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 rc = dt_declare_ref_add(env, orphan, th);
1402 rec->rec_type = S_IFDIR;
1403 rec->rec_fid = cfid;
1404 rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
1405 (const struct dt_key *)dot, th);
1409 rec->rec_fid = lfsck_dto2fid(parent);
1410 rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
1411 (const struct dt_key *)dotdot, th);
1416 lmv->lmv_magic = LMV_MAGIC;
1417 lmv->lmv_master_mdt_index = lfsck_dev_idx(lfsck);
1418 lfsck_lmv_header_cpu_to_le(lmv2, lmv);
1419 lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
1420 rc = dt_declare_xattr_set(env, orphan, &lmv_buf,
1421 XATTR_NAME_LMV, 0, th);
1426 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
1427 ldata.ld_leh->leh_len);
1428 rc = dt_declare_xattr_set(env, orphan, &linkea_buf,
1429 XATTR_NAME_LINK, 0, th);
1433 rec->rec_fid = cfid;
1434 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1435 (const struct dt_key *)name, th);
1437 rc = dt_declare_ref_add(env, parent, th);
1442 rc = dt_trans_start_local(env, dev, th);
1446 dt_write_lock(env, orphan, 0);
1447 rc = dt_create(env, orphan, la, hint, dof, th);
1451 rc = dt_ref_add(env, orphan, th);
1455 rec->rec_fid = cfid;
1456 rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1457 (const struct dt_key *)dot, th, 1);
1461 rec->rec_fid = lfsck_dto2fid(parent);
1462 rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1463 (const struct dt_key *)dotdot, th, 1);
1468 rc = dt_xattr_set(env, orphan, &lmv_buf, XATTR_NAME_LMV, 0, th);
1473 rc = dt_xattr_set(env, orphan, &linkea_buf,
1474 XATTR_NAME_LINK, 0, th);
1475 dt_write_unlock(env, orphan);
1479 rec->rec_fid = cfid;
1480 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1481 (const struct dt_key *)name, th, 1);
1483 dt_write_lock(env, parent, 0);
1484 rc = dt_ref_add(env, parent, th);
1485 dt_write_unlock(env, parent);
1488 GOTO(stop, rc = (rc == 0 ? 1 : rc));
1491 dt_write_unlock(env, orphan);
1494 rc1 = dt_trans_stop(env, dev, th);
1495 if (rc1 != 0 && rc > 0)
1502 CDEBUG(D_LFSCK, "%s: namespace LFSCK create orphan dir for "
1503 "the object "DFID", name = %s: rc = %d\n",
1504 lfsck_lfsck2name(lfsck), PFID(cfid),
1505 cname->ln_name != NULL ? cname->ln_name : "<NULL>", rc);
1507 if (parent != NULL && !IS_ERR(parent) && parent != lfsck->li_lpf_obj)
1508 lfsck_object_put(env, parent);
1511 ns->ln_flags |= LF_INCONSISTENT;
1517 * Remove the specified entry from the linkEA.
1519 * Locate the linkEA entry with the given @cname and @pfid, then
1520 * remove this entry or the other entries those are repeated with
1523 * \param[in] env pointer to the thread context
1524 * \param[in] com pointer to the lfsck component
1525 * \param[in] obj pointer to the dt_object to be handled
1526 * \param[in,out]ldata pointer to the buffer that holds the linkEA
1527 * \param[in] cname the name for the child in the parent directory
1528 * \param[in] pfid the parent directory's FID for the linkEA
1529 * \param[in] next if true, then remove the first found linkEA
1530 * entry, and move the ldata->ld_lee to next entry
1532 * \retval positive number for repaired cases
1533 * \retval 0 if nothing to be repaired
1534 * \retval negative error number on failure
1536 static int lfsck_namespace_shrink_linkea(const struct lu_env *env,
1537 struct lfsck_component *com,
1538 struct dt_object *obj,
1539 struct linkea_data *ldata,
1540 struct lu_name *cname,
1541 struct lu_fid *pfid,
1544 struct lfsck_instance *lfsck = com->lc_lfsck;
1545 struct dt_device *dev = lfsck_obj2dev(obj);
1546 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
1547 struct thandle *th = NULL;
1548 struct lustre_handle lh = { 0 };
1549 struct linkea_data ldata_new = { NULL };
1550 struct lu_buf linkea_buf;
1554 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
1555 MDS_INODELOCK_UPDATE | MDS_INODELOCK_XATTR,
1561 linkea_del_buf(ldata, cname);
1563 lfsck_namespace_filter_linkea_entry(ldata, cname, pfid,
1565 lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
1566 ldata->ld_leh->leh_len);
1569 th = dt_trans_create(env, dev);
