need the object based on the lfsck->li_next. */
1780 parent = lfsck_object_locate(dev, parent);
1782 GOTO(log, rc = PTR_ERR(parent));
1784 if (unlikely(!dt_try_as_dir(env, parent)))
1785 GOTO(log, rc = -ENOTDIR);
1787 rc = lfsck_lock(env, lfsck, parent, name, pllh,
1788 MDS_INODELOCK_UPDATE, LCK_PW);
1792 if (!fid_is_sane(cfid)) {
1797 cobj = lfsck_object_find_by_dev(env, dev, cfid);
1800 if (rc == -ENOENT) {
1808 if (!dt_object_exists(cobj)) {
1813 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
1814 (const struct dt_key *)name);
1815 if (rc == -ENOENT) {
1823 /* Someone changed the name entry, cannot replace it. */
1824 if (!lu_fid_eq(cfid, &tfid))
1827 /* lock the object to be destroyed. */
1828 rc = lfsck_ibits_lock(env, lfsck, cobj, &clh,
1829 MDS_INODELOCK_UPDATE |
1830 MDS_INODELOCK_UPDATE | MDS_INODELOCK_XATTR,
1835 if (unlikely(lfsck_is_dead_obj(cobj))) {
1840 rc = dt_attr_get(env, cobj, la);
1844 /* The object has been modified by other(s), or it is not created by
1845 * LFSCK, the two cases are indistinguishable. So cannot replace it. */
1846 if (la->la_ctime != 0)
1849 if (S_ISREG(la->la_mode)) {
1850 rc = dt_xattr_get(env, cobj, &LU_BUF_NULL, XATTR_NAME_LOV);
1851 /* If someone has created related OST-object(s),
1853 if ((rc > 0) || (rc < 0 && rc != -ENODATA))
1854 GOTO(log, rc = (rc > 0 ? 0 : rc));
1858 dt_read_lock(env, child, 0);
1859 rc = lfsck_links_read2(env, child, &ldata);
1860 dt_read_unlock(env, child);
1862 /* Someone changed the child, no need to replace. */
1869 rc = linkea_links_find(&ldata, cname, pfid);
1870 /* Someone moved the child, no need to replace. */
1874 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1877 th = dt_trans_create(env, dev);
1879 GOTO(log, rc = PTR_ERR(th));
1882 rc = dt_declare_destroy(env, cobj, th);
1887 rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
1891 rec->rec_type = S_IFDIR;
1892 rec->rec_fid = lfsck_dto2fid(child);
1893 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1894 (const struct dt_key *)name, th);
1898 rc = dt_trans_start_local(env, dev, th);
1903 rc = dt_destroy(env, cobj, th);
1908 /* The old name entry maybe not exist. */
1909 rc = dt_delete(env, parent, (const struct dt_key *)name, th);
1910 if (rc != 0 && rc != -ENOENT)
1913 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1914 (const struct dt_key *)name, th, 1);
1916 GOTO(stop, rc = (rc == 0 ? 1 : rc));
1919 dt_trans_stop(env, dev, th);
1922 lfsck_ibits_unlock(&clh, LCK_EX);
1925 if (cobj != NULL && !IS_ERR(cobj))
1926 lfsck_object_put(env, cobj);
1928 CDEBUG(D_LFSCK, "%s: namespace LFSCK conditionally destroy the "
1929 "object "DFID" because of conflict with the object "DFID
1930 " under the parent "DFID" with name %s: rc = %d\n",
1931 lfsck_lfsck2name(lfsck), PFID(cfid),
1932 PFID(lfsck_dto2fid(child)), PFID(pfid), name, rc);
1938 * Overwrite the linkEA for the object with the given ldata.
1940 * The caller should take the ldlm lock before the calling.
1942 * \param[in] env pointer to the thread context
1943 * \param[in] com pointer to the lfsck component
1944 * \param[in] obj pointer to the dt_object to be handled
1945 * \param[in] ldata pointer to the new linkEA data
1947 * \retval positive number for repaired cases
1948 * \retval 0 if nothing to be repaired
1949 * \retval negative error number on failure
1951 int lfsck_namespace_rebuild_linkea(const struct lu_env *env,
1952 struct lfsck_component *com,
1953 struct dt_object *obj,
1954 struct linkea_data *ldata)
1956 struct lfsck_instance *lfsck = com->lc_lfsck;
1957 struct dt_device *dev = lfsck_obj2dev(obj);
1958 struct thandle *th = NULL;
1959 struct lu_buf linkea_buf;
1963 th = dt_trans_create(env, dev);
1965 GOTO(log, rc = PTR_ERR(th));
1967 lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
1968 ldata->ld_leh->leh_len);
1969 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
1970 XATTR_NAME_LINK, 0, th);
1974 rc = dt_trans_start_local(env, dev, th);
1978 dt_write_lock(env, obj, 0);
1979 if (unlikely(lfsck_is_dead_obj(obj)))
1980 GOTO(unlock, rc = 0);
1982 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1983 GOTO(unlock, rc = 1);
1985 rc = dt_xattr_set(env, obj, &linkea_buf,
1986 XATTR_NAME_LINK, 0, th);
1988 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
1991 dt_write_unlock(env, obj);
1994 dt_trans_stop(env, dev, th);
1997 CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild linkEA for the "
1998 "object "DFID": rc = %d\n",
1999 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
2002 struct lfsck_namespace *ns = com->lc_file_ram;
2004 ns->ln_flags |= LF_INCONSISTENT;
2011 * Repair invalid name entry.
2013 * If the name entry contains invalid information, such as bad file type
2014 * or (and) corrupted object FID, then either remove the name entry or
2015 * udpate the name entry with the given (right) information.
2017 * \param[in] env pointer to the thread context
2018 * \param[in] com pointer to the lfsck component
2019 * \param[in] parent pointer to the parent directory
2020 * \param[in] child pointer to the object referenced by the name entry
2021 * \param[in] name the old name of the child under the parent directory
2022 * \param[in] name2 the new name of the child under the parent directory
2023 * \param[in] type the type claimed by the name entry
2024 * \param[in] update update the name entry if true; otherwise, remove it
2025 * \param[in] dec decrease the parent nlink count if true
2027 * \retval positive number for repaired successfully
2028 * \retval 0 if nothing to be repaired
2029 * \retval negative error number on failure
2031 int lfsck_namespace_repair_dirent(const struct lu_env *env,
2032 struct lfsck_component *com,
2033 struct dt_object *parent,
2034 struct dt_object *child,
2035 const char *name, const char *name2,
2036 __u16 type, bool update, bool dec)
2038 struct lfsck_thread_info *info = lfsck_env_info(env);
2039 struct dt_insert_rec *rec = &info->lti_dt_rec;
2040 const struct lu_fid *pfid = lfsck_dto2fid(parent);
2041 const struct lu_fid *cfid = lfsck_dto2fid(child);
2043 struct lfsck_instance *lfsck = com->lc_lfsck;
2044 struct dt_device *dev = lfsck->li_next;
2045 struct thandle *th = NULL;
2046 struct lfsck_lock_handle *llh = &info->lti_llh;
2047 struct lustre_handle lh = { 0 };
2051 parent = lfsck_object_locate(dev, parent);
2053 GOTO(log, rc = PTR_ERR(parent));
2055 if (unlikely(!dt_try_as_dir(env, parent)))
2056 GOTO(log, rc = -ENOTDIR);
2058 if (!update || strcmp(name, name2) == 0)
2059 rc = lfsck_lock(env, lfsck, parent, name, llh,
2060 MDS_INODELOCK_UPDATE, LCK_PW);
2062 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
2063 MDS_INODELOCK_UPDATE, LCK_PW);
2067 th = dt_trans_create(env, dev);
2069 GOTO(unlock1, rc = PTR_ERR(th));
2071 rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
2076 rec->rec_type = lfsck_object_type(child) & S_IFMT;
2077 rec->rec_fid = cfid;
2078 rc = dt_declare_insert(env, parent,
2079 (const struct dt_rec *)rec,
2080 (const struct dt_key *)name2, th);
2086 rc = dt_declare_ref_del(env, parent, th);
2091 rc = dt_trans_start_local(env, dev, th);
2096 dt_write_lock(env, parent, 0);
2097 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
2098 (const struct dt_key *)name);
2099 /* Someone has removed the bad name entry by race. */
2101 GOTO(unlock2, rc = 0);
2106 /* Someone has removed the bad name entry and reused it for other
2107 * object by race. */
2108 if (!lu_fid_eq(&tfid, cfid))
2109 GOTO(unlock2, rc = 0);
2111 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2112 GOTO(unlock2, rc = 1);
2114 rc = dt_delete(env, parent, (const struct dt_key *)name, th);
2119 rc = dt_insert(env, parent,
2120 (const struct dt_rec *)rec,
2121 (const struct dt_key *)name2, th, 1);
2127 rc = dt_ref_del(env, parent, th);
2132 GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
2135 dt_write_unlock(env, parent);
2138 dt_trans_stop(env, dev, th);
2140 /* We are not sure whether the child will become orphan or not.
2141 * Record it in the LFSCK trace file for further checking in
2142 * the second-stage scanning. */
2143 if (!update && !dec && rc == 0)
2144 lfsck_namespace_trace_update(env, com, cfid,
2145 LNTF_CHECK_LINKEA, true);
2148 /* It is harmless even if unlock the unused lock_handle */
2149 lfsck_ibits_unlock(&lh, LCK_PW);
2153 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found bad name "
2154 "entry for: parent "DFID", child "DFID", name %s, type "
2155 "in name entry %o, type claimed by child %o. repair it "
2156 "by %s with new name2 %s: rc = %d\n",
2157 lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
2158 name, type, update ? lfsck_object_type(child) : 0,
2159 update ? "updating" : "removing", name2, rc);
2162 struct lfsck_namespace *ns = com->lc_file_ram;
2164 ns->ln_flags |= LF_INCONSISTENT;
2171 * Update the ".." name entry for the given object.
2173 * The object's ".." is corrupted, this function will update the ".." name
2174 * entry with the given pfid, and the linkEA with the given ldata.
2176 * The caller should take the ldlm lock before the calling.
2178 * \param[in] env pointer to the thread context
2179 * \param[in] com pointer to the lfsck component
2180 * \param[in] obj pointer to the dt_object to be handled
2181 * \param[in] pfid the new fid for the object's ".." name entry
2182 * \param[in] cname the name for the @obj in the parent directory
2184 * \retval positive number for repaired cases
2185 * \retval 0 if nothing to be repaired
2186 * \retval negative error number on failure
2188 static int lfsck_namespace_repair_unmatched_pairs(const struct lu_env *env,
2189 struct lfsck_component *com,
2190 struct dt_object *obj,
2191 const struct lu_fid *pfid,
2192 struct lu_name *cname)
2194 struct lfsck_thread_info *info = lfsck_env_info(env);
2195 struct dt_insert_rec *rec = &info->lti_dt_rec;
2196 struct lfsck_instance *lfsck = com->lc_lfsck;
2197 struct dt_device *dev = lfsck_obj2dev(obj);
2198 struct thandle *th = NULL;
2199 struct linkea_data ldata = { NULL };
2200 struct lu_buf linkea_buf;
2204 LASSERT(!dt_object_remote(obj));
2205 LASSERT(S_ISDIR(lfsck_object_type(obj)));
2207 rc = linkea_data_new(&ldata, &info->lti_big_buf);
2211 rc = linkea_add_buf(&ldata, cname, pfid);
2215 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
2216 ldata.ld_leh->leh_len);
2218 th = dt_trans_create(env, dev);
2220 GOTO(log, rc = PTR_ERR(th));
2222 rc = dt_declare_delete(env, obj, (const struct dt_key *)dotdot, th);
2226 rec->rec_type = S_IFDIR;
2227 rec->rec_fid = pfid;
2228 rc = dt_declare_insert(env, obj, (const struct dt_rec *)rec,
2229 (const struct dt_key *)dotdot, th);
2233 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
2234 XATTR_NAME_LINK, 0, th);
2238 rc = dt_trans_start_local(env, dev, th);
2242 dt_write_lock(env, obj, 0);
2243 if (unlikely(lfsck_is_dead_obj(obj)))
2244 GOTO(unlock, rc = 0);
2246 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2247 GOTO(unlock, rc = 1);
2249 /* The old ".." name entry maybe not exist. */
2250 dt_delete(env, obj, (const struct dt_key *)dotdot, th);
2252 rc = dt_insert(env, obj, (const struct dt_rec *)rec,
2253 (const struct dt_key *)dotdot, th, 1);
2257 rc = dt_xattr_set(env, obj, &linkea_buf,
2258 XATTR_NAME_LINK, 0, th);
2260 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2263 dt_write_unlock(env, obj);
2266 dt_trans_stop(env, dev, th);
2269 CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild dotdot name entry for "
2270 "the object "DFID", new parent "DFID": rc = %d\n",
2271 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)),
2275 struct lfsck_namespace *ns = com->lc_file_ram;
2277 ns->ln_flags |= LF_INCONSISTENT;
2284 * Handle orphan @obj during Double Scan Directory.
2286 * Remove the @obj's current (invalid) linkEA entries, and insert
2287 * it in the directory .lustre/lost+found/MDTxxxx/ with the name:
2288 * ${FID}-${PFID}-D-${conflict_version}
2290 * The caller should take the ldlm lock before the calling.
2292 * \param[in] env pointer to the thread context
2293 * \param[in] com pointer to the lfsck component
2294 * \param[in] obj pointer to the orphan object to be handled
2295 * \param[in] pfid the new fid for the object's ".." name entry
2296 * \param[in,out] lh ldlm lock handler for the given @obj
2297 * \param[out] type to tell the caller what the inconsistency is
2299 * \retval positive number for repaired cases
2300 * \retval 0 if nothing to be repaired
2301 * \retval negative error number on failure
2304 lfsck_namespace_dsd_orphan(const struct lu_env *env,
2305 struct lfsck_component *com,
2306 struct dt_object *obj,
2307 const struct lu_fid *pfid,
2308 struct lustre_handle *lh,
2309 enum lfsck_namespace_inconsistency_type *type)
2311 struct lfsck_thread_info *info = lfsck_env_info(env);
2312 struct lfsck_namespace *ns = com->lc_file_ram;
2316 /* Remove the unrecognized linkEA. */
2317 rc = lfsck_namespace_links_remove(env, com, obj);
2318 lfsck_ibits_unlock(lh, LCK_EX);
2319 if (rc < 0 && rc != -ENODATA)
2322 *type = LNIT_MUL_REF;
2324 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
2325 * ever tried to verify some remote MDT-object that resides on this
2326 * MDT, but this MDT failed to respond such request. So means there
2327 * may be some remote name entry on other MDT that references this
2328 * object with another name, so we cannot know whether this linkEA
2329 * is valid or not. So keep it there and maybe resolved when next
2331 if (ns->ln_flags & LF_INCOMPLETE)
2334 /* The unique linkEA is invalid, even if the ".." name entry may be
2335 * valid, we still cannot know via which name entry this directory
2336 * will be referenced. Then handle it as pure orphan. */
2337 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2338 "-"DFID, PFID(pfid));
2339 rc = lfsck_namespace_insert_orphan(env, com, obj,
2340 info->lti_tmpbuf, "D", NULL);
2346 * Double Scan Directory object for single linkEA entry case.