1571 GOTO(unlock1, rc = PTR_ERR(th));
1573 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
1574 XATTR_NAME_LINK, 0, th);
1578 rc = dt_trans_start_local(env, dev, th);
1582 dt_write_lock(env, obj, 0);
1583 if (unlikely(lfsck_is_dead_obj(obj)))
1584 GOTO(unlock2, rc = -ENOENT);
1586 rc = lfsck_links_read2(env, obj, &ldata_new);
1590 /* The specified linkEA entry has been removed by race. */
1591 rc = linkea_links_find(&ldata_new, cname, pfid);
1593 GOTO(unlock2, rc = 0);
1595 if (bk->lb_param & LPF_DRYRUN)
1596 GOTO(unlock2, rc = 1);
1599 linkea_del_buf(&ldata_new, cname);
1601 lfsck_namespace_filter_linkea_entry(&ldata_new, cname, pfid,
1604 if (linkea_buf.lb_len < ldata_new.ld_leh->leh_len) {
1605 dt_write_unlock(env, obj);
1606 dt_trans_stop(env, dev, th);
1607 lfsck_buf_init(&linkea_buf, ldata_new.ld_buf->lb_buf,
1608 ldata_new.ld_leh->leh_len);
1612 lfsck_buf_init(&linkea_buf, ldata_new.ld_buf->lb_buf,
1613 ldata_new.ld_leh->leh_len);
1614 rc = dt_xattr_set(env, obj, &linkea_buf, XATTR_NAME_LINK, 0, th);
1616 GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
1619 dt_write_unlock(env, obj);
1622 dt_trans_stop(env, dev, th);
1625 lfsck_ibits_unlock(&lh, LCK_EX);
1628 CDEBUG(D_LFSCK, "%s: namespace LFSCK remove %s linkEA entry "
1629 "for the object: "DFID", parent "DFID", name %.*s\n",
1630 lfsck_lfsck2name(lfsck), next ? "invalid" : "redundant",
1631 PFID(lfsck_dto2fid(obj)), PFID(pfid), cname->ln_namelen,
1635 struct lfsck_namespace *ns = com->lc_file_ram;
1637 ns->ln_flags |= LF_INCONSISTENT;
1644 * Conditionally remove the specified entry from the linkEA.
1646 * Take the parent lock firstly, then check whether the specified
1647 * name entry exists or not: if yes, do nothing; otherwise, call
1648 * lfsck_namespace_shrink_linkea() to remove the linkea entry.
1650 * \param[in] env pointer to the thread context
1651 * \param[in] com pointer to the lfsck component
1652 * \param[in] parent pointer to the parent directory
1653 * \param[in] child pointer to the child object that holds the linkEA
1654 * \param[in,out]ldata pointer to the buffer that holds the linkEA
1655 * \param[in] cname the name for the child in the parent directory
1656 * \param[in] pfid the parent directory's FID for the linkEA
1658 * \retval positive number for repaired cases
1659 * \retval 0 if nothing to be repaired
1660 * \retval negative error number on failure
1662 static int lfsck_namespace_shrink_linkea_cond(const struct lu_env *env,
1663 struct lfsck_component *com,
1664 struct dt_object *parent,
1665 struct dt_object *child,
1666 struct linkea_data *ldata,
1667 struct lu_name *cname,
1668 struct lu_fid *pfid)
1670 struct lfsck_thread_info *info = lfsck_env_info(env);
1671 struct lu_fid *cfid = &info->lti_fid3;
1672 struct lfsck_lock_handle *llh = &info->lti_llh;
1676 rc = lfsck_lock(env, com->lc_lfsck, parent, cname->ln_name, llh,
1677 MDS_INODELOCK_UPDATE, LCK_PR);
1681 dt_read_lock(env, parent, 0);
1682 if (unlikely(lfsck_is_dead_obj(parent))) {
1683 dt_read_unlock(env, parent);
1685 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
1691 rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
1692 (const struct dt_key *)cname->ln_name);
1693 dt_read_unlock(env, parent);
1695 /* It is safe to release the ldlm lock, because when the logic come
1696 * here, we have got all the needed information above whether the
1697 * linkEA entry is valid or not. It is not important that others
1698 * may add new linkEA entry after the ldlm lock released. If other
1699 * has removed the specified linkEA entry by race, then it is OK,
1700 * because the subsequent lfsck_namespace_shrink_linkea() can handle
1703 if (rc == -ENOENT) {
1704 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
1713 /* The LFSCK just found some internal status of cross-MDTs
1714 * create operation. That is normal. */
1715 if (lu_fid_eq(cfid, lfsck_dto2fid(child))) {
1716 linkea_next_entry(ldata);
1721 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata, cname,
1728 * Conditionally replace name entry in the parent.