2348 * The given @child has unique linkEA entry. If the linkEA entry is valid,
2349 * then check whether the name is in the namespace or not, if not, add the
2350 * missing name entry back to namespace. If the linkEA entry is invalid,
2351 * then remove it and insert the @child in the .lustre/lost+found/MDTxxxx/
2354 * \param[in] env pointer to the thread context
2355 * \param[in] com pointer to the lfsck component
2356 * \param[in] child pointer to the directory to be double scanned
2357 * \param[in] pfid the FID corresponding to the ".." entry
2358 * \param[in] ldata pointer to the linkEA data for the given @child
2359 * \param[in,out] lh ldlm lock handler for the given @child
2360 * \param[out] type to tell the caller what the inconsistency is
2361 * \param[in] retry if found inconsistency, but the caller does not hold
2362 * ldlm lock on the @child, then set @retry as true
2364 * \retval positive number for repaired cases
2365 * \retval 0 if nothing to be repaired
2366 * \retval negative error number on failure
2369 lfsck_namespace_dsd_single(const struct lu_env *env,
2370 struct lfsck_component *com,
2371 struct dt_object *child,
2372 const struct lu_fid *pfid,
2373 struct linkea_data *ldata,
2374 struct lustre_handle *lh,
2375 enum lfsck_namespace_inconsistency_type *type,
2378 struct lfsck_thread_info *info = lfsck_env_info(env);
2379 struct lu_name *cname = &info->lti_name;
2380 const struct lu_fid *cfid = lfsck_dto2fid(child);
2382 struct lfsck_namespace *ns = com->lc_file_ram;
2383 struct lfsck_instance *lfsck = com->lc_lfsck;
2384 struct dt_object *parent = NULL;
2385 struct lmv_mds_md_v1 *lmv;
2389 lfsck_namespace_unpack_linkea_entry(ldata, cname, &tfid, info->lti_key);
2390 /* The unique linkEA entry with bad parent will be handled as orphan. */
2391 if (!fid_is_sane(&tfid)) {
2392 if (!lustre_handle_is_used(lh) && retry != NULL)
2395 rc = lfsck_namespace_dsd_orphan(env, com, child,
2401 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
2403 GOTO(out, rc = PTR_ERR(parent));
2405 /* We trust the unique linkEA entry in spite of whether it matches the
2406 * ".." name entry or not. Because even if the linkEA entry is wrong
2407 * and the ".." name entry is right, we still cannot know via which
2408 * name entry the child will be referenced, since all known entries
2409 * have been verified during the first-stage scanning. */
2410 if (!dt_object_exists(parent)) {
2411 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2412 * has ever tried to verify some remote MDT-object that resides
2413 * on this MDT, but this MDT failed to respond such request. So
2414 * means there may be some remote name entry on other MDT that
2415 * references this object with another name, so we cannot know
2416 * whether this linkEA is valid or not. So keep it there and
2417 * maybe resolved when next LFSCK run. */
2418 if (ns->ln_flags & LF_INCOMPLETE)
2421 if (!lustre_handle_is_used(lh) && retry != NULL) {
2427 lfsck_ibits_unlock(lh, LCK_EX);
2430 lmv = &info->lti_lmv;
2431 rc = lfsck_read_stripe_lmv(env, child, lmv);
2432 if (rc != 0 && rc != -ENODATA)
2435 if (rc == -ENODATA || lmv->lmv_magic != LMV_MAGIC_STRIPE) {
2437 } else if (lfsck_shard_name_to_index(env,
2438 cname->ln_name, cname->ln_namelen,
2439 S_IFDIR, cfid) < 0) {
2440 /* It is an invalid name entry, we
2441 * cannot trust the parent also. */
2442 rc = lfsck_namespace_shrink_linkea(env, com, child,
2443 ldata, cname, &tfid, true);
2447 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2448 "-"DFID, PFID(pfid));
2449 rc = lfsck_namespace_insert_orphan(env, com, child,
2450 info->lti_tmpbuf, "S", NULL);
2455 /* Create the lost parent as an orphan. */
2456 rc = lfsck_namespace_create_orphan_dir(env, com, parent, lmv);
2458 /* Add the missing name entry to the parent. */
2459 rc = lfsck_namespace_insert_normal(env, com, parent,
2460 child, cname->ln_name);
2461 if (unlikely(rc == -EEXIST)) {
2462 /* Unfortunately, someone reused the name
2463 * under the parent by race. So we have
2464 * to remove the linkEA entry from
2465 * current child object. It means that the
2466 * LFSCK cannot recover the system
2467 * totally back to its original status,
2468 * but it is necessary to make the
2469 * current system to be consistent. */
2470 rc = lfsck_namespace_shrink_linkea(env,
2472 cname, &tfid, true);
2474 snprintf(info->lti_tmpbuf,
2475 sizeof(info->lti_tmpbuf),
2476 "-"DFID, PFID(pfid));
2477 rc = lfsck_namespace_insert_orphan(env,
2478 com, child, info->lti_tmpbuf,
2487 /* The unique linkEA entry with bad parent will be handled as orphan. */
2488 if (unlikely(!dt_try_as_dir(env, parent))) {
2489 if (!lustre_handle_is_used(lh) && retry != NULL)
2492 rc = lfsck_namespace_dsd_orphan(env, com, child,
2498 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
2499 (const struct dt_key *)cname->ln_name);
2500 if (rc == -ENOENT) {
2501 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2502 * has ever tried to verify some remote MDT-object that resides
2503 * on this MDT, but this MDT failed to respond such request. So
2504 * means there may be some remote name entry on other MDT that
2505 * references this object with another name, so we cannot know
2506 * whether this linkEA is valid or not. So keep it there and
2507 * maybe resolved when next LFSCK run. */
2508 if (ns->ln_flags & LF_INCOMPLETE)
2511 if (!lustre_handle_is_used(lh) && retry != NULL) {
2517 lfsck_ibits_unlock(lh, LCK_EX);
2518 rc = lfsck_namespace_check_name(env, parent, child, cname);
2525 /* It is an invalid name entry, drop it. */
2526 if (unlikely(rc > 0)) {
2527 rc = lfsck_namespace_shrink_linkea(env, com, child,
2528 ldata, cname, &tfid, true);
2530 snprintf(info->lti_tmpbuf,
2531 sizeof(info->lti_tmpbuf),
2532 "-"DFID, PFID(pfid));
2533 rc = lfsck_namespace_insert_orphan(env, com,
2534 child, info->lti_tmpbuf, "D", NULL);
2540 /* Add the missing name entry back to the namespace. */
2541 rc = lfsck_namespace_insert_normal(env, com, parent, child,
2543 if (unlikely(rc == -ESTALE))
2544 /* It may happen when the remote object has been
2545 * removed, but the local MDT is not aware of that. */
2548 if (unlikely(rc == -EEXIST)) {
2549 /* Unfortunately, someone reused the name under the
2550 * parent by race. So we have to remove the linkEA
2551 * entry from current child object. It means that the
2552 * LFSCK cannot recover the system totally back to
2553 * its original status, but it is necessary to make
2554 * the current system to be consistent.
2556 * It also may be because of the LFSCK found some
2557 * internal status of create operation. Under such
2558 * case, nothing to be done. */
2559 rc = lfsck_namespace_shrink_linkea_cond(env, com,
2560 parent, child, ldata, cname, &tfid);
2562 snprintf(info->lti_tmpbuf,
2563 sizeof(info->lti_tmpbuf),
2564 "-"DFID, PFID(pfid));
2565 rc = lfsck_namespace_insert_orphan(env, com,
2566 child, info->lti_tmpbuf, "D", NULL);
2576 if (!lu_fid_eq(&tfid, cfid)) {
2577 if (!lustre_handle_is_used(lh) && retry != NULL) {
2583 lfsck_ibits_unlock(lh, LCK_EX);
2584 /* The name entry references another MDT-object that
2585 * may be created by the LFSCK for repairing dangling
2586 * name entry. Try to replace it. */
2587 rc = lfsck_namespace_replace_cond(env, com, parent, child,
2590 rc = lfsck_namespace_dsd_orphan(env, com, child,
2596 if (fid_is_zero(pfid))
2599 /* The ".." name entry is wrong, update it. */
2600 if (!lu_fid_eq(pfid, lfsck_dto2fid(parent))) {
2601 if (!lustre_handle_is_used(lh) && retry != NULL) {
2607 *type = LNIT_UNMATCHED_PAIRS;
2608 rc = lfsck_namespace_repair_unmatched_pairs(env, com, child,
2609 lfsck_dto2fid(parent), cname);
2615 if (parent != NULL && !IS_ERR(parent))
2616 lfsck_object_put(env, parent);
2622 * Double Scan Directory object for multiple linkEA entries case.
2624 * The given @child has multiple linkEA entries. There is at most one linkEA
2625 * entry will be valid, all the others will be removed. Firstly, the function
2626 * will try to find out the linkEA entry for which the name entry exists under
2627 * the given parent (@pfid). If there is no linkEA entry that matches the given
2628 * ".." name entry, then tries to find out the first linkEA entry that both the
2629 * parent and the name entry exist to rebuild a new ".." name entry.
2631 * \param[in] env pointer to the thread context
2632 * \param[in] com pointer to the lfsck component
2633 * \param[in] child pointer to the directory to be double scanned
2634 * \param[in] pfid the FID corresponding to the ".." entry
2635 * \param[in] ldata pointer to the linkEA data for the given @child
2636 * \param[in,out] lh ldlm lock handler for the given @child
2637 * \param[out] type to tell the caller what the inconsistency is
2638 * \param[in] lpf true if the ".." entry is under lost+found/MDTxxxx/
2640 * \retval positive number for repaired cases
2641 * \retval 0 if nothing to be repaired
2642 * \retval negative error number on failure
2645 lfsck_namespace_dsd_multiple(const struct lu_env *env,
2646 struct lfsck_component *com,
2647 struct dt_object *child,
2648 const struct lu_fid *pfid,
2649 struct linkea_data *ldata,
2650 struct lustre_handle *lh,
2651 enum lfsck_namespace_inconsistency_type *type,
2654 struct lfsck_thread_info *info = lfsck_env_info(env);
2655 struct lu_name *cname = &info->lti_name;
2656 const struct lu_fid *cfid = lfsck_dto2fid(child);
2657 struct lu_fid *pfid2 = &info->lti_fid3;
2659 struct lfsck_namespace *ns = com->lc_file_ram;
2660 struct lfsck_instance *lfsck = com->lc_lfsck;
2661 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
2662 struct dt_object *parent = NULL;
2663 struct linkea_data ldata_new = { NULL };
2664 int dirent_count = 0;
2665 int linkea_count = 0;
2671 while (ldata->ld_lee != NULL) {
2672 lfsck_namespace_unpack_linkea_entry(ldata, cname, &tfid,
2674 /* Drop repeated linkEA entries. */
2675 lfsck_namespace_filter_linkea_entry(ldata, cname, &tfid, true);
2676 /* Drop invalid linkEA entry. */
2677 if (!fid_is_sane(&tfid)) {
2678 linkea_del_buf(ldata, cname);
2683 /* If current dotdot is the .lustre/lost+found/MDTxxxx/,
2684 * then it is possible that: the directry object has ever
2685 * been lost, but its name entry was there. In the former
2686 * LFSCK run, during the first-stage scanning, the LFSCK
2687 * found the dangling name entry, but it did not recreate
2688 * the lost object, and when moved to the second-stage
2689 * scanning, some children objects of the lost directory
2690 * object were found, then the LFSCK recreated such lost
2691 * directory object as an orphan.
2693 * When the LFSCK runs again, if the dangling name is still
2694 * there, the LFSCK should move the orphan directory object
2695 * back to the normal namespace. */
2696 if (!lpf && !lu_fid_eq(pfid, &tfid) && once) {
2697 linkea_next_entry(ldata);
2701 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
2703 RETURN(PTR_ERR(parent));
2705 if (!dt_object_exists(parent)) {
2706 lfsck_object_put(env, parent);
2707 if (ldata->ld_leh->leh_reccount > 1) {
2708 /* If it is NOT the last linkEA entry, then
2709 * there is still other chance to make the
2710 * child to be visible via other parent, then
2711 * remove this linkEA entry. */
2712 linkea_del_buf(ldata, cname);
2720 /* The linkEA entry with bad parent will be removed. */
2721 if (unlikely(!dt_try_as_dir(env, parent))) {
2722 lfsck_object_put(env, parent);
2723 linkea_del_buf(ldata, cname);
2728 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
2729 (const struct dt_key *)cname->ln_name);
2730 *pfid2 = *lfsck_dto2fid(parent);
2731 if (rc == -ENOENT) {
2732 lfsck_object_put(env, parent);
2733 linkea_next_entry(ldata);
2738 lfsck_object_put(env, parent);
2743 if (lu_fid_eq(&tfid, cfid)) {
2744 lfsck_object_put(env, parent);
2745 if (!lu_fid_eq(pfid, pfid2)) {
2746 *type = LNIT_UNMATCHED_PAIRS;
2747 rc = lfsck_namespace_repair_unmatched_pairs(env,
2748 com, child, pfid2, cname);
2754 /* It is the most common case that we find the
2755 * name entry corresponding to the linkEA entry
2756 * that matches the ".." name entry. */
2757 rc = linkea_data_new(&ldata_new, &info->lti_big_buf);
2761 rc = linkea_add_buf(&ldata_new, cname, pfid2);
2765 rc = lfsck_namespace_rebuild_linkea(env, com, child,
2770 linkea_del_buf(ldata, cname);
2772 linkea_first_entry(ldata);
2773 /* There may be some invalid dangling name entries under
2774 * other parent directories, remove all of them. */
2775 while (ldata->ld_lee != NULL) {
2776 lfsck_namespace_unpack_linkea_entry(ldata,
2777 cname, &tfid, info->lti_key);
2778 if (!fid_is_sane(&tfid))
2781 parent = lfsck_object_find_bottom(env, lfsck,
2783 if (IS_ERR(parent)) {
2784 rc = PTR_ERR(parent);
2785 if (rc != -ENOENT &&
2786 bk->lb_param & LPF_FAILOUT)
2792 if (!dt_object_exists(parent)) {
2793 lfsck_object_put(env, parent);
2797 rc = lfsck_namespace_repair_dirent(env, com,
2798 parent, child, cname->ln_name,
2799 cname->ln_name, S_IFDIR, false, true);
2800 lfsck_object_put(env, parent);
2802 if (bk->lb_param & LPF_FAILOUT)
2811 linkea_del_buf(ldata, cname);
2814 ns->ln_dirent_repaired += dirent_count;
2819 lfsck_ibits_unlock(lh, LCK_EX);
2820 /* The name entry references another MDT-object that may be
2821 * created by the LFSCK for repairing dangling name entry.
2822 * Try to replace it. */
2823 rc = lfsck_namespace_replace_cond(env, com, parent, child,
2825 lfsck_object_put(env, parent);
2832 linkea_del_buf(ldata, cname);
2835 linkea_first_entry(ldata);
2836 if (ldata->ld_leh->leh_reccount == 1) {
2837 rc = lfsck_namespace_dsd_single(env, com, child, pfid, ldata,
2840 if (rc == 0 && fid_is_zero(pfid) && linkea_count > 0)
2841 rc = lfsck_namespace_rebuild_linkea(env, com, child,
2847 /* All linkEA entries are invalid and removed, then handle the @child
2849 if (ldata->ld_leh->leh_reccount == 0) {
2850 rc = lfsck_namespace_dsd_orphan(env, com, child, pfid, lh,
2856 /* If the dangling name entry for the orphan directory object has
2857 * been remvoed, then just check whether the directory object is
2858 * still under the .lustre/lost+found/MDTxxxx/ or not. */
2864 /* There is no linkEA entry that matches the ".." name entry. Find
2865 * the first linkEA entry that both parent and name entry exist to
2866 * rebuild a new ".." name entry. */
2876 * Repair the object's nlink attribute.
2878 * If all the known name entries have been verified, then the object's hard
2879 * link attribute should match the object's linkEA entries count unless the
2880 * object's has too much hard link to be recorded in the linkEA. Such cases
2881 * should have been marked in the LFSCK trace file. Otherwise, trust the
2882 * linkEA to update the object's nlink attribute.
2884 * \param[in] env pointer to the thread context
2885 * \param[in] com pointer to the lfsck component
2886 * \param[in] obj pointer to the dt_object to be handled
2887 * \param[in,out] la pointer to buffer to object's attribute before
2888 * and after the repairing
2890 * \retval positive number for repaired cases
2891 * \retval 0 if nothing to be repaired
2892 * \retval negative error number on failure
2894 static int lfsck_namespace_repair_nlink(const struct lu_env *env,
2895 struct lfsck_component *com,
2896 struct dt_object *obj,
2899 struct lfsck_thread_info *info = lfsck_env_info(env);
2900 struct lu_fid *tfid = &info->lti_fid3;
2901 struct lfsck_namespace *ns = com->lc_file_ram;
2902 struct lfsck_instance *lfsck = com->lc_lfsck;
2903 struct dt_device *dev = lfsck_obj2dev(obj);
2904 const struct lu_fid *cfid = lfsck_dto2fid(obj);
2905 struct thandle *th = NULL;
2906 struct linkea_data ldata = { NULL };
2907 struct lustre_handle lh = { 0 };
2908 __u32 old = la->la_nlink;
2914 LASSERT(!dt_object_remote(obj));
2915 LASSERT(S_ISREG(lfsck_object_type(obj)));
2917 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
2918 MDS_INODELOCK_UPDATE, LCK_PW);
2922 th = dt_trans_create(env, dev);
2924 GOTO(log, rc = PTR_ERR(th));
2926 la->la_valid = LA_NLINK;
2927 rc = dt_declare_attr_set(env, obj, la, th);
2931 rc = dt_trans_start_local(env, dev, th);
2935 dt_write_lock(env, obj, 0);
2936 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
2937 * ever tried to verify some remote MDT-object that resides on this
2938 * MDT, but this MDT failed to respond such request. So means there
2939 * may be some remote name entry on other MDT that references this
2940 * object with another name, so we cannot know whether this linkEA
2941 * is valid or not. So keep it there and maybe resolved when next
2943 if (ns->ln_flags & LF_INCOMPLETE)
2944 GOTO(unlock, rc = 0);
2946 fid_cpu_to_be(tfid, cfid);
2947 idx = lfsck_sub_trace_file_fid2idx(cfid);
2948 rc = dt_lookup(env, com->lc_sub_trace_objs[idx].lsto_obj,
2949 (struct dt_rec *)&flags, (const struct dt_key *)tfid);
2953 if (flags & LNTF_SKIP_NLINK)
2954 GOTO(unlock, rc = 0);
2956 rc = dt_attr_get(env, obj, la);
2958 GOTO(unlock, rc = (rc == -ENOENT ? 0 : rc));
2960 rc = lfsck_links_read2(env, obj, &ldata);
2962 GOTO(unlock, rc = (rc == -ENODATA ? 0 : rc));
2964 if (la->la_nlink == ldata.ld_leh->leh_reccount ||
2965 unlikely(la->la_nlink == 0))
2966 GOTO(unlock, rc = 0);
2968 la->la_nlink = ldata.ld_leh->leh_reccount;
2969 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2970 GOTO(unlock, rc = 1);
2972 rc = dt_attr_set(env, obj, la, th);
2974 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2977 dt_write_unlock(env, obj);
2980 dt_trans_stop(env, dev, th);
2983 lfsck_ibits_unlock(&lh, LCK_PW);
2984 CDEBUG(D_LFSCK, "%s: namespace LFSCK repaired the object "DFID"'s "
2985 "nlink count from %u to %u: rc = %d\n",
2986 lfsck_lfsck2name(lfsck), PFID(cfid), old, la->la_nlink, rc);
2989 ns->ln_flags |= LF_INCONSISTENT;
2995 * Double scan the directory object for namespace LFSCK.
2997 * This function will verify the <parent, child> pairs in the namespace tree:
2998 * the parent references the child via some name entry that should be in the
2999 * child's linkEA entry, the child should back references the parent via its
3002 * The LFSCK will scan every linkEA entry in turn until find out the first
3003 * matched pairs. If found, then all other linkEA entries will be dropped.
3004 * If all the linkEA entries cannot match the ".." name entry, then there
3005 * are serveral possible cases:
3007 * 1) If there is only one linkEA entry, then trust it as long as the PFID
3008 * in the linkEA entry is valid.
3010 * 2) If there are multiple linkEA entries, then try to find the linkEA
3011 * that matches the ".." name entry. If found, then all other entries
3012 * are invalid; otherwise, it is quite possible that the ".." name entry
3013 * is corrupted. Under such case, the LFSCK will rebuild the ".." name
3014 * entry according to the first valid linkEA entry (both the parent and
3015 * the name entry should exist).
3017 * 3) If the directory object has no (valid) linkEA entry, then the
3018 * directory object will be handled as pure orphan and inserted
3019 * in the .lustre/lost+found/MDTxxxx/ with the name:
3020 * ${self_FID}-${PFID}-D-${conflict_version}
3022 * \param[in] env pointer to the thread context
3023 * \param[in] com pointer to the lfsck component
3024 * \param[in] child pointer to the directory object to be handled
3025 * \param[in] flags to indicate the specical checking on the @child
3027 * \retval positive number for repaired cases
3028 * \retval 0 if nothing to be repaired
3029 * \retval negative error number on failure
3031 static int lfsck_namespace_double_scan_dir(const struct lu_env *env,
3032 struct lfsck_component *com,
3033 struct dt_object *child, __u8 flags)
3035 struct lfsck_thread_info *info = lfsck_env_info(env);
3036 const struct lu_fid *cfid = lfsck_dto2fid(child);
3037 struct lu_fid *pfid = &info->lti_fid2;
3038 struct lfsck_namespace *ns = com->lc_file_ram;
3039 struct lfsck_instance *lfsck = com->lc_lfsck;
3040 struct lustre_handle lh = { 0 };
3041 struct linkea_data ldata = { NULL };
3042 bool unknown = false;
3045 enum lfsck_namespace_inconsistency_type type = LNIT_BAD_LINKEA;
3049 LASSERT(!dt_object_remote(child));
3051 if (flags & LNTF_UNCERTAIN_LMV) {
3052 if (flags & LNTF_RECHECK_NAME_HASH) {
3053 rc = lfsck_namespace_scan_shard(env, com, child);
3057 ns->ln_striped_shards_scanned++;
3059 ns->ln_striped_shards_skipped++;
3063 flags &= ~(LNTF_RECHECK_NAME_HASH | LNTF_UNCERTAIN_LMV);
3067 if (flags & (LNTF_CHECK_LINKEA | LNTF_CHECK_PARENT) &&
3068 !(lfsck->li_bookmark_ram.lb_param & LPF_ALL_TGT)) {
3069 CDEBUG(D_LFSCK, "%s: some MDT(s) maybe NOT take part in the"
3070 "the namespace LFSCK, then the LFSCK cannot guarantee"
3071 "all the name entries have been verified in first-stage"
3072 "scanning. So have to skip orphan related handling for"
3073 "the directory object "DFID" with remote name entry\n",
3074 lfsck_lfsck2name(lfsck), PFID(cfid));
3079 if (unlikely(!dt_try_as_dir(env, child)))
3080 GOTO(out, rc = -ENOTDIR);
3082 /* We only take ldlm lock on the @child when required. When the
3083 * logic comes here for the first time, it is always false. */
3087 rc = lfsck_ibits_lock(env, lfsck, child, &lh,
3088 MDS_INODELOCK_UPDATE |
3089 MDS_INODELOCK_XATTR, LCK_EX);
3094 dt_read_lock(env, child, 0);
3095 if (unlikely(lfsck_is_dead_obj(child))) {
3096 dt_read_unlock(env, child);
3101 rc = dt_lookup(env, child, (struct dt_rec *)pfid,
3102 (const struct dt_key *)dotdot);
3104 if (rc != -ENOENT && rc != -ENODATA && rc != -EINVAL) {
3105 dt_read_unlock(env, child);
3110 if (!lustre_handle_is_used(&lh)) {
3111 dt_read_unlock(env, child);
3116 } else if (lfsck->li_lpf_obj != NULL &&
3117 lu_fid_eq(pfid, lfsck_dto2fid(lfsck->li_lpf_obj))) {
3119 } else if (unlikely(!fid_is_sane(pfid))) {
3123 rc = lfsck_links_read(env, child, &ldata);
3124 dt_read_unlock(env, child);
3126 if (rc != -ENODATA && rc != -EINVAL)
3129 if (!lustre_handle_is_used(&lh))
3132 if (rc == -EINVAL && !fid_is_zero(pfid)) {
3133 /* Remove the corrupted linkEA. */
3134 rc = lfsck_namespace_links_remove(env, com, child);
3136 /* Here, because of the crashed linkEA, we
3137 * cannot know whether there is some parent
3138 * that references the child directory via
3139 * some name entry or not. So keep it there,
3140 * when the LFSCK run next time, if there is
3141 * some parent that references this object,
3142 * then the LFSCK can rebuild the linkEA;
3143 * otherwise, this object will be handled
3144 * as orphan as above. */
3147 /* 1. If we have neither ".." nor linkEA,
3148 * then it is an orphan.
3150 * 2. If we only have the ".." name entry,
3151 * but no parent references this child
3152 * directory, then handle it as orphan. */
3153 lfsck_ibits_unlock(&lh, LCK_EX);
3154 type = LNIT_MUL_REF;
3156 /* If the LFSCK is marked as LF_INCOMPLETE,
3157 * then means some MDT has ever tried to
3158 * verify some remote MDT-object that resides
3159 * on this MDT, but this MDT failed to respond
3160 * such request. So means there may be some
3161 * remote name entry on other MDT that
3162 * references this object with another name,
3163 * so we cannot know whether this linkEA is
3164 * valid or not. So keep it there and maybe
3165 * resolved when next LFSCK run. */
3166 if (ns->ln_flags & LF_INCOMPLETE)
3169 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
3170 "-"DFID, PFID(pfid));
3171 rc = lfsck_namespace_insert_orphan(env, com, child,
3172 info->lti_tmpbuf, "D", NULL);
3178 linkea_first_entry(&ldata);
3179 /* This is the most common case: the object has unique linkEA entry. */
3180 if (ldata.ld_leh->leh_reccount == 1) {
3181 rc = lfsck_namespace_dsd_single(env, com, child, pfid, &ldata,
3182 &lh, &type, &retry);
3184 LASSERT(!lustre_handle_is_used(&lh));
3193 if (!lustre_handle_is_used(&lh))
3196 if (unlikely(ldata.ld_leh->leh_reccount == 0)) {
3197 rc = lfsck_namespace_dsd_orphan(env, com, child, pfid, &lh,
3203 /* When we come here, the cases usually like that:
3204 * 1) The directory object has a corrupted linkEA entry. During the
3205 * first-stage scanning, the LFSCK cannot know such corruption,
3206 * then it appends the right linkEA entry according to the found
3207 * name entry after the bad one.
3209 * 2) The directory object has a right linkEA entry. During the
3210 * first-stage scanning, the LFSCK finds some bad name entry,
3211 * but the LFSCK cannot aware that at that time, then it adds
3212 * the bad linkEA entry for further processing. */
3213 rc = lfsck_namespace_dsd_multiple(env, com, child, pfid, &ldata,
3219 lfsck_ibits_unlock(&lh, LCK_EX);
3222 case LNIT_BAD_LINKEA:
3223 ns->ln_linkea_repaired++;
3225 case LNIT_UNMATCHED_PAIRS:
3226 ns->ln_unmatched_pairs_repaired++;
3229 ns->ln_mul_ref_repaired++;
3237 ns->ln_unknown_inconsistency++;
3243 * Double scan the MDT-object for namespace LFSCK.
3245 * If the MDT-object contains invalid or repeated linkEA entries, then drop
3246 * those entries from the linkEA; if the linkEA becomes empty or the object
3247 * has no linkEA, then it is an orphan and will be added into the directory
3248 * .lustre/lost+found/MDTxxxx/; if the remote parent is lost, then recreate
3249 * the remote parent; if the name entry corresponding to some linkEA entry
3250 * is lost, then add the name entry back to the namespace.
3252 * \param[in] env pointer to the thread context
3253 * \param[in] com pointer to the lfsck component
3254 * \param[in] child pointer to the dt_object to be handled
3255 * \param[in] flags some hints to indicate how the @child should be handled
3257 * \retval positive number for repaired cases
3258 * \retval 0 if nothing to be repaired
3259 * \retval negative error number on failure
3261 static int lfsck_namespace_double_scan_one(const struct lu_env *env,
3262 struct lfsck_component *com,
3263 struct dt_object *child, __u8 flags)
3265 struct lfsck_thread_info *info = lfsck_env_info(env);
3266 struct lu_attr *la = &info->lti_la;
3267 struct lu_name *cname = &info->lti_name;
3268 struct lu_fid *pfid = &info->lti_fid;
3269 struct lu_fid *cfid = &info->lti_fid2;
3270 struct lfsck_instance *lfsck = com->lc_lfsck;
3271 struct lfsck_namespace *ns = com->lc_file_ram;
3272 struct dt_object *parent = NULL;
3273 struct linkea_data ldata = { NULL };
3274 bool repaired = false;
3279 dt_read_lock(env, child, 0);
3280 if (unlikely(lfsck_is_dead_obj(child))) {
3281 dt_read_unlock(env, child);
3286 if (S_ISDIR(lfsck_object_type(child))) {
3287 dt_read_unlock(env, child);
3288 rc = lfsck_namespace_double_scan_dir(env, com, child, flags);
3293 rc = lfsck_links_read(env, child, &ldata);
3294 dt_read_unlock(env, child);
3298 linkea_first_entry(&ldata);
3299 while (ldata.ld_lee != NULL) {
3300 lfsck_namespace_unpack_linkea_entry(&ldata, cname, pfid,
3302 rc = lfsck_namespace_filter_linkea_entry(&ldata, cname, pfid,
3304 /* Found repeated linkEA entries */
3306 rc = lfsck_namespace_shrink_linkea(env, com, child,
3307 &ldata, cname, pfid, false);
3319 /* Invalid PFID in the linkEA entry. */
3320 if (!fid_is_sane(pfid)) {
3321 rc = lfsck_namespace_shrink_linkea(env, com, child,
3322 &ldata, cname, pfid, true);
3332 parent = lfsck_object_find_bottom(env, lfsck, pfid);
3334 GOTO(out, rc = PTR_ERR(parent));
3336 if (!dt_object_exists(parent)) {
3339 if (ldata.ld_leh->leh_reccount > 1) {
3340 /* If it is NOT the last linkEA entry, then
3341 * there is still other chance to make the
3342 * child to be visible via other parent, then
3343 * remove this linkEA entry. */
3344 rc = lfsck_namespace_shrink_linkea(env, com,
3345 child, &ldata, cname, pfid, true);
3347 /* If the LFSCK is marked as LF_INCOMPLETE,
3348 * then means some MDT has ever tried to
3349 * verify some remote MDT-object that resides
3350 * on this MDT, but this MDT failed to respond
3351 * such request. So means there may be some
3352 * remote name entry on other MDT that
3353 * references this object with another name,
3354 * so we cannot know whether this linkEA is
3355 * valid or not. So keep it there and maybe
3356 * resolved when next LFSCK run. */
3357 if (ns->ln_flags & LF_INCOMPLETE) {
3358 lfsck_object_put(env, parent);
3363 /* Create the lost parent as an orphan. */
3364 rc = lfsck_namespace_create_orphan_dir(env, com,
3367 lfsck_object_put(env, parent);
3375 /* Add the missing name entry to the parent. */
3376 rc = lfsck_namespace_insert_normal(env, com,
3377 parent, child, cname->ln_name);
3378 if (unlikely(rc == -EEXIST))
3379 /* Unfortunately, someone reused the
3380 * name under the parent by race. So we
3381 * have to remove the linkEA entry from
3382 * current child object. It means that
3383 * the LFSCK cannot recover the system
3384 * totally back to its original status,
3385 * but it is necessary to make the
3386 * current system to be consistent. */
3387 rc = lfsck_namespace_shrink_linkea(env,
3391 linkea_next_entry(&ldata);
3394 lfsck_object_put(env, parent);
3404 /* The linkEA entry with bad parent will be removed. */
3405 if (unlikely(!dt_try_as_dir(env, parent))) {
3406 lfsck_object_put(env, parent);
3407 rc = lfsck_namespace_shrink_linkea(env, com, child,
3408 &ldata, cname, pfid, true);
3418 rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
3419 (const struct dt_key *)cname->ln_name);
3420 if (rc != 0 && rc != -ENOENT) {
3421 lfsck_object_put(env, parent);
3427 if (lu_fid_eq(cfid, lfsck_dto2fid(child))) {
3428 /* It is the most common case that we
3429 * find the name entry corresponding
3430 * to the linkEA entry. */
3431 lfsck_object_put(env, parent);
3432 linkea_next_entry(&ldata);
3434 /* The name entry references another
3435 * MDT-object that may be created by
3436 * the LFSCK for repairing dangling
3437 * name entry. Try to replace it. */
3438 rc = lfsck_namespace_replace_cond(env, com,
3439 parent, child, cfid, cname);
3440 lfsck_object_put(env, parent);
3446 linkea_next_entry(&ldata);
3448 rc = lfsck_namespace_shrink_linkea(env,
3462 rc = dt_attr_get(env, child, la);
3466 /* If there is no name entry in the parent dir and the object
3467 * link count is less than the linkea entries count, then the
3468 * linkea entry should be removed. */
3469 if (ldata.ld_leh->leh_reccount > la->la_nlink) {
3470 rc = lfsck_namespace_shrink_linkea_cond(env, com,
3471 parent, child, &ldata, cname, pfid);
3472 lfsck_object_put(env, parent);
3482 /* If the LFSCK is marked as LF_INCOMPLETE, then means some
3483 * MDT has ever tried to verify some remote MDT-object that
3484 * resides on this MDT, but this MDT failed to respond such
3485 * request. So means there may be some remote name entry on
3486 * other MDT that references this object with another name,
3487 * so we cannot know whether this linkEA is valid or not.
3488 * So keep it there and maybe resolved when next LFSCK run. */
3489 if (ns->ln_flags & LF_INCOMPLETE) {
3490 lfsck_object_put(env, parent);
3495 rc = lfsck_namespace_check_name(env, parent, child, cname);
3500 lfsck_object_put(env, parent);
3505 /* It is an invalid name entry, drop it. */
3506 if (unlikely(rc > 0)) {
3507 lfsck_object_put(env, parent);
3508 rc = lfsck_namespace_shrink_linkea(env, com, child,
3509 &ldata, cname, pfid, true);
3519 /* Add the missing name entry back to the namespace. */
3520 rc = lfsck_namespace_insert_normal(env, com, parent, child,
3522 if (unlikely(rc == -ESTALE))
3523 /* It may happen when the remote object has been
3524 * removed, but the local MDT is not aware of that. */
3527 if (unlikely(rc == -EEXIST))
3528 /* Unfortunately, someone reused the name under the
3529 * parent by race. So we have to remove the linkEA
3530 * entry from current child object. It means that the
3531 * LFSCK cannot recover the system totally back to
3532 * its original status, but it is necessary to make
3533 * the current system to be consistent.
3535 * It also may be because of the LFSCK found some
3536 * internal status of create operation. Under such
3537 * case, nothing to be done. */
3538 rc = lfsck_namespace_shrink_linkea_cond(env, com,
3539 parent, child, &ldata, cname, pfid);
3541 linkea_next_entry(&ldata);
3543 lfsck_object_put(env, parent);
3554 if (rc < 0 && rc != -ENODATA)
3558 LASSERT(ldata.ld_leh != NULL);
3560 count = ldata.ld_leh->leh_reccount;
3564 /* If the LFSCK is marked as LF_INCOMPLETE, then means some
3565 * MDT has ever tried to verify some remote MDT-object that
3566 * resides on this MDT, but this MDT failed to respond such
3567 * request. So means there may be some remote name entry on
3568 * other MDT that references this object with another name,
3569 * so we cannot know whether this linkEA is valid or not.
3570 * So keep it there and maybe resolved when next LFSCK run. */
3571 if (!(ns->ln_flags & LF_INCOMPLETE)) {
3572 /* If the child becomes orphan, then insert it into
3573 * the global .lustre/lost+found/MDTxxxx directory. */
3574 rc = lfsck_namespace_insert_orphan(env, com, child,
3580 ns->ln_mul_ref_repaired++;
3585 rc = dt_attr_get(env, child, la);
3589 if (la->la_nlink != 0 && la->la_nlink != count) {
3590 rc = lfsck_namespace_repair_nlink(env, com, child, la);
3592 ns->ln_objs_nlink_repaired++;
3599 if (la->la_nlink > 1)
3600 ns->ln_mul_linked_repaired++;
3609 static void lfsck_namespace_dump_statistics(struct seq_file *m,
3610 struct lfsck_namespace *ns,
3611 __u64 checked_phase1,
3612 __u64 checked_phase2,
3616 seq_printf(m, "checked_phase1: "LPU64"\n"
3617 "checked_phase2: "LPU64"\n"
3618 "updated_phase1: "LPU64"\n"
3619 "updated_phase2: "LPU64"\n"
3620 "failed_phase1: "LPU64"\n"
3621 "failed_phase2: "LPU64"\n"
3622 "directories: "LPU64"\n"
3623 "dirent_repaired: "LPU64"\n"
3624 "linkea_repaired: "LPU64"\n"
3625 "nlinks_repaired: "LPU64"\n"
3626 "multiple_linked_checked: "LPU64"\n"
3627 "multiple_linked_repaired: "LPU64"\n"
3628 "unknown_inconsistency: "LPU64"\n"
3629 "unmatched_pairs_repaired: "LPU64"\n"
3630 "dangling_repaired: "LPU64"\n"
3631 "multiple_referenced_repaired: "LPU64"\n"
3632 "bad_file_type_repaired: "LPU64"\n"
3633 "lost_dirent_repaired: "LPU64"\n"
3634 "local_lost_found_scanned: "LPU64"\n"
3635 "local_lost_found_moved: "LPU64"\n"
3636 "local_lost_found_skipped: "LPU64"\n"
3637 "local_lost_found_failed: "LPU64"\n"
3638 "striped_dirs_scanned: "LPU64"\n"
3639 "striped_dirs_repaired: "LPU64"\n"
3640 "striped_dirs_failed: "LPU64"\n"
3641 "striped_dirs_disabled: "LPU64"\n"
3642 "striped_dirs_skipped: "LPU64"\n"
3643 "striped_shards_scanned: "LPU64"\n"
3644 "striped_shards_repaired: "LPU64"\n"
3645 "striped_shards_failed: "LPU64"\n"
3646 "striped_shards_skipped: "LPU64"\n"
3647 "name_hash_repaired: "LPU64"\n"
3648 "success_count: %u\n"
3649 "run_time_phase1: %u seconds\n"
3650 "run_time_phase2: %u seconds\n",
3653 ns->ln_items_repaired,
3654 ns->ln_objs_repaired_phase2,
3655 ns->ln_items_failed,
3656 ns->ln_objs_failed_phase2,
3657 ns->ln_dirs_checked,
3658 ns->ln_dirent_repaired,
3659 ns->ln_linkea_repaired,
3660 ns->ln_objs_nlink_repaired,
3661 ns->ln_mul_linked_checked,
3662 ns->ln_mul_linked_repaired,
3663 ns->ln_unknown_inconsistency,
3664 ns->ln_unmatched_pairs_repaired,
3665 ns->ln_dangling_repaired,
3666 ns->ln_mul_ref_repaired,
3667 ns->ln_bad_type_repaired,
3668 ns->ln_lost_dirent_repaired,
3669 ns->ln_local_lpf_scanned,
3670 ns->ln_local_lpf_moved,
3671 ns->ln_local_lpf_skipped,
3672 ns->ln_local_lpf_failed,
3673 ns->ln_striped_dirs_scanned,
3674 ns->ln_striped_dirs_repaired,
3675 ns->ln_striped_dirs_failed,
3676 ns->ln_striped_dirs_disabled,
3677 ns->ln_striped_dirs_skipped,
3678 ns->ln_striped_shards_scanned,
3679 ns->ln_striped_shards_repaired,
3680 ns->ln_striped_shards_failed,
3681 ns->ln_striped_shards_skipped,
3682 ns->ln_name_hash_repaired,
3683 ns->ln_success_count,
3688 static void lfsck_namespace_release_lmv(const struct lu_env *env,
3689 struct lfsck_component *com)
3691 struct lfsck_instance *lfsck = com->lc_lfsck;
3692 struct lfsck_namespace *ns = com->lc_file_ram;
3694 while (!list_empty(&lfsck->li_list_lmv)) {
3695 struct lfsck_lmv_unit *llu;
3696 struct lfsck_lmv *llmv;
3698 llu = list_entry(lfsck->li_list_lmv.next,
3699 struct lfsck_lmv_unit, llu_link);
3700 llmv = &llu->llu_lmv;
3702 LASSERTF(atomic_read(&llmv->ll_ref) == 1,
3703 "still in using: %u\n",
3704 atomic_read(&llmv->ll_ref));
3706 ns->ln_striped_dirs_skipped++;
3707 lfsck_lmv_put(env, llmv);
3711 static int lfsck_namespace_check_for_double_scan(const struct lu_env *env,
3712 struct lfsck_component *com,
3713 struct dt_object *obj)
3715 struct lu_attr *la = &lfsck_env_info(env)->lti_la;
3718 rc = dt_attr_get(env, obj, la);
3722 /* zero-linkEA object may be orphan, but it also maybe because
3723 * of upgrading. Currently, we cannot record it for double scan.
3724 * Because it may cause the LFSCK trace file to be too large. */
3726 /* "la_ctime" == 1 means that it has ever been removed from
3727 * backend /lost+found directory but not been added back to
3728 * the normal namespace yet. */
3730 if ((S_ISREG(lfsck_object_type(obj)) && la->la_nlink > 1) ||
3731 unlikely(la->la_ctime == 1))
3732 rc = lfsck_namespace_trace_update(env, com, lfsck_dto2fid(obj),
3733 LNTF_CHECK_LINKEA, true);
3738 /* namespace APIs */
3740 static int lfsck_namespace_reset(const struct lu_env *env,
3741 struct lfsck_component *com, bool init)
3743 struct lfsck_instance *lfsck = com->lc_lfsck;
3744 struct lfsck_namespace *ns = com->lc_file_ram;
3745 struct lfsck_assistant_data *lad = com->lc_data;
3746 struct dt_object *root;
3747 struct dt_object *dto;
3751 root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
3753 GOTO(log, rc = PTR_ERR(root));
3755 if (unlikely(!dt_try_as_dir(env, root)))
3756 GOTO(put, rc = -ENOTDIR);
3758 down_write(&com->lc_sem);
3760 memset(ns, 0, sizeof(*ns));
3762 __u32 count = ns->ln_success_count;
3763 __u64 last_time = ns->ln_time_last_complete;
3765 memset(ns, 0, sizeof(*ns));
3766 ns->ln_success_count = count;
3767 ns->ln_time_last_complete = last_time;
3769 ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
3770 ns->ln_status = LS_INIT;
3772 lfsck_object_put(env, com->lc_obj);
3774 dto = lfsck_namespace_load_one_trace_file(env, com, root,
3775 LFSCK_NAMESPACE, true);
3777 GOTO(out, rc = PTR_ERR(dto));
3780 rc = lfsck_namespace_load_sub_trace_files(env, com, true);
3784 lad->lad_incomplete = 0;
3785 CFS_RESET_BITMAP(lad->lad_bitmap);
3787 rc = lfsck_namespace_store(env, com, true);
3792 up_write(&com->lc_sem);
3795 lfsck_object_put(env, root);
3797 CDEBUG(D_LFSCK, "%s: namespace LFSCK reset: rc = %d\n",
3798 lfsck_lfsck2name(lfsck), rc);
3803 lfsck_namespace_fail(const struct lu_env *env, struct lfsck_component *com,
3806 struct lfsck_namespace *ns = com->lc_file_ram;
3808 down_write(&com->lc_sem);
3810 com->lc_new_checked++;
3811 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
3812 up_write(&com->lc_sem);
3815 static void lfsck_namespace_close_dir(const struct lu_env *env,
3816 struct lfsck_component *com)
3818 struct lfsck_namespace *ns = com->lc_file_ram;
3819 struct lfsck_assistant_data *lad = com->lc_data;
3820 struct lfsck_assistant_object *lso = NULL;
3821 struct lfsck_instance *lfsck = com->lc_lfsck;
3822 struct lfsck_lmv *llmv = lfsck->li_lmv;
3823 struct lfsck_namespace_req *lnr;
3825 sizeof(*lnr) + LFSCK_TMPBUF_LEN;
3826 bool wakeup = false;
3832 OBD_ALLOC(lnr, size);
3834 ns->ln_striped_dirs_skipped++;
3839 lso = lfsck_assistant_object_init(env, lfsck_dto2fid(lfsck->li_obj_dir),
3840 NULL, lfsck->li_pos_current.lp_oit_cookie, true);
3842 OBD_FREE(lnr, size);
3843 ns->ln_striped_dirs_skipped++;
3848 /* Generate a dummy request to indicate that all shards' name entry
3849 * in this striped directory has been scanned for the first time. */
3850 INIT_LIST_HEAD(&lnr->lnr_lar.lar_list);
3851 lnr->lnr_lar.lar_parent = lso;
3852 lnr->lnr_lmv = lfsck_lmv_get(llmv);
3853 lnr->lnr_fid = *lfsck_dto2fid(lfsck->li_obj_dir);
3854 lnr->lnr_dir_cookie = MDS_DIR_END_OFF;
3855 lnr->lnr_size = size;
3857 spin_lock(&lad->lad_lock);
3858 if (lad->lad_assistant_status < 0) {
3859 spin_unlock(&lad->lad_lock);
3860 lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar);
3861 ns->ln_striped_dirs_skipped++;
3866 list_add_tail(&lnr->lnr_lar.lar_list, &lad->lad_req_list);
3867 if (lad->lad_prefetched == 0)
3870 lad->lad_prefetched++;
3871 spin_unlock(&lad->lad_lock);
3873 wake_up_all(&lad->lad_thread.t_ctl_waitq);
3878 static int lfsck_namespace_open_dir(const struct lu_env *env,
3879 struct lfsck_component *com)
3881 struct lfsck_instance *lfsck = com->lc_lfsck;
3882 struct lfsck_namespace *ns = com->lc_file_ram;
3883 struct lfsck_lmv *llmv = lfsck->li_lmv;
3890 if (llmv->ll_lmv_master) {
3891 struct lmv_mds_md_v1 *lmv = &llmv->ll_lmv;
3893 if (lmv->lmv_master_mdt_index != lfsck_dev_idx(lfsck)) {
3894 lmv->lmv_master_mdt_index =
3895 lfsck_dev_idx(lfsck);
3896 ns->ln_flags |= LF_INCONSISTENT;
3897 llmv->ll_lmv_updated = 1;
3900 rc = lfsck_namespace_verify_stripe_slave(env, com,
3901 lfsck->li_obj_dir, llmv);
3904 RETURN(rc > 0 ? 0 : rc);
3907 static int lfsck_namespace_checkpoint(const struct lu_env *env,
3908 struct lfsck_component *com, bool init)
3910 struct lfsck_instance *lfsck = com->lc_lfsck;
3911 struct lfsck_namespace *ns = com->lc_file_ram;
3915 rc = lfsck_checkpoint_generic(env, com);
3920 down_write(&com->lc_sem);
3922 ns->ln_pos_latest_start = lfsck->li_pos_checkpoint;
3924 ns->ln_pos_last_checkpoint = lfsck->li_pos_checkpoint;
3925 ns->ln_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
3926 HALF_SEC - lfsck->li_time_last_checkpoint);
3927 ns->ln_time_last_checkpoint = cfs_time_current_sec();
3928 ns->ln_items_checked += com->lc_new_checked;
3929 com->lc_new_checked = 0;
3932 rc = lfsck_namespace_store(env, com, false);
3933 up_write(&com->lc_sem);
3936 CDEBUG(D_LFSCK, "%s: namespace LFSCK checkpoint at the pos ["LPU64
3937 ", "DFID", "LPX64"]: rc = %d\n", lfsck_lfsck2name(lfsck),
3938 lfsck->li_pos_current.lp_oit_cookie,
3939 PFID(&lfsck->li_pos_current.lp_dir_parent),
3940 lfsck->li_pos_current.lp_dir_cookie, rc);
3942 return rc > 0 ? 0 : rc;
3945 static int lfsck_namespace_prep(const struct lu_env *env,
3946 struct lfsck_component *com,
3947 struct lfsck_start_param *lsp)
3949 struct lfsck_instance *lfsck = com->lc_lfsck;
3950 struct lfsck_namespace *ns = com->lc_file_ram;
3951 struct lfsck_position *pos = &com->lc_pos_start;
3954 rc = lfsck_namespace_load_bitmap(env, com);
3955 if (rc != 0 || ns->ln_status == LS_COMPLETED) {
3956 rc = lfsck_namespace_reset(env, com, false);
3958 rc = lfsck_set_param(env, lfsck, lsp->lsp_start, true);
3961 CDEBUG(D_LFSCK, "%s: namespace LFSCK prep failed: "
3962 "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
3968 down_write(&com->lc_sem);
3969 ns->ln_time_latest_start = cfs_time_current_sec();
3970 spin_lock(&lfsck->li_lock);
3972 if (ns->ln_flags & LF_SCANNED_ONCE) {
3973 if (!lfsck->li_drop_dryrun ||
3974 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent)) {
3975 ns->ln_status = LS_SCANNING_PHASE2;
3976 list_move_tail(&com->lc_link,
3977 &lfsck->li_list_double_scan);
3978 if (!list_empty(&com->lc_link_dir))
3979 list_del_init(&com->lc_link_dir);
3980 lfsck_pos_set_zero(pos);
3982 ns->ln_status = LS_SCANNING_PHASE1;
3983 ns->ln_run_time_phase1 = 0;
3984 ns->ln_run_time_phase2 = 0;
3985 ns->ln_items_checked = 0;
3986 ns->ln_items_repaired = 0;
3987 ns->ln_items_failed = 0;
3988 ns->ln_dirs_checked = 0;
3989 ns->ln_objs_checked_phase2 = 0;
3990 ns->ln_objs_repaired_phase2 = 0;
3991 ns->ln_objs_failed_phase2 = 0;
3992 ns->ln_objs_nlink_repaired = 0;
3993 ns->ln_dirent_repaired = 0;
3994 ns->ln_linkea_repaired = 0;
3995 ns->ln_mul_linked_checked = 0;
3996 ns->ln_mul_linked_repaired = 0;
3997 ns->ln_unknown_inconsistency = 0;
3998 ns->ln_unmatched_pairs_repaired = 0;
3999 ns->ln_dangling_repaired = 0;
4000 ns->ln_mul_ref_repaired = 0;
4001 ns->ln_bad_type_repaired = 0;
4002 ns->ln_lost_dirent_repaired = 0;
4003 ns->ln_striped_dirs_scanned = 0;
4004 ns->ln_striped_dirs_repaired = 0;
4005 ns->ln_striped_dirs_failed = 0;
4006 ns->ln_striped_dirs_disabled = 0;
4007 ns->ln_striped_dirs_skipped = 0;
4008 ns->ln_striped_shards_scanned = 0;
4009 ns->ln_striped_shards_repaired = 0;
4010 ns->ln_striped_shards_failed = 0;
4011 ns->ln_striped_shards_skipped = 0;
4012 ns->ln_name_hash_repaired = 0;
4013 fid_zero(&ns->ln_fid_latest_scanned_phase2);
4014 if (list_empty(&com->lc_link_dir))
4015 list_add_tail(&com->lc_link_dir,
4016 &lfsck->li_list_dir);
4017 *pos = ns->ln_pos_first_inconsistent;
4020 ns->ln_status = LS_SCANNING_PHASE1;
4021 if (list_empty(&com->lc_link_dir))
4022 list_add_tail(&com->lc_link_dir,
4023 &lfsck->li_list_dir);
4024 if (!lfsck->li_drop_dryrun ||
4025 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent)) {
4026 *pos = ns->ln_pos_last_checkpoint;
4027 pos->lp_oit_cookie++;
4029 *pos = ns->ln_pos_first_inconsistent;
4033 spin_unlock(&lfsck->li_lock);
4034 up_write(&com->lc_sem);
4036 rc = lfsck_start_assistant(env, com, lsp);
4038 CDEBUG(D_LFSCK, "%s: namespace LFSCK prep done, start pos ["LPU64", "
4039 DFID", "LPX64"]: rc = %d\n",
4040 lfsck_lfsck2name(lfsck), pos->lp_oit_cookie,
4041 PFID(&pos->lp_dir_parent), pos->lp_dir_cookie, rc);
4046 static int lfsck_namespace_exec_oit(const struct lu_env *env,
4047 struct lfsck_component *com,
4048 struct dt_object *obj)
4050 struct lfsck_thread_info *info = lfsck_env_info(env);
4051 struct lfsck_namespace *ns = com->lc_file_ram;
4052 struct lfsck_instance *lfsck = com->lc_lfsck;
4053 const struct lu_fid *fid = lfsck_dto2fid(obj);
4054 struct lu_fid *pfid = &info->lti_fid2;
4055 struct lu_name *cname = &info->lti_name;
4056 struct lu_seq_range *range = &info->lti_range;
4057 struct seq_server_site *ss = lfsck_dev_site(lfsck);
4058 struct linkea_data ldata = { NULL };
4059 __u32 idx = lfsck_dev_idx(lfsck);
4063 rc = lfsck_links_read(env, obj, &ldata);
4067 /* -EINVAL means crashed linkEA, should be verified. */
4068 if (rc == -EINVAL) {
4069 rc = lfsck_namespace_trace_update(env, com, fid,
4070 LNTF_CHECK_LINKEA, true);
4072 struct lustre_handle lh = { 0 };
4074 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
4075 MDS_INODELOCK_UPDATE |
4076 MDS_INODELOCK_XATTR, LCK_EX);
4078 rc = lfsck_namespace_links_remove(env, com,
4080 lfsck_ibits_unlock(&lh, LCK_EX);
4084 GOTO(out, rc = (rc == -ENOENT ? 0 : rc));
4087 if (rc == -ENODATA) {
4088 rc = lfsck_namespace_check_for_double_scan(env, com, obj);
4096 /* Record multiple-linked object. */
4097 if (ldata.ld_leh->leh_reccount > 1) {
4098 rc = lfsck_namespace_trace_update(env, com, fid,
4099 LNTF_CHECK_LINKEA, true);
4104 linkea_first_entry(&ldata);
4105 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, cname, pfid);
4106 if (!fid_is_sane(pfid)) {
4107 rc = lfsck_namespace_trace_update(env, com, fid,
4108 LNTF_CHECK_PARENT, true);
4110 fld_range_set_mdt(range);
4111 rc = fld_local_lookup(env, ss->ss_server_fld,
4112 fid_seq(pfid), range);
4113 if ((rc == -ENOENT) ||
4114 (rc == 0 && range->lsr_index != idx))
4115 rc = lfsck_namespace_trace_update(env, com, fid,
4116 LNTF_CHECK_LINKEA, true);
4118 rc = lfsck_namespace_check_for_double_scan(env, com,
4125 down_write(&com->lc_sem);
4126 if (S_ISDIR(lfsck_object_type(obj)))
4127 ns->ln_dirs_checked++;
4129 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
4130 up_write(&com->lc_sem);
4135 static int lfsck_namespace_exec_dir(const struct lu_env *env,
4136 struct lfsck_component *com,
4137 struct lfsck_assistant_object *lso,
4138 struct lu_dirent *ent, __u16 type)
4140 struct lfsck_assistant_data *lad = com->lc_data;
4141 struct lfsck_instance *lfsck = com->lc_lfsck;
4142 struct lfsck_namespace_req *lnr;
4143 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4144 struct ptlrpc_thread *mthread = &lfsck->li_thread;
4145 struct ptlrpc_thread *athread = &lad->lad_thread;
4146 struct l_wait_info lwi = { 0 };
4147 bool wakeup = false;
4149 l_wait_event(mthread->t_ctl_waitq,
4150 lad->lad_prefetched < bk->lb_async_windows ||
4151 !thread_is_running(mthread) ||
4152 thread_is_stopped(athread),
4155 if (unlikely(!thread_is_running(mthread)) ||
4156 thread_is_stopped(athread))
4159 if (unlikely(lfsck_is_dead_obj(lfsck->li_obj_dir)))
4162 lnr = lfsck_namespace_assistant_req_init(com->lc_lfsck, lso, ent, type);
4164 struct lfsck_namespace *ns = com->lc_file_ram;
4166 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
4167 return PTR_ERR(lnr);
4170 spin_lock(&lad->lad_lock);
4171 if (lad->lad_assistant_status < 0) {
4172 spin_unlock(&lad->lad_lock);
4173 lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar);
4174 return lad->lad_assistant_status;
4177 list_add_tail(&lnr->lnr_lar.lar_list, &lad->lad_req_list);
4178 if (lad->lad_prefetched == 0)
4181 lad->lad_prefetched++;
4182 spin_unlock(&lad->lad_lock);
4184 wake_up_all(&lad->lad_thread.t_ctl_waitq);
4186 down_write(&com->lc_sem);
4187 com->lc_new_checked++;
4188 up_write(&com->lc_sem);
4193 static int lfsck_namespace_post(const struct lu_env *env,
4194 struct lfsck_component *com,
4195 int result, bool init)
4197 struct lfsck_instance *lfsck = com->lc_lfsck;
4198 struct lfsck_namespace *ns = com->lc_file_ram;
4202 lfsck_post_generic(env, com, &result);
4204 down_write(&com->lc_sem);
4205 lfsck_namespace_release_lmv(env, com);
4207 spin_lock(&lfsck->li_lock);
4209 ns->ln_pos_last_checkpoint = lfsck->li_pos_checkpoint;
4211 ns->ln_status = LS_SCANNING_PHASE2;
4212 ns->ln_flags |= LF_SCANNED_ONCE;
4213 ns->ln_flags &= ~LF_UPGRADE;
4214 list_del_init(&com->lc_link_dir);
4215 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
4216 } else if (result == 0) {
4217 if (lfsck->li_status != 0)
4218 ns->ln_status = lfsck->li_status;
4220 ns->ln_status = LS_STOPPED;
4221 if (ns->ln_status != LS_PAUSED) {
4222 list_del_init(&com->lc_link_dir);
4223 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4226 ns->ln_status = LS_FAILED;
4227 list_del_init(&com->lc_link_dir);
4228 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4230 spin_unlock(&lfsck->li_lock);
4233 ns->ln_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
4234 HALF_SEC - lfsck->li_time_last_checkpoint);
4235 ns->ln_time_last_checkpoint = cfs_time_current_sec();
4236 ns->ln_items_checked += com->lc_new_checked;
4237 com->lc_new_checked = 0;
4240 rc = lfsck_namespace_store(env, com, false);
4241 up_write(&com->lc_sem);
4243 CDEBUG(D_LFSCK, "%s: namespace LFSCK post done: rc = %d\n",
4244 lfsck_lfsck2name(lfsck), rc);
4250 lfsck_namespace_dump(const struct lu_env *env, struct lfsck_component *com,
4253 struct lfsck_instance *lfsck = com->lc_lfsck;
4254 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4255 struct lfsck_namespace *ns = com->lc_file_ram;
4258 down_read(&com->lc_sem);
4259 seq_printf(m, "name: lfsck_namespace\n"
4265 lfsck_status2names(ns->ln_status));
4267 rc = lfsck_bits_dump(m, ns->ln_flags, lfsck_flags_names, "flags");
4271 rc = lfsck_bits_dump(m, bk->lb_param, lfsck_param_names, "param");
4275 rc = lfsck_time_dump(m, ns->ln_time_last_complete,
4280 rc = lfsck_time_dump(m, ns->ln_time_latest_start,
4285 rc = lfsck_time_dump(m, ns->ln_time_last_checkpoint,
4290 rc = lfsck_pos_dump(m, &ns->ln_pos_latest_start,
4291 "latest_start_position");
4295 rc = lfsck_pos_dump(m, &ns->ln_pos_last_checkpoint,
4296 "last_checkpoint_position");
4300 rc = lfsck_pos_dump(m, &ns->ln_pos_first_inconsistent,
4301 "first_failure_position");
4305 if (ns->ln_status == LS_SCANNING_PHASE1) {
4306 struct lfsck_position pos;
4307 const struct dt_it_ops *iops;
4308 cfs_duration_t duration = cfs_time_current() -
4309 lfsck->li_time_last_checkpoint;
4310 __u64 checked = ns->ln_items_checked + com->lc_new_checked;
4311 __u64 speed = checked;
4312 __u64 new_checked = com->lc_new_checked *
4313 msecs_to_jiffies(MSEC_PER_SEC);
4314 __u32 rtime = ns->ln_run_time_phase1 +
4315 cfs_duration_sec(duration + HALF_SEC);
4318 do_div(new_checked, duration);
4321 do_div(speed, rtime);
4323 lfsck_namespace_dump_statistics(m, ns, checked, 0, rtime, 0);
4324 seq_printf(m, "average_speed_phase1: "LPU64" items/sec\n"
4325 "average_speed_phase2: N/A\n"
4326 "average_speed_total: "LPU64" items/sec\n"
4327 "real_time_speed_phase1: "LPU64" items/sec\n"
4328 "real_time_speed_phase2: N/A\n",
4333 LASSERT(lfsck->li_di_oit != NULL);
4335 iops = &lfsck->li_obj_oit->do_index_ops->dio_it;
4337 /* The low layer otable-based iteration position may NOT
4338 * exactly match the namespace-based directory traversal
4339 * cookie. Generally, it is not a serious issue. But the
4340 * caller should NOT make assumption on that. */
4341 pos.lp_oit_cookie = iops->store(env, lfsck->li_di_oit);
4342 if (!lfsck->li_current_oit_processed)
4343 pos.lp_oit_cookie--;
4345 spin_lock(&lfsck->li_lock);
4346 if (lfsck->li_di_dir != NULL) {
4347 pos.lp_dir_cookie = lfsck->li_cookie_dir;
4348 if (pos.lp_dir_cookie >= MDS_DIR_END_OFF) {
4349 fid_zero(&pos.lp_dir_parent);
4350 pos.lp_dir_cookie = 0;
4353 *lfsck_dto2fid(lfsck->li_obj_dir);
4356 fid_zero(&pos.lp_dir_parent);
4357 pos.lp_dir_cookie = 0;
4359 spin_unlock(&lfsck->li_lock);
4360 lfsck_pos_dump(m, &pos, "current_position");
4361 } else if (ns->ln_status == LS_SCANNING_PHASE2) {
4362 cfs_duration_t duration = cfs_time_current() -
4363 com->lc_time_last_checkpoint;
4364 __u64 checked = ns->ln_objs_checked_phase2 +
4365 com->lc_new_checked;
4366 __u64 speed1 = ns->ln_items_checked;
4367 __u64 speed2 = checked;
4368 __u64 speed0 = speed1 + speed2;
4369 __u64 new_checked = com->lc_new_checked *
4370 msecs_to_jiffies(MSEC_PER_SEC);
4371 __u32 rtime = ns->ln_run_time_phase2 +
4372 cfs_duration_sec(duration + HALF_SEC);
4373 __u32 time0 = ns->ln_run_time_phase1 + rtime;
4376 do_div(new_checked, duration);
4378 if (ns->ln_run_time_phase1 != 0)
4379 do_div(speed1, ns->ln_run_time_phase1);
4380 else if (ns->ln_items_checked != 0)
4384 do_div(speed2, rtime);
4385 else if (checked != 0)
4389 do_div(speed0, time0);
4391 lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked,
4393 ns->ln_run_time_phase1, rtime);
4394 seq_printf(m, "average_speed_phase1: "LPU64" items/sec\n"
4395 "average_speed_phase2: "LPU64" objs/sec\n"
4396 "average_speed_total: "LPU64" items/sec\n"
4397 "real_time_speed_phase1: N/A\n"
4398 "real_time_speed_phase2: "LPU64" objs/sec\n"
4399 "current_position: "DFID"\n",
4404 PFID(&ns->ln_fid_latest_scanned_phase2));
4406 __u64 speed1 = ns->ln_items_checked;
4407 __u64 speed2 = ns->ln_objs_checked_phase2;
4408 __u64 speed0 = speed1 + speed2;
4409 __u32 time0 = ns->ln_run_time_phase1 + ns->ln_run_time_phase2;
4411 if (ns->ln_run_time_phase1 != 0)
4412 do_div(speed1, ns->ln_run_time_phase1);
4413 else if (ns->ln_items_checked != 0)
4416 if (ns->ln_run_time_phase2 != 0)
4417 do_div(speed2, ns->ln_run_time_phase2);
4418 else if (ns->ln_objs_checked_phase2 != 0)
4422 do_div(speed0, time0);
4424 lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked,
4425 ns->ln_objs_checked_phase2,
4426 ns->ln_run_time_phase1,
4427 ns->ln_run_time_phase2);
4428 seq_printf(m, "average_speed_phase1: "LPU64" items/sec\n"
4429 "average_speed_phase2: "LPU64" objs/sec\n"
4430 "average_speed_total: "LPU64" items/sec\n"
4431 "real_time_speed_phase1: N/A\n"
4432 "real_time_speed_phase2: N/A\n"
4433 "current_position: N/A\n",
4439 up_read(&com->lc_sem);
4443 static int lfsck_namespace_double_scan(const struct lu_env *env,
4444 struct lfsck_component *com)
4446 struct lfsck_namespace *ns = com->lc_file_ram;
4447 struct lfsck_assistant_data *lad = com->lc_data;
4448 struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
4449 struct lfsck_tgt_desc *ltd;
4450 struct lfsck_tgt_desc *next;
4453 rc = lfsck_double_scan_generic(env, com, ns->ln_status);
4454 if (thread_is_stopped(&lad->lad_thread)) {
4455 LASSERT(list_empty(&lad->lad_req_list));
4456 LASSERT(list_empty(&lad->lad_mdt_phase1_list));
4458 spin_lock(<ds->ltd_lock);
4459 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4460 ltd_namespace_phase_list) {
4461 list_del_init(<d->ltd_namespace_phase_list);
4463 spin_unlock(<ds->ltd_lock);
4469 static void lfsck_namespace_data_release(const struct lu_env *env,
4470 struct lfsck_component *com)
4472 struct lfsck_assistant_data *lad = com->lc_data;
4473 struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
4474 struct lfsck_tgt_desc *ltd;
4475 struct lfsck_tgt_desc *next;
4477 LASSERT(lad != NULL);
4478 LASSERT(thread_is_init(&lad->lad_thread) ||
4479 thread_is_stopped(&lad->lad_thread));
4480 LASSERT(list_empty(&lad->lad_req_list));
4482 com->lc_data = NULL;
4483 lfsck_namespace_release_lmv(env, com);
4485 spin_lock(<ds->ltd_lock);
4486 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
4487 ltd_namespace_phase_list) {
4488 list_del_init(<d->ltd_namespace_phase_list);
4490 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4491 ltd_namespace_phase_list) {
4492 list_del_init(<d->ltd_namespace_phase_list);
4494 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_list,
4495 ltd_namespace_list) {
4496 list_del_init(<d->ltd_namespace_list);
4498 spin_unlock(<ds->ltd_lock);
4500 if (likely(lad->lad_bitmap != NULL))
4501 CFS_FREE_BITMAP(lad->lad_bitmap);
4506 static void lfsck_namespace_quit(const struct lu_env *env,
4507 struct lfsck_component *com)
4509 struct lfsck_assistant_data *lad = com->lc_data;
4510 struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
4511 struct lfsck_tgt_desc *ltd;
4512 struct lfsck_tgt_desc *next;
4514 LASSERT(lad != NULL);
4516 lfsck_quit_generic(env, com);
4518 LASSERT(thread_is_init(&lad->lad_thread) ||
4519 thread_is_stopped(&lad->lad_thread));
4520 LASSERT(list_empty(&lad->lad_req_list));
4522 lfsck_namespace_release_lmv(env, com);
4524 spin_lock(<ds->ltd_lock);
4525 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
4526 ltd_namespace_phase_list) {
4527 list_del_init(<d->ltd_namespace_phase_list);
4529 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4530 ltd_namespace_phase_list) {
4531 list_del_init(<d->ltd_namespace_phase_list);
4533 spin_unlock(<ds->ltd_lock);
4536 static int lfsck_namespace_in_notify(const struct lu_env *env,
4537 struct lfsck_component *com,
4538 struct lfsck_request *lr,
4541 struct lfsck_instance *lfsck = com->lc_lfsck;
4542 struct lfsck_namespace *ns = com->lc_file_ram;
4543 struct lfsck_assistant_data *lad = com->lc_data;
4544 struct lfsck_tgt_descs *ltds = &lfsck->li_mdt_descs;
4545 struct lfsck_tgt_desc *ltd;
4550 switch (lr->lr_event) {
4551 case LE_SKIP_NLINK_DECLARE: {
4552 struct dt_object *obj;
4553 struct lu_fid *key = &lfsck_env_info(env)->lti_fid3;
4557 LASSERT(th != NULL);
4559 idx = lfsck_sub_trace_file_fid2idx(&lr->lr_fid);
4560 obj = com->lc_sub_trace_objs[idx].lsto_obj;
4561 fid_cpu_to_be(key, &lr->lr_fid);
4562 mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
4563 rc = dt_declare_delete(env, obj,
4564 (const struct dt_key *)key, th);
4566 rc = dt_declare_insert(env, obj,
4567 (const struct dt_rec *)&flags,
4568 (const struct dt_key *)key, th);
4569 mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
4573 case LE_SKIP_NLINK: {
4574 struct dt_object *obj;
4575 struct lu_fid *key = &lfsck_env_info(env)->lti_fid3;
4581 LASSERT(th != NULL);
4583 idx = lfsck_sub_trace_file_fid2idx(&lr->lr_fid);
4584 obj = com->lc_sub_trace_objs[idx].lsto_obj;
4585 fid_cpu_to_be(key, &lr->lr_fid);
4586 mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
4587 rc = dt_lookup(env, obj, (struct dt_rec *)&flags,
4588 (const struct dt_key *)key);
4590 if (flags & LNTF_SKIP_NLINK) {
4592 &com->lc_sub_trace_objs[idx].lsto_mutex);
4598 } else if (rc != -ENOENT) {
4602 flags |= LNTF_SKIP_NLINK;
4604 rc = dt_delete(env, obj, (const struct dt_key *)key,
4610 rc = dt_insert(env, obj, (const struct dt_rec *)&flags,
4611 (const struct dt_key *)key, th, 1);
4616 mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
4617 CDEBUG(D_LFSCK, "%s: RPC service thread mark the "DFID
4618 " to be skipped for namespace double scan: rc = %d\n",
4619 lfsck_lfsck2name(com->lc_lfsck), PFID(&lr->lr_fid), rc);
4622 /* If we cannot record this object in the LFSCK tracing,
4623 * we have to mark the LFSC as LF_INCOMPLETE, then the
4624 * LFSCK will skip nlink attribute verification for
4626 ns->ln_flags |= LF_INCOMPLETE;
4630 case LE_SET_LMV_MASTER: {
4631 struct dt_object *obj;
4633 obj = lfsck_object_find_bottom(env, lfsck, &lr->lr_fid);
4635 RETURN(PTR_ERR(obj));
4637 if (likely(dt_object_exists(obj)))
4638 rc = lfsck_namespace_notify_lmv_master_local(env, com,
4641 lfsck_object_put(env, obj);
4643 RETURN(rc > 0 ? 0 : rc);
4645 case LE_SET_LMV_SLAVE: {
4646 if (!(lr->lr_flags & LEF_RECHECK_NAME_HASH))
4647 ns->ln_striped_shards_repaired++;
4649 rc = lfsck_namespace_trace_update(env, com, &lr->lr_fid,
4650 LNTF_RECHECK_NAME_HASH, true);
4652 RETURN(rc > 0 ? 0 : rc);
4654 case LE_PHASE1_DONE:
4655 case LE_PHASE2_DONE:
4662 CDEBUG(D_LFSCK, "%s: namespace LFSCK handles notify %u from MDT %x, "
4663 "status %d, flags %x\n", lfsck_lfsck2name(lfsck), lr->lr_event,
4664 lr->lr_index, lr->lr_status, lr->lr_flags2);
4666 spin_lock(<ds->ltd_lock);
4667 ltd = lfsck_ltd2tgt(ltds, lr->lr_index);
4669 spin_unlock(<ds->ltd_lock);
4674 list_del_init(<d->ltd_namespace_phase_list);
4675 switch (lr->lr_event) {
4676 case LE_PHASE1_DONE:
4677 if (lr->lr_status <= 0) {
4678 ltd->ltd_namespace_done = 1;
4679 list_del_init(<d->ltd_namespace_list);
4680 CDEBUG(D_LFSCK, "%s: MDT %x failed/stopped at "
4681 "phase1 for namespace LFSCK: rc = %d.\n",
4682 lfsck_lfsck2name(lfsck),
4683 ltd->ltd_index, lr->lr_status);
4684 ns->ln_flags |= LF_INCOMPLETE;
4689 if (lr->lr_flags2 & LF_INCOMPLETE)
4690 ns->ln_flags |= LF_INCOMPLETE;
4692 if (list_empty(<d->ltd_namespace_list))
4693 list_add_tail(<d->ltd_namespace_list,
4694 &lad->lad_mdt_list);
4695 list_add_tail(<d->ltd_namespace_phase_list,
4696 &lad->lad_mdt_phase2_list);
4698 case LE_PHASE2_DONE:
4699 ltd->ltd_namespace_done = 1;
4700 list_del_init(<d->ltd_namespace_list);
4704 ltd->ltd_namespace_done = 1;
4705 list_del_init(<d->ltd_namespace_list);
4706 if (!(lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT)) {
4708 "%s: the peer MDT %x exit namespace LFSCK\n",
4709 lfsck_lfsck2name(lfsck), ltd->ltd_index);
4710 ns->ln_flags |= LF_INCOMPLETE;
4716 spin_unlock(<ds->ltd_lock);
4718 if (fail && lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) {
4719 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
4721 memset(stop, 0, sizeof(*stop));
4722 stop->ls_status = lr->lr_status;
4723 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
4724 lfsck_stop(env, lfsck->li_bottom, stop);
4725 } else if (lfsck_phase2_next_ready(lad)) {
4726 wake_up_all(&lad->lad_thread.t_ctl_waitq);
4732 static int lfsck_namespace_query(const struct lu_env *env,
4733 struct lfsck_component *com)
4735 struct lfsck_namespace *ns = com->lc_file_ram;
4737 return ns->ln_status;
4740 static struct lfsck_operations lfsck_namespace_ops = {
4741 .lfsck_reset = lfsck_namespace_reset,
4742 .lfsck_fail = lfsck_namespace_fail,
4743 .lfsck_close_dir = lfsck_namespace_close_dir,
4744 .lfsck_open_dir = lfsck_namespace_open_dir,
4745 .lfsck_checkpoint = lfsck_namespace_checkpoint,
4746 .lfsck_prep = lfsck_namespace_prep,
4747 .lfsck_exec_oit = lfsck_namespace_exec_oit,
4748 .lfsck_exec_dir = lfsck_namespace_exec_dir,
4749 .lfsck_post = lfsck_namespace_post,
4750 .lfsck_dump = lfsck_namespace_dump,
4751 .lfsck_double_scan = lfsck_namespace_double_scan,
4752 .lfsck_data_release = lfsck_namespace_data_release,
4753 .lfsck_quit = lfsck_namespace_quit,
4754 .lfsck_in_notify = lfsck_namespace_in_notify,
4755 .lfsck_query = lfsck_namespace_query,
4759 * Repair dangling name entry.
4761 * For the name entry with dangling reference, we need to repare the
4762 * inconsistency according to the LFSCK sponsor's requirement:
4764 * 1) Keep the inconsistency there and report the inconsistency case,
4765 * then give the chance to the application to find related issues,
4766 * and the users can make the decision about how to handle it with
4767 * more human knownledge. (by default)
4769 * 2) Re-create the missing MDT-object with the FID information.
4771 * \param[in] env pointer to the thread context
4772 * \param[in] com pointer to the lfsck component
4773 * \param[in] parent pointer to the dir object that contains the dangling
4775 * \param[in] child pointer to the object corresponding to the dangling
4777 * \param[in] lnr pointer to the namespace request that contains the
4778 * name's name, parent object, parent's LMV, and ect.
4780 * \retval positive number if no need to repair
4781 * \retval zero for repaired successfully
4782 * \retval negative error number on failure
4784 int lfsck_namespace_repair_dangling(const struct lu_env *env,
4785 struct lfsck_component *com,
4786 struct dt_object *parent,
4787 struct dt_object *child,
4788 struct lfsck_namespace_req *lnr)
4790 struct lfsck_thread_info *info = lfsck_env_info(env);
4791 struct lu_attr *la = &info->lti_la;
4792 struct dt_allocation_hint *hint = &info->lti_hint;
4793 struct dt_object_format *dof = &info->lti_dof;
4794 struct dt_insert_rec *rec = &info->lti_dt_rec;
4795 struct lmv_mds_md_v1 *lmv2 = &info->lti_lmv2;
4796 const struct lu_name *cname;
4797 const struct lu_fid *pfid = lfsck_dto2fid(parent);
4798 const struct lu_fid *cfid = lfsck_dto2fid(child);
4799 struct linkea_data ldata = { NULL };
4800 struct lfsck_lock_handle *llh = &info->lti_llh;
4801 struct lu_buf linkea_buf;
4802 struct lu_buf lmv_buf;
4803 struct lfsck_instance *lfsck = com->lc_lfsck;
4804 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4805 struct dt_device *dev = lfsck->li_next;
4806 struct thandle *th = NULL;
4808 __u16 type = lnr->lnr_type;
4812 cname = lfsck_name_get_const(env, lnr->lnr_name, lnr->lnr_namelen);
4813 if (bk->lb_param & LPF_CREATE_MDTOBJ)
4818 if (!create || bk->lb_param & LPF_DRYRUN)
4821 /* We may need to create the sub-objects of the @child via LOD,
4822 * so make the modification based on lfsck->li_next. */
4824 parent = lfsck_object_locate(dev, parent);
4826 GOTO(log, rc = PTR_ERR(parent));
4828 if (unlikely(!dt_try_as_dir(env, parent)))
4829 GOTO(log, rc = -ENOTDIR);
4831 child = lfsck_object_locate(dev, child);
4833 GOTO(log, rc = PTR_ERR(child));
4835 rc = linkea_data_new(&ldata, &info->lti_linkea_buf2);
4839 rc = linkea_add_buf(&ldata, cname, pfid);
4843 rc = lfsck_lock(env, lfsck, parent, lnr->lnr_name, llh,
4844 MDS_INODELOCK_UPDATE, LCK_PR);
4848 rc = lfsck_namespace_check_exist(env, parent, child, lnr->lnr_name);
4852 /* Set the ctime as zero, then others can know it is created for
4853 * repairing dangling name entry by LFSCK. And if the LFSCK made
4854 * wrong decision and the real MDT-object has been found later,
4855 * then the LFSCK has chance to fix the incosistency properly. */
4856 memset(la, 0, sizeof(*la));
4857 la->la_mode = (type & S_IFMT) | 0600;
4858 la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
4859 LA_ATIME | LA_MTIME | LA_CTIME;
4861 child->do_ops->do_ah_init(env, hint, parent, child,
4862 la->la_mode & S_IFMT);
4864 memset(dof, 0, sizeof(*dof));
4865 dof->dof_type = dt_mode_to_dft(type);
4866 /* If the target is a regular file, then the LFSCK will only create
4867 * the MDT-object without stripes (dof->dof_reg.striped = 0). related
4868 * OST-objects will be created when write open. */
4870 th = dt_trans_create(env, dev);
4872 GOTO(log, rc = PTR_ERR(th));
4874 /* 1a. create child. */
4875 rc = dt_declare_create(env, child, la, hint, dof, th);
4879 if (S_ISDIR(type)) {
4880 if (unlikely(!dt_try_as_dir(env, child)))
4881 GOTO(stop, rc = -ENOTDIR);
4883 /* 2a. insert dot into child dir */
4884 rec->rec_type = S_IFDIR;
4885 rec->rec_fid = cfid;
4886 rc = dt_declare_insert(env, child,
4887 (const struct dt_rec *)rec,
4888 (const struct dt_key *)dot, th);
4892 /* 3a. insert dotdot into child dir */
4893 rec->rec_fid = pfid;
4894 rc = dt_declare_insert(env, child,
4895 (const struct dt_rec *)rec,
4896 (const struct dt_key *)dotdot, th);
4900 /* 4a. increase child nlink */
4901 rc = dt_declare_ref_add(env, child, th);
4905 /* 5a. generate slave LMV EA. */
4906 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
4909 idx = lfsck_shard_name_to_index(env,
4910 lnr->lnr_name, lnr->lnr_namelen,
4912 if (unlikely(idx < 0))
4913 GOTO(stop, rc = idx);
4915 *lmv2 = lnr->lnr_lmv->ll_lmv;
4916 lmv2->lmv_magic = LMV_MAGIC_STRIPE;
4917 lmv2->lmv_master_mdt_index = idx;
4919 lfsck_lmv_header_cpu_to_le(lmv2, lmv2);
4920 lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
4921 rc = dt_declare_xattr_set(env, child, &lmv_buf,
4922 XATTR_NAME_LMV, 0, th);
4928 /* 6a. insert linkEA for child */
4929 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
4930 ldata.ld_leh->leh_len);
4931 rc = dt_declare_xattr_set(env, child, &linkea_buf,
4932 XATTR_NAME_LINK, 0, th);
4936 rc = dt_trans_start_local(env, dev, th);
4938 GOTO(stop, rc = (rc == -EEXIST ? 1 : rc));
4940 dt_write_lock(env, child, 0);
4941 /* 1b. create child */
4942 rc = dt_create(env, child, la, hint, dof, th);
4944 GOTO(unlock, rc = (rc == -EEXIST ? 1 : rc));
4946 if (S_ISDIR(type)) {
4947 /* 2b. insert dot into child dir */
4948 rec->rec_type = S_IFDIR;
4949 rec->rec_fid = cfid;
4950 rc = dt_insert(env, child, (const struct dt_rec *)rec,
4951 (const struct dt_key *)dot, th, 1);
4955 /* 3b. insert dotdot into child dir */
4956 rec->rec_fid = pfid;
4957 rc = dt_insert(env, child, (const struct dt_rec *)rec,
4958 (const struct dt_key *)dotdot, th, 1);
4962 /* 4b. increase child nlink */
4963 rc = dt_ref_add(env, child, th);
4967 /* 5b. generate slave LMV EA. */
4968 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
4969 rc = dt_xattr_set(env, child, &lmv_buf, XATTR_NAME_LMV,
4976 /* 6b. insert linkEA for child. */
4977 rc = dt_xattr_set(env, child, &linkea_buf,
4978 XATTR_NAME_LINK, 0, th);
4983 dt_write_unlock(env, child);
4986 dt_trans_stop(env, dev, th);
4990 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found dangling "
4991 "reference for: parent "DFID", child "DFID", type %u, "
4992 "name %s. %s: rc = %d\n", lfsck_lfsck2name(lfsck),
4993 PFID(pfid), PFID(cfid), type, cname->ln_name,
4994 create ? "Create the lost MDT-object as required" :
4995 "Keep the MDT-object there by default", rc);
4998 struct lfsck_namespace *ns = com->lc_file_ram;
5000 ns->ln_flags |= LF_INCONSISTENT;
5006 static int lfsck_namespace_assistant_handler_p1(const struct lu_env *env,
5007 struct lfsck_component *com,
5008 struct lfsck_assistant_req *lar)
5010 struct lfsck_thread_info *info = lfsck_env_info(env);
5011 struct lu_attr *la = &info->lti_la;
5012 struct lfsck_instance *lfsck = com->lc_lfsck;
5013 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5014 struct lfsck_namespace *ns = com->lc_file_ram;
5015 struct lfsck_assistant_data *lad = com->lc_data;
5016 struct linkea_data ldata = { NULL };
5017 const struct lu_name *cname;
5018 struct thandle *handle = NULL;
5019 struct lfsck_namespace_req *lnr =
5020 container_of0(lar, struct lfsck_namespace_req, lnr_lar);
5021 struct dt_object *dir = NULL;
5022 struct dt_object *obj = NULL;
5023 struct lfsck_assistant_object *lso = lar->lar_parent;
5024 const struct lu_fid *pfid = &lso->lso_fid;
5025 struct dt_device *dev = NULL;
5026 struct lustre_handle lh = { 0 };
5027 bool repaired = false;
5028 bool dtlocked = false;
5032 bool bad_hash = false;
5033 bool bad_linkea = false;
5037 enum lfsck_namespace_inconsistency_type type = LNIT_NONE;
5044 if (lnr->lnr_attr & LUDA_UPGRADE) {
5045 ns->ln_flags |= LF_UPGRADE;
5046 ns->ln_dirent_repaired++;
5048 } else if (lnr->lnr_attr & LUDA_REPAIR) {
5049 ns->ln_flags |= LF_INCONSISTENT;
5050 ns->ln_dirent_repaired++;
5054 if (unlikely(fid_is_zero(&lnr->lnr_fid))) {
5055 if (strcmp(lnr->lnr_name, dotdot) != 0)
5058 rc = lfsck_namespace_trace_update(env, com, pfid,
5059 LNTF_CHECK_PARENT, true);
5064 if (unlikely(!fid_is_sane(&lnr->lnr_fid))) {
5065 CDEBUG(D_LFSCK, "%s: dir scan find invalid FID "DFID
5066 " for the name entry %.*s under "DFID"\n",
5067 lfsck_lfsck2name(lfsck), PFID(&lnr->lnr_fid),
5068 lnr->lnr_namelen, lnr->lnr_name, PFID(pfid));
5070 if (strcmp(lnr->lnr_name, dotdot) != 0)
5071 /* invalid FID means bad name entry, remove it. */
5072 type = LNIT_BAD_DIRENT;
5074 /* If the parent FID is invalid, we cannot remove
5075 * the ".." entry directly. */
5076 rc = lfsck_namespace_trace_update(env, com, pfid,
5077 LNTF_CHECK_PARENT, true);
5082 if (unlikely(lnr->lnr_dir_cookie == MDS_DIR_END_OFF)) {
5083 rc = lfsck_namespace_striped_dir_rescan(env, com, lnr);
5088 if (fid_seq_is_dot(fid_seq(&lnr->lnr_fid)))
5091 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
5092 rc = lfsck_namespace_handle_striped_master(env, com, lnr);
5097 idx = lfsck_find_mdt_idx_by_fid(env, lfsck, &lnr->lnr_fid);
5099 GOTO(out, rc = idx);
5101 if (idx == lfsck_dev_idx(lfsck)) {
5102 if (unlikely(strcmp(lnr->lnr_name, dotdot) == 0))
5105 dev = lfsck->li_bottom;
5107 struct lfsck_tgt_desc *ltd;
5109 /* Usually, some local filesystem consistency verification
5110 * tools can guarantee the local namespace tree consistenct.
5111 * So the LFSCK will only verify the remote directory. */
5112 if (unlikely(strcmp(lnr->lnr_name, dotdot) == 0)) {
5113 rc = lfsck_namespace_trace_update(env, com, pfid,
5114 LNTF_CHECK_PARENT, true);
5119 ltd = lfsck_ltd2tgt(&lfsck->li_mdt_descs, idx);
5120 if (unlikely(ltd == NULL)) {
5121 CDEBUG(D_LFSCK, "%s: cannot talk with MDT %x which "
5122 "did not join the namespace LFSCK\n",
5123 lfsck_lfsck2name(lfsck), idx);
5124 lfsck_lad_set_bitmap(env, com, idx);
5126 GOTO(out, rc = -ENODEV);
5132 obj = lfsck_object_find_by_dev(env, dev, &lnr->lnr_fid);
5134 GOTO(out, rc = PTR_ERR(obj));
5136 cname = lfsck_name_get_const(env, lnr->lnr_name, lnr->lnr_namelen);
5137 if (dt_object_exists(obj) == 0) {
5141 dir = lfsck_assistant_object_load(env, lfsck, lso);
5145 GOTO(trace, rc == -ENOENT ? 0 : rc);
5149 rc = lfsck_namespace_check_exist(env, dir, obj, lnr->lnr_name);
5151 if (!lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5152 lnr->lnr_name, lnr->lnr_namelen)) {
5153 type = LNIT_BAD_DIRENT;
5158 type = LNIT_DANGLING;
5159 rc = lfsck_namespace_repair_dangling(env, com, dir,
5168 if (!(bk->lb_param & LPF_DRYRUN) && lad->lad_advance_lock) {
5171 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
5172 MDS_INODELOCK_UPDATE |
5173 MDS_INODELOCK_XATTR, LCK_EX);
5177 handle = dt_trans_create(env, dev);
5179 GOTO(out, rc = PTR_ERR(handle));
5181 rc = lfsck_declare_namespace_exec_dir(env, obj, handle);
5185 rc = dt_trans_start_local(env, dev, handle);
5189 dt_write_lock(env, obj, 0);
5193 rc = lfsck_links_read(env, obj, &ldata);
5194 if (unlikely(rc == -ENOENT)) {
5195 if (handle != NULL) {
5196 dt_write_unlock(env, obj);
5199 dt_trans_stop(env, dev, handle);
5202 lfsck_ibits_unlock(&lh, LCK_EX);
5205 /* It may happen when the remote object has been removed,
5206 * but the local MDT is not aware of that. */
5208 } else if (rc == 0) {
5209 count = ldata.ld_leh->leh_reccount;
5210 rc = linkea_links_find(&ldata, cname, pfid);
5212 (count == 1 || !S_ISDIR(lfsck_object_type(obj)))) {
5213 if ((lfsck_object_type(obj) & S_IFMT) !=
5215 ns->ln_flags |= LF_INCONSISTENT;
5216 type = LNIT_BAD_TYPE;
5222 /* If the name entry hash does not match the slave striped
5223 * directory, and the name entry does not match also, then
5224 * it is quite possible that name entry is corrupted. */
5225 if (!lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5226 lnr->lnr_name, lnr->lnr_namelen)) {
5227 ns->ln_flags |= LF_INCONSISTENT;
5228 type = LNIT_BAD_DIRENT;
5233 /* If the file type stored in the name entry does not match
5234 * the file type claimed by the object, and the object does
5235 * not recognize the name entry, then it is quite possible
5236 * that the name entry is corrupted. */
5237 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type) {
5238 ns->ln_flags |= LF_INCONSISTENT;
5239 type = LNIT_BAD_DIRENT;
5244 /* For sub-dir object, we cannot make sure whether the sub-dir
5245 * back references the parent via ".." name entry correctly or
5246 * not in the LFSCK first-stage scanning. It may be that the
5247 * (remote) sub-dir ".." name entry has no parent FID after
5248 * file-level backup/restore and its linkEA may be wrong.
5249 * So under such case, we should replace the linkEA according
5250 * to current name entry. But this needs to be done during the
5251 * LFSCK second-stage scanning. The LFSCK will record the name
5252 * entry for further possible using. */
5256 } else if (unlikely(rc == -EINVAL)) {
5257 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type)
5258 type = LNIT_BAD_TYPE;
5261 /* The magic crashed, we are not sure whether there are more
5262 * corrupt data in the linkea, so remove all linkea entries. */
5266 } else if (rc == -ENODATA) {
5267 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type)
5268 type = LNIT_BAD_TYPE;
5275 if (bk->lb_param & LPF_DRYRUN) {
5277 ns->ln_flags |= LF_UPGRADE;
5279 ns->ln_flags |= LF_INCONSISTENT;
5280 ns->ln_linkea_repaired++;
5286 if (!lustre_handle_is_used(&lh)) {
5295 dir = lfsck_assistant_object_load(env, lfsck, lso);
5299 GOTO(stop, rc == -ENOENT ? 0 : rc);
5303 rc = lfsck_namespace_check_exist(env, dir, obj, lnr->lnr_name);
5308 if (!remove && newdata)
5309 ns->ln_flags |= LF_UPGRADE;
5310 else if (remove || !(ns->ln_flags & LF_UPGRADE))
5311 ns->ln_flags |= LF_INCONSISTENT;
5316 rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, handle);
5322 rc = linkea_data_new(&ldata,
5323 &lfsck_env_info(env)->lti_linkea_buf);
5328 rc = linkea_add_buf(&ldata, cname, pfid);
5332 rc = lfsck_links_write(env, obj, &ldata, handle);
5333 if (unlikely(rc == -ENOSPC) &&
5334 S_ISREG(lfsck_object_type(obj)) && !dt_object_remote(obj)) {
5335 if (handle != NULL) {
5336 LASSERT(dt_write_locked(env, obj));
5338 dt_write_unlock(env, obj);
5341 dt_trans_stop(env, dev, handle);
5344 lfsck_ibits_unlock(&lh, LCK_EX);
5347 rc = lfsck_namespace_trace_update(env, com,
5348 &lnr->lnr_fid, LNTF_SKIP_NLINK, true);
5350 /* If we cannot record this object in the
5351 * LFSCK tracing, we have to mark the LFSCK
5352 * as LF_INCOMPLETE, then the LFSCK will
5353 * skip nlink attribute verification for
5355 ns->ln_flags |= LF_INCOMPLETE;
5363 count = ldata.ld_leh->leh_reccount;
5364 if (!S_ISDIR(lfsck_object_type(obj)) ||
5365 !dt_object_remote(obj)) {
5366 ns->ln_linkea_repaired++;
5376 dt_write_unlock(env, obj);
5378 if (handle != NULL && !IS_ERR(handle))
5379 dt_trans_stop(env, dev, handle);
5382 lfsck_ibits_unlock(&lh, LCK_EX);
5384 if (!name_is_dot_or_dotdot(lnr->lnr_name, lnr->lnr_namelen) &&
5385 !lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5386 lnr->lnr_name, lnr->lnr_namelen) &&
5387 type != LNIT_BAD_DIRENT) {
5388 ns->ln_flags |= LF_INCONSISTENT;
5392 dir = lfsck_assistant_object_load(env, lfsck, lso);
5396 GOTO(trace, rc == -ENOENT ? 0 : rc);
5400 rc = lfsck_namespace_repair_bad_name_hash(env, com, dir,
5401 lnr->lnr_lmv, lnr->lnr_name);
5407 if (type != LNIT_NONE && dir == NULL) {
5408 dir = lfsck_assistant_object_load(env, lfsck, lso);
5412 GOTO(trace, rc == -ENOENT ? 0 : rc);
5419 rc = lfsck_namespace_repair_dirent(env, com, dir,
5420 obj, lnr->lnr_name, lnr->lnr_name,
5421 lnr->lnr_type, true, false);
5425 case LNIT_BAD_DIRENT:
5427 /* XXX: This is a bad dirent, we do not know whether
5428 * the original name entry reference a regular
5429 * file or a directory, then keep the parent's
5430 * nlink count unchanged here. */
5431 rc = lfsck_namespace_repair_dirent(env, com, dir,
5432 obj, lnr->lnr_name, lnr->lnr_name,
5433 lnr->lnr_type, false, false);
5441 if (count == 1 && S_ISREG(lfsck_object_type(obj)))
5442 dt_attr_get(env, obj, la);
5446 down_write(&com->lc_sem);
5448 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail to handle "
5449 "the entry: "DFID", parent "DFID", name %.*s: rc = %d\n",
5450 lfsck_lfsck2name(lfsck), PFID(&lnr->lnr_fid), PFID(pfid),
5451 lnr->lnr_namelen, lnr->lnr_name, rc);
5453 lfsck_namespace_record_failure(env, lfsck, ns);
5454 if ((rc == -ENOTCONN || rc == -ESHUTDOWN || rc == -EREMCHG ||
5455 rc == -ETIMEDOUT || rc == -EHOSTDOWN ||
5456 rc == -EHOSTUNREACH || rc == -EINPROGRESS) &&
5457 dev != NULL && dev != lfsck->li_bottom)
5458 lfsck_lad_set_bitmap(env, com, idx);
5460 if (!(bk->lb_param & LPF_FAILOUT))
5464 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant "
5465 "repaired the entry: "DFID", parent "DFID
5466 ", name %.*s\n", lfsck_lfsck2name(lfsck),
5467 PFID(&lnr->lnr_fid), PFID(pfid),
5468 lnr->lnr_namelen, lnr->lnr_name);
5471 ns->ln_items_repaired++;
5475 ns->ln_dangling_repaired++;
5478 ns->ln_bad_type_repaired++;
5480 case LNIT_BAD_DIRENT:
5481 ns->ln_dirent_repaired++;
5487 if (bk->lb_param & LPF_DRYRUN &&
5488 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent))
5489 lfsck_pos_fill(env, lfsck,
5490 &ns->ln_pos_first_inconsistent,
5495 ns->ln_name_hash_repaired++;
5497 /* Not count repeatedly. */
5499 ns->ln_items_repaired++;
5501 if (bk->lb_param & LPF_DRYRUN &&
5502 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent))
5503 lfsck_pos_fill(env, lfsck,
5504 &ns->ln_pos_first_inconsistent,
5511 if (count > 1 || la->la_nlink > 1)
5512 ns->ln_mul_linked_checked++;
5514 up_write(&com->lc_sem);
5516 if (obj != NULL && !IS_ERR(obj))
5517 lfsck_object_put(env, obj);
5519 if (dir != NULL && !IS_ERR(dir))
5520 lfsck_object_put(env, dir);
5522 lad->lad_advance_lock = bad_linkea;
5528 * Handle one orphan under the backend /lost+found directory
5530 * Insert the orphan FID into the namespace LFSCK trace file for further
5531 * processing (via the subsequent namespace LFSCK second-stage scanning).
5532 * At the same time, remove the orphan name entry from backend /lost+found
5533 * directory. There is an interval between the orphan name entry removed
5534 * from the backend /lost+found directory and the orphan FID in the LFSCK
5535 * trace file handled. In such interval, the LFSCK can be reset, then
5536 * all the FIDs recorded in the namespace LFSCK trace file will be dropped.
5537 * To guarantee that the orphans can be found when LFSCK run next time
5538 * without e2fsck again, when remove the orphan name entry, the LFSCK
5539 * will set the orphan's ctime attribute as 1. Since normal applications
5540 * cannot change the object's ctime attribute as 1. Then when LFSCK run
5541 * next time, it can record the object (that ctime is 1) in the namespace
5542 * LFSCK trace file during the first-stage scanning.
5544 * \param[in] env pointer to the thread context
5545 * \param[in] com pointer to the lfsck component
5546 * \param[in] parent pointer to the object for the backend /lost+found
5547 * \param[in] ent pointer to the name entry for the target under the
5548 * backend /lost+found
5550 * \retval positive for repaired
5551 * \retval 0 if needs to repair nothing
5552 * \retval negative error number on failure
5554 static int lfsck_namespace_scan_local_lpf_one(const struct lu_env *env,
5555 struct lfsck_component *com,
5556 struct dt_object *parent,
5557 struct lu_dirent *ent)
5559 struct lfsck_thread_info *info = lfsck_env_info(env);
5560 struct lu_fid *key = &info->lti_fid;
5561 struct lu_attr *la = &info->lti_la;
5562 struct lfsck_instance *lfsck = com->lc_lfsck;
5563 struct dt_object *obj;
5564 struct dt_device *dev = lfsck->li_bottom;
5565 struct dt_object *child = NULL;
5566 struct thandle *th = NULL;
5573 child = lfsck_object_find_by_dev(env, dev, &ent->lde_fid);
5575 RETURN(PTR_ERR(child));
5577 LASSERT(dt_object_exists(child));
5578 LASSERT(!dt_object_remote(child));
5580 idx = lfsck_sub_trace_file_fid2idx(&ent->lde_fid);
5581 obj = com->lc_sub_trace_objs[idx].lsto_obj;
5582 fid_cpu_to_be(key, &ent->lde_fid);
5583 rc = dt_lookup(env, obj, (struct dt_rec *)&flags,
5584 (const struct dt_key *)key);
5587 flags |= LNTF_CHECK_ORPHAN;
5588 } else if (rc == -ENOENT) {
5589 flags = LNTF_CHECK_ORPHAN;
5594 th = dt_trans_create(env, dev);
5596 GOTO(out, rc = PTR_ERR(th));
5598 /* a1. remove name entry from backend /lost+found */
5599 rc = dt_declare_delete(env, parent,
5600 (const struct dt_key *)ent->lde_name, th);
5604 if (S_ISDIR(lfsck_object_type(child))) {
5605 /* a2. decrease parent's nlink */
5606 rc = dt_declare_ref_del(env, parent, th);
5612 /* a3. remove child's FID from the LFSCK trace file. */
5613 rc = dt_declare_delete(env, obj,
5614 (const struct dt_key *)key, th);
5618 /* a4. set child's ctime as 1 */
5619 memset(la, 0, sizeof(*la));
5621 la->la_valid = LA_CTIME;
5622 rc = dt_declare_attr_set(env, child, la, th);
5627 /* a5. insert child's FID into the LFSCK trace file. */
5628 rc = dt_declare_insert(env, obj, (const struct dt_rec *)&flags,
5629 (const struct dt_key *)key, th);
5633 rc = dt_trans_start_local(env, dev, th);
5637 /* b1. remove name entry from backend /lost+found */
5638 rc = dt_delete(env, parent, (const struct dt_key *)ent->lde_name, th);
5642 if (S_ISDIR(lfsck_object_type(child))) {
5643 /* b2. decrease parent's nlink */
5644 dt_write_lock(env, parent, 0);
5645 rc = dt_ref_del(env, parent, th);
5646 dt_write_unlock(env, parent);
5652 /* a3. remove child's FID from the LFSCK trace file. */
5653 rc = dt_delete(env, obj, (const struct dt_key *)key, th);
5657 /* b4. set child's ctime as 1 */
5658 rc = dt_attr_set(env, child, la, th);
5663 /* b5. insert child's FID into the LFSCK trace file. */
5664 rc = dt_insert(env, obj, (const struct dt_rec *)&flags,
5665 (const struct dt_key *)key, th, 1);
5667 GOTO(stop, rc = (rc == 0 ? 1 : rc));
5670 dt_trans_stop(env, dev, th);
5673 lfsck_object_put(env, child);
5679 * Handle orphans under the backend /lost+found directory
5681 * Some backend checker, such as e2fsck for ldiskfs may find some orphans
5682 * and put them under the backend /lost+found directory that is invisible
5683 * to client. The LFSCK will scan such directory, for the original client
5684 * visible orphans, add their fids into the namespace LFSCK trace file,
5685 * then the subsenquent namespace LFSCK second-stage scanning can handle
5686 * them as other objects to be double scanned: either move back to normal
5687 * namespace, or to the global visible orphan directory:
5688 * /ROOT/.lustre/lost+found/MDTxxxx/
5690 * \param[in] env pointer to the thread context
5691 * \param[in] com pointer to the lfsck component
5693 static void lfsck_namespace_scan_local_lpf(const struct lu_env *env,
5694 struct lfsck_component *com)
5696 struct lfsck_thread_info *info = lfsck_env_info(env);
5697 struct lu_dirent *ent =
5698 (struct lu_dirent *)info->lti_key;
5699 struct lu_seq_range *range = &info->lti_range;
5700 struct lfsck_instance *lfsck = com->lc_lfsck;
5701 struct ptlrpc_thread *thread = &lfsck->li_thread;
5702 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5703 struct lfsck_namespace *ns = com->lc_file_ram;
5704 struct dt_object *parent;
5705 const struct dt_it_ops *iops;
5707 struct seq_server_site *ss = lfsck_dev_site(lfsck);
5709 __u32 idx = lfsck_dev_idx(lfsck);
5714 parent = lfsck_object_find_by_dev(env, lfsck->li_bottom,
5715 &LU_BACKEND_LPF_FID);
5716 if (IS_ERR(parent)) {
5717 CERROR("%s: fail to find backend /lost+found: rc = %ld\n",
5718 lfsck_lfsck2name(lfsck), PTR_ERR(parent));
5722 /* It is normal that the /lost+found does not exist for ZFS backend. */
5723 if (!dt_object_exists(parent))
5726 if (unlikely(!dt_try_as_dir(env, parent)))
5727 GOTO(out, rc = -ENOTDIR);
5729 CDEBUG(D_LFSCK, "%s: start to scan backend /lost+found\n",
5730 lfsck_lfsck2name(lfsck));
5732 com->lc_new_scanned = 0;
5733 iops = &parent->do_index_ops->dio_it;
5734 di = iops->init(env, parent, LUDA_64BITHASH | LUDA_TYPE);
5736 GOTO(out, rc = PTR_ERR(di));
5738 rc = iops->load(env, di, 0);
5740 rc = iops->next(env, di);
5745 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
5746 unlikely(!thread_is_running(thread)))
5749 rc = iops->rec(env, di, (struct dt_rec *)ent,
5750 LUDA_64BITHASH | LUDA_TYPE);
5752 rc = lfsck_unpack_ent(ent, &cookie, &type);
5754 if (unlikely(rc != 0)) {
5755 CDEBUG(D_LFSCK, "%s: fail to iterate backend "
5756 "/lost+found: rc = %d\n",
5757 lfsck_lfsck2name(lfsck), rc);
5762 /* skip dot and dotdot entries */
5763 if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen))
5766 if (!fid_seq_in_fldb(fid_seq(&ent->lde_fid)))
5769 if (fid_is_norm(&ent->lde_fid)) {
5770 fld_range_set_mdt(range);
5771 rc = fld_local_lookup(env, ss->ss_server_fld,
5772 fid_seq(&ent->lde_fid), range);
5775 } else if (idx != 0) {
5776 /* If the returned FID is IGIF, then there are three
5779 * 1) The object is upgraded from old Lustre-1.8 with
5780 * IGIF assigned to such object.
5781 * 2) The object is a backend local object and is
5782 * invisible to client.
5783 * 3) The object lost its LMV EA, and since there is
5784 * no FID-in-dirent for the orphan in the backend
5785 * /lost+found directory, then the low layer will
5786 * return IGIF for such object.
5788 * For MDTx (x != 0), it is either case 2) or case 3),
5789 * but from the LFSCK view, they are indistinguishable.
5790 * To be safe, the LFSCK will keep it there and report
5791 * some message, then the adminstrator can handle that
5794 * For MDT0, it is more possible the case 1). The LFSCK
5795 * will handle the orphan as an upgraded object. */
5796 CDEBUG(D_LFSCK, "%s: the orphan %.*s with IGIF "DFID
5797 "in the backend /lost+found on the MDT %04x, "
5798 "to be safe, skip it.\n",
5799 lfsck_lfsck2name(lfsck), ent->lde_namelen,
5800 ent->lde_name, PFID(&ent->lde_fid), idx);
5804 rc = lfsck_namespace_scan_local_lpf_one(env, com, parent, ent);
5807 down_write(&com->lc_sem);
5808 com->lc_new_scanned++;
5809 ns->ln_local_lpf_scanned++;
5811 ns->ln_local_lpf_moved++;
5813 ns->ln_local_lpf_skipped++;
5815 ns->ln_local_lpf_failed++;
5816 up_write(&com->lc_sem);
5818 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
5822 lfsck_control_speed_by_self(com);
5823 if (unlikely(!thread_is_running(thread))) {
5828 rc = iops->next(env, di);
5832 iops->fini(env, di);
5837 CDEBUG(D_LFSCK, "%s: stop to scan backend /lost+found: rc = %d\n",
5838 lfsck_lfsck2name(lfsck), rc);
5840 lfsck_object_put(env, parent);
5844 * Rescan the striped directory after the master LMV EA reset.
5846 * Sometimes, the master LMV EA of the striped directory maybe lost, so when
5847 * the namespace LFSCK engine scan the striped directory for the first time,
5848 * it will be regarded as a normal directory. As the LFSCK processing, some
5849 * other LFSCK instance on other MDT will find the shard of this striped dir,
5850 * and find that the master MDT-object of the striped directory lost its LMV
5851 * EA, then such remote LFSCK instance will regenerate the master LMV EA and
5852 * notify the LFSCK instance on this MDT to rescan the striped directory.
5854 * \param[in] env pointer to the thread context
5855 * \param[in] com pointer to the lfsck component
5856 * \param[in] llu the lfsck_lmv_unit that contains the striped directory
5859 * \retval positive number for success
5860 * \retval 0 for LFSCK stopped/paused
5861 * \retval negative error number on failure
5863 static int lfsck_namespace_rescan_striped_dir(const struct lu_env *env,
5864 struct lfsck_component *com,
5865 struct lfsck_lmv_unit *llu)
5867 struct lfsck_thread_info *info = lfsck_env_info(env);
5868 struct lfsck_instance *lfsck = com->lc_lfsck;
5869 struct lfsck_assistant_data *lad = com->lc_data;
5870 struct dt_object *dir;
5871 const struct dt_it_ops *iops;
5873 struct lu_dirent *ent =
5874 (struct lu_dirent *)info->lti_key;
5875 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5876 struct ptlrpc_thread *thread = &lfsck->li_thread;
5877 struct lfsck_assistant_object *lso = NULL;
5878 struct lfsck_namespace_req *lnr;
5879 struct lfsck_assistant_req *lar;
5884 LASSERT(list_empty(&lad->lad_req_list));
5886 lfsck->li_lmv = &llu->llu_lmv;
5887 lfsck->li_obj_dir = lfsck_object_get(llu->llu_obj);
5888 rc = lfsck_open_dir(env, lfsck, 0);
5892 dir = lfsck->li_obj_dir;
5893 di = lfsck->li_di_dir;
5894 iops = &dir->do_index_ops->dio_it;
5896 rc = iops->rec(env, di, (struct dt_rec *)ent,
5897 lfsck->li_args_dir);
5899 rc = lfsck_unpack_ent(ent, &lfsck->li_cookie_dir,
5903 if (bk->lb_param & LPF_FAILOUT)
5909 if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen))
5913 lso = lfsck_assistant_object_init(env,
5914 lfsck_dto2fid(dir), NULL,
5915 lfsck->li_pos_current.lp_oit_cookie, true);
5917 if (bk->lb_param & LPF_FAILOUT)
5918 GOTO(out, rc = PTR_ERR(lso));
5925 lnr = lfsck_namespace_assistant_req_init(lfsck, lso, ent, type);
5927 if (bk->lb_param & LPF_FAILOUT)
5928 GOTO(out, rc = PTR_ERR(lnr));
5933 lar = &lnr->lnr_lar;
5934 rc = lfsck_namespace_assistant_handler_p1(env, com, lar);
5935 lfsck_namespace_assistant_req_fini(env, lar);
5936 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
5939 if (unlikely(!thread_is_running(thread)))
5943 rc = iops->next(env, di);
5947 if (lso != NULL && !IS_ERR(lso))
5948 lfsck_assistant_object_put(env, lso);
5950 lfsck_close_dir(env, lfsck, rc);
5954 /* The close_dir() may insert a dummy lnr in the lad->lad_req_list. */
5955 if (list_empty(&lad->lad_req_list))
5958 spin_lock(&lad->lad_lock);
5959 lar = list_entry(lad->lad_req_list.next, struct lfsck_assistant_req,
5961 list_del_init(&lar->lar_list);
5962 spin_unlock(&lad->lad_lock);
5964 rc = lfsck_namespace_assistant_handler_p1(env, com, lar);
5965 lfsck_namespace_assistant_req_fini(env, lar);
5967 RETURN(rc == 0 ? 1 : rc);
5971 lfsck_namespace_double_scan_one_trace_file(const struct lu_env *env,
5972 struct lfsck_component *com,
5973 struct dt_object *obj, bool first)
5975 struct lfsck_instance *lfsck = com->lc_lfsck;
5976 struct ptlrpc_thread *thread = &lfsck->li_thread;
5977 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5978 struct lfsck_namespace *ns = com->lc_file_ram;
5979 const struct dt_it_ops *iops = &obj->do_index_ops->dio_it;
5980 struct dt_object *target;
5988 di = iops->init(env, obj, 0);
5990 RETURN(PTR_ERR(di));
5993 fid_cpu_to_be(&fid, &ns->ln_fid_latest_scanned_phase2);
5996 rc = iops->get(env, di, (const struct dt_key *)&fid);
6001 /* The start one either has been processed or does not exist,
6003 rc = iops->next(env, di);
6009 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
6010 unlikely(!thread_is_running(thread)))
6013 key = iops->key(env, di);
6022 fid_be_to_cpu(&fid, (const struct lu_fid *)key);
6023 if (!fid_is_sane(&fid)) {
6028 target = lfsck_object_find_bottom(env, lfsck, &fid);
6029 if (IS_ERR(target)) {
6030 rc = PTR_ERR(target);
6034 if (dt_object_exists(target)) {
6035 rc = iops->rec(env, di, (struct dt_rec *)&flags, 0);
6037 rc = lfsck_namespace_double_scan_one(env, com,
6044 lfsck_object_put(env, target);
6047 down_write(&com->lc_sem);
6048 com->lc_new_checked++;
6049 com->lc_new_scanned++;
6050 if (rc >= 0 && fid_is_sane(&fid))
6051 ns->ln_fid_latest_scanned_phase2 = fid;
6053 ns->ln_objs_repaired_phase2++;
6055 ns->ln_objs_failed_phase2++;
6056 up_write(&com->lc_sem);
6058 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
6061 if (unlikely(cfs_time_beforeq(com->lc_time_next_checkpoint,
6062 cfs_time_current())) &&
6063 com->lc_new_checked != 0) {
6064 down_write(&com->lc_sem);
6065 ns->ln_run_time_phase2 +=
6066 cfs_duration_sec(cfs_time_current() +
6067 HALF_SEC - com->lc_time_last_checkpoint);
6068 ns->ln_time_last_checkpoint = cfs_time_current_sec();
6069 ns->ln_objs_checked_phase2 += com->lc_new_checked;
6070 com->lc_new_checked = 0;
6071 rc = lfsck_namespace_store(env, com, false);
6072 up_write(&com->lc_sem);
6076 com->lc_time_last_checkpoint = cfs_time_current();
6077 com->lc_time_next_checkpoint =
6078 com->lc_time_last_checkpoint +
6079 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
6082 lfsck_control_speed_by_self(com);
6083 if (unlikely(!thread_is_running(thread)))
6086 rc = iops->next(env, di);
6095 iops->fini(env, di);
6100 static int lfsck_namespace_assistant_handler_p2(const struct lu_env *env,
6101 struct lfsck_component *com)
6103 struct lfsck_instance *lfsck = com->lc_lfsck;
6104 struct lfsck_namespace *ns = com->lc_file_ram;
6109 while (!list_empty(&lfsck->li_list_lmv)) {
6110 struct lfsck_lmv_unit *llu;
6112 spin_lock(&lfsck->li_lock);
6113 llu = list_entry(lfsck->li_list_lmv.next,
6114 struct lfsck_lmv_unit, llu_link);
6115 list_del_init(&llu->llu_link);
6116 spin_unlock(&lfsck->li_lock);
6118 rc = lfsck_namespace_rescan_striped_dir(env, com, llu);
6123 CDEBUG(D_LFSCK, "%s: namespace LFSCK phase2 scan start\n",
6124 lfsck_lfsck2name(lfsck));
6126 lfsck_namespace_scan_local_lpf(env, com);
6128 com->lc_new_checked = 0;
6129 com->lc_new_scanned = 0;
6130 com->lc_time_last_checkpoint = cfs_time_current();
6131 com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
6132 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
6134 i = lfsck_sub_trace_file_fid2idx(&ns->ln_fid_latest_scanned_phase2);
6135 rc = lfsck_namespace_double_scan_one_trace_file(env, com,
6136 com->lc_sub_trace_objs[i].lsto_obj, true);
6137 while (rc > 0 && ++i < LFSCK_STF_COUNT)
6138 rc = lfsck_namespace_double_scan_one_trace_file(env, com,
6139 com->lc_sub_trace_objs[i].lsto_obj, false);
6141 CDEBUG(D_LFSCK, "%s: namespace LFSCK phase2 scan stop at the No. %d "
6142 "trace file: rc = %d\n", lfsck_lfsck2name(lfsck), i, rc);
6147 static void lfsck_namespace_assistant_fill_pos(const struct lu_env *env,
6148 struct lfsck_component *com,
6149 struct lfsck_position *pos)
6151 struct lfsck_assistant_data *lad = com->lc_data;
6152 struct lfsck_namespace_req *lnr;
6154 if (list_empty(&lad->lad_req_list))
6157 lnr = list_entry(lad->lad_req_list.next,
6158 struct lfsck_namespace_req,
6160 pos->lp_oit_cookie = lnr->lnr_lar.lar_parent->lso_oit_cookie;
6161 pos->lp_dir_cookie = lnr->lnr_dir_cookie - 1;
6162 pos->lp_dir_parent = lnr->lnr_lar.lar_parent->lso_fid;
6165 static int lfsck_namespace_double_scan_result(const struct lu_env *env,
6166 struct lfsck_component *com,
6169 struct lfsck_instance *lfsck = com->lc_lfsck;
6170 struct lfsck_namespace *ns = com->lc_file_ram;
6172 down_write(&com->lc_sem);
6173 ns->ln_run_time_phase2 += cfs_duration_sec(cfs_time_current() +
6174 HALF_SEC - com->lc_time_last_checkpoint);
6175 ns->ln_time_last_checkpoint = cfs_time_current_sec();
6176 ns->ln_objs_checked_phase2 += com->lc_new_checked;
6177 com->lc_new_checked = 0;
6180 if (ns->ln_flags & LF_INCOMPLETE)
6181 ns->ln_status = LS_PARTIAL;
6183 ns->ln_status = LS_COMPLETED;
6184 if (!(lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN))
6185 ns->ln_flags &= ~(LF_SCANNED_ONCE | LF_INCONSISTENT);
6186 ns->ln_time_last_complete = ns->ln_time_last_checkpoint;
6187 ns->ln_success_count++;
6188 } else if (rc == 0) {
6189 if (lfsck->li_status != 0)
6190 ns->ln_status = lfsck->li_status;
6192 ns->ln_status = LS_STOPPED;
6194 ns->ln_status = LS_FAILED;
6197 rc = lfsck_namespace_store(env, com, false);
6198 up_write(&com->lc_sem);
6204 lfsck_namespace_assistant_sync_failures_interpret(const struct lu_env *env,
6205 struct ptlrpc_request *req,
6209 struct lfsck_async_interpret_args *laia = args;
6210 struct lfsck_tgt_desc *ltd = laia->laia_ltd;
6212 ltd->ltd_synced_failures = 1;
6219 * Notify remote LFSCK instances about former failures.
6221 * The local LFSCK instance has recorded which MDTs have ever failed to respond
6222 * some LFSCK verification requests (maybe because of network issues or the MDT
6223 * itself trouble). During the respond gap the MDT may missed some name entries
6224 * verification, then the MDT cannot know whether related MDT-objects have been
6225 * referenced by related name entries or not, then in the second-stage scanning,
6226 * these MDT-objects will be regarded as orphan, if the MDT-object contains bad
6227 * linkEA for back reference, then it will misguide the LFSCK to generate wrong
6228 * name entry for repairing the orphan.
6230 * To avoid above trouble, when layout LFSCK finishes the first-stage scanning,
6231 * it will scan the bitmap for the ever failed MDTs, and notify them that they
6232 * have ever missed some name entries verification and should skip the handling
6233 * for orphan MDT-objects.
6235 * \param[in] env pointer to the thread context
6236 * \param[in] com pointer to the lfsck component
6237 * \param[in] lr pointer to the lfsck request
6239 static void lfsck_namespace_assistant_sync_failures(const struct lu_env *env,
6240 struct lfsck_component *com,
6241 struct lfsck_request *lr)
6243 struct lfsck_async_interpret_args *laia =
6244 &lfsck_env_info(env)->lti_laia2;
6245 struct lfsck_assistant_data *lad = com->lc_data;
6246 struct lfsck_namespace *ns = com->lc_file_ram;
6247 struct lfsck_instance *lfsck = com->lc_lfsck;
6248 struct lfsck_tgt_descs *ltds = &lfsck->li_mdt_descs;
6249 struct lfsck_tgt_desc *ltd;
6250 struct ptlrpc_request_set *set;
6255 if (!lad->lad_incomplete)
6258 set = ptlrpc_prep_set();
6260 GOTO(out, rc = -ENOMEM);
6262 lr->lr_flags2 = ns->ln_flags | LF_INCOMPLETE;
6263 memset(laia, 0, sizeof(*laia));
6264 lad->lad_touch_gen++;
6266 down_read(<ds->ltd_rw_sem);
6267 cfs_foreach_bit(lad->lad_bitmap, idx) {
6268 ltd = lfsck_ltd2tgt(ltds, idx);
6269 LASSERT(ltd != NULL);
6271 laia->laia_ltd = ltd;
6272 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
6273 lfsck_namespace_assistant_sync_failures_interpret,
6274 laia, LFSCK_NOTIFY);
6276 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail "
6277 "to sync failure with MDT %x: rc = %d\n",
6278 lfsck_lfsck2name(lfsck), ltd->ltd_index, rc);
6280 up_read(<ds->ltd_rw_sem);
6282 rc = ptlrpc_set_wait(set);
6283 ptlrpc_set_destroy(set);
6289 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail "
6290 "to sync failure with MDTs, and related MDTs "
6291 "may handle orphan improperly: rc = %d\n",
6292 lfsck_lfsck2name(lfsck), rc);
6297 struct lfsck_assistant_operations lfsck_namespace_assistant_ops = {
6298 .la_handler_p1 = lfsck_namespace_assistant_handler_p1,
6299 .la_handler_p2 = lfsck_namespace_assistant_handler_p2,
6300 .la_fill_pos = lfsck_namespace_assistant_fill_pos,
6301 .la_double_scan_result = lfsck_namespace_double_scan_result,
6302 .la_req_fini = lfsck_namespace_assistant_req_fini,
6303 .la_sync_failures = lfsck_namespace_assistant_sync_failures,
6307 * Verify the specified linkEA entry for the given directory object.
6308 * If the object has no such linkEA entry or it has more other linkEA
6309 * entries, then re-generate the linkEA with the given information.
6311 * \param[in] env pointer to the thread context
6312 * \param[in] obj pointer to the dt_object to be handled
6313 * \param[in] cname the name for the child in the parent directory
6314 * \param[in] pfid the parent directory's FID for the linkEA
6316 * \retval 0 for success
6317 * \retval negative error number on failure
6319 int lfsck_verify_linkea(const struct lu_env *env, struct dt_object *obj,
6320 const struct lu_name *cname, const struct lu_fid *pfid)
6322 struct dt_device *dev = lfsck_obj2dev(obj);
6323 struct linkea_data ldata = { NULL };
6324 struct lu_buf linkea_buf;
6327 int fl = LU_XATTR_CREATE;
6331 LASSERT(S_ISDIR(lfsck_object_type(obj)));
6333 rc = lfsck_links_read(env, obj, &ldata);
6334 if (rc == -ENODATA) {
6336 } else if (rc == 0) {
6337 fl = LU_XATTR_REPLACE;
6338 if (ldata.ld_leh->leh_reccount != 1) {
6341 rc = linkea_links_find(&ldata, cname, pfid);
6350 rc = linkea_data_new(&ldata, &lfsck_env_info(env)->lti_linkea_buf);
6354 rc = linkea_add_buf(&ldata, cname, pfid);
6358 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
6359 ldata.ld_leh->leh_len);
6360 th = dt_trans_create(env, dev);
6362 RETURN(PTR_ERR(th));
6364 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
6365 XATTR_NAME_LINK, fl, th);
6369 rc = dt_trans_start_local(env, dev, th);
6373 dt_write_lock(env, obj, 0);
6374 rc = dt_xattr_set(env, obj, &linkea_buf,
6375 XATTR_NAME_LINK, fl, th);
6376 dt_write_unlock(env, obj);
6381 dt_trans_stop(env, dev, th);
6386 * Get the name and parent directory's FID from the first linkEA entry.
6388 * \param[in] env pointer to the thread context
6389 * \param[in] obj pointer to the object which get linkEA from
6390 * \param[out] name pointer to the buffer to hold the name
6391 * in the first linkEA entry
6392 * \param[out] pfid pointer to the buffer to hold the parent
6393 * directory's FID in the first linkEA entry
6395 * \retval 0 for success
6396 * \retval negative error number on failure
6398 int lfsck_links_get_first(const struct lu_env *env, struct dt_object *obj,
6399 char *name, struct lu_fid *pfid)
6401 struct lu_name *cname = &lfsck_env_info(env)->lti_name;
6402 struct linkea_data ldata = { NULL };
6405 rc = lfsck_links_read(env, obj, &ldata);
6409 linkea_first_entry(&ldata);
6410 if (ldata.ld_lee == NULL)
6413 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, cname, pfid);
6414 /* To guarantee the 'name' is terminated with '0'. */
6415 memcpy(name, cname->ln_name, cname->ln_namelen);
6416 name[cname->ln_namelen] = 0;
6422 * Update the object's name entry with the given FID.
6424 * \param[in] env pointer to the thread context
6425 * \param[in] lfsck pointer to the lfsck instance
6426 * \param[in] dir pointer to the directory that holds
6428 * \param[in] name the name for the entry to be updated
6429 * \param[in] fid the new FID for the name entry referenced
6430 * \param[in] type the type for the name entry to be updated
6432 * \retval 0 for success
6433 * \retval negative error number on failure
6435 int lfsck_update_name_entry(const struct lu_env *env,
6436 struct lfsck_instance *lfsck,
6437 struct dt_object *dir, const char *name,
6438 const struct lu_fid *fid, __u32 type)
6440 struct lfsck_thread_info *info = lfsck_env_info(env);
6441 struct dt_insert_rec *rec = &info->lti_dt_rec;
6442 struct lfsck_lock_handle *llh = &info->lti_llh;
6443 struct dt_device *dev = lfsck_obj2dev(dir);
6449 rc = lfsck_lock(env, lfsck, dir, name, llh,
6450 MDS_INODELOCK_UPDATE, LCK_PW);
6454 th = dt_trans_create(env, dev);
6456 GOTO(unlock, rc = PTR_ERR(th));
6458 rc = dt_declare_delete(env, dir, (const struct dt_key *)name, th);
6462 rec->rec_type = type;
6464 rc = dt_declare_insert(env, dir, (const struct dt_rec *)rec,
6465 (const struct dt_key *)name, th);
6469 rc = dt_declare_ref_add(env, dir, th);
6473 rc = dt_trans_start_local(env, dev, th);
6477 rc = dt_delete(env, dir, (const struct dt_key *)name, th);
6478 if (rc == -ENOENT) {
6486 rc = dt_insert(env, dir, (const struct dt_rec *)rec,
6487 (const struct dt_key *)name, th, 1);
6488 if (rc == 0 && S_ISDIR(type) && !exists) {
6489 dt_write_lock(env, dir, 0);
6490 rc = dt_ref_add(env, dir, th);
6491 dt_write_unlock(env, dir);
6497 dt_trans_stop(env, dev, th);
6501 CDEBUG(D_LFSCK, "%s: update name entry "DFID"/%s with the FID "DFID
6502 " and the type %o: rc = %d\n", lfsck_lfsck2name(lfsck),
6503 PFID(lfsck_dto2fid(dir)), name, PFID(fid), type, rc);
6508 int lfsck_namespace_setup(const struct lu_env *env,
6509 struct lfsck_instance *lfsck)
6511 struct lfsck_component *com;
6512 struct lfsck_namespace *ns;
6513 struct dt_object *root = NULL;
6514 struct dt_object *obj;
6519 LASSERT(lfsck->li_master);
6525 INIT_LIST_HEAD(&com->lc_link);
6526 INIT_LIST_HEAD(&com->lc_link_dir);
6527 init_rwsem(&com->lc_sem);
6528 atomic_set(&com->lc_ref, 1);
6529 com->lc_lfsck = lfsck;
6530 com->lc_type = LFSCK_TYPE_NAMESPACE;
6531 com->lc_ops = &lfsck_namespace_ops;
6532 com->lc_data = lfsck_assistant_data_init(
6533 &lfsck_namespace_assistant_ops,
6535 if (com->lc_data == NULL)
6536 GOTO(out, rc = -ENOMEM);
6538 com->lc_file_size = sizeof(struct lfsck_namespace);
6539 OBD_ALLOC(com->lc_file_ram, com->lc_file_size);
6540 if (com->lc_file_ram == NULL)
6541 GOTO(out, rc = -ENOMEM);
6543 OBD_ALLOC(com->lc_file_disk, com->lc_file_size);
6544 if (com->lc_file_disk == NULL)
6545 GOTO(out, rc = -ENOMEM);
6547 for (i = 0; i < LFSCK_STF_COUNT; i++)
6548 mutex_init(&com->lc_sub_trace_objs[i].lsto_mutex);
6550 root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
6552 GOTO(out, rc = PTR_ERR(root));
6554 if (unlikely(!dt_try_as_dir(env, root)))
6555 GOTO(out, rc = -ENOTDIR);
6557 obj = local_index_find_or_create(env, lfsck->li_los, root,
6559 S_IFREG | S_IRUGO | S_IWUSR,
6560 &dt_lfsck_features);
6562 GOTO(out, rc = PTR_ERR(obj));
6565 rc = lfsck_namespace_load(env, com);
6567 rc = lfsck_namespace_init(env, com);
6569 rc = lfsck_namespace_reset(env, com, true);
6571 rc = lfsck_namespace_load_sub_trace_files(env, com, false);
6575 ns = com->lc_file_ram;
6576 switch (ns->ln_status) {
6581 spin_lock(&lfsck->li_lock);
6582 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
6583 spin_unlock(&lfsck->li_lock);
6586 CERROR("%s: unknown lfsck_namespace status %d\n",
6587 lfsck_lfsck2name(lfsck), ns->ln_status);
6589 case LS_SCANNING_PHASE1:
6590 case LS_SCANNING_PHASE2:
6591 /* No need to store the status to disk right now.
6592 * If the system crashed before the status stored,
6593 * it will be loaded back when next time. */
6594 ns->ln_status = LS_CRASHED;
6598 spin_lock(&lfsck->li_lock);
6599 list_add_tail(&com->lc_link, &lfsck->li_list_scan);
6600 list_add_tail(&com->lc_link_dir, &lfsck->li_list_dir);
6601 spin_unlock(&lfsck->li_lock);
6608 if (root != NULL && !IS_ERR(root))
6609 lfsck_object_put(env, root);
6611 lfsck_component_cleanup(env, com);
6612 CERROR("%s: fail to init namespace LFSCK component: rc = %d\n",
6613 lfsck_lfsck2name(lfsck), rc);