1730 * As required, the LFSCK may re-create the lost MDT-object for dangling
1731 * name entry, but such repairing may be wrong because of bad FID in the
1732 * name entry. As the LFSCK processing, the real MDT-object may be found,
1733 * then the LFSCK should check whether the former re-created MDT-object
1734 * has been modified or not, if not, then destroy it and update the name
1735 * entry in the parent to reference the real MDT-object.
1737 * \param[in] env pointer to the thread context
1738 * \param[in] com pointer to the lfsck component
1739 * \param[in] parent pointer to the parent directory
1740 * \param[in] child pointer to the MDT-object that may be the real
1741 * MDT-object corresponding to the name entry in parent
1742 * \param[in] cfid the current FID in the name entry
1743 * \param[in] cname contains the name of the child in the parent directory
1745 * \retval positive number for repaired cases
1746 * \retval 0 if nothing to be repaired
1747 * \retval negative error number on failure
1749 static int lfsck_namespace_replace_cond(const struct lu_env *env,
1750 struct lfsck_component *com,
1751 struct dt_object *parent,
1752 struct dt_object *child,
1753 const struct lu_fid *cfid,
1754 const struct lu_name *cname)
1756 struct lfsck_thread_info *info = lfsck_env_info(env);
1757 struct lu_attr *la = &info->lti_la;
1758 struct dt_insert_rec *rec = &info->lti_dt_rec;
1760 struct lfsck_instance *lfsck = com->lc_lfsck;
1761 /* The child and its name may be on different MDTs. */
1762 struct dt_device *dev = lfsck->li_next;
1763 const char *name = cname->ln_name;
1764 const struct lu_fid *pfid = lfsck_dto2fid(parent);
1765 struct dt_object *cobj = NULL;
1766 struct lfsck_lock_handle *pllh = &info->lti_llh;
1767 struct lustre_handle clh = { 0 };
1768 struct linkea_data ldata = { NULL };
1769 struct thandle *th = NULL;
1774 /* @parent/@child may be based on lfsck->li_bottom,
1775 * but here we need the object based on the lfsck->li_next. */
1777 parent = lfsck_object_locate(dev, parent);
1779 GOTO(log, rc = PTR_ERR(parent));
1781 if (unlikely(!dt_try_as_dir(env, parent)))
1782 GOTO(log, rc = -ENOTDIR);
1784 rc = lfsck_lock(env, lfsck, parent, name, pllh,
1785 MDS_INODELOCK_UPDATE, LCK_PW);
1789 if (!fid_is_sane(cfid)) {
1794 cobj = lfsck_object_find_by_dev(env, dev, cfid);
1797 if (rc == -ENOENT) {
1805 if (!dt_object_exists(cobj)) {
1810 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
1811 (const struct dt_key *)name);
1812 if (rc == -ENOENT) {
1820 /* Someone changed the name entry, cannot replace it. */
1821 if (!lu_fid_eq(cfid, &tfid))
1824 /* lock the object to be destroyed. */
1825 rc = lfsck_ibits_lock(env, lfsck, cobj, &clh,
1826 MDS_INODELOCK_UPDATE |
1827 MDS_INODELOCK_UPDATE | MDS_INODELOCK_XATTR,
1832 if (unlikely(lfsck_is_dead_obj(cobj))) {
1837 rc = dt_attr_get(env, cobj, la);
1841 /* The object has been modified by other(s), or it is not created by
1842 * LFSCK, the two cases are indistinguishable. So cannot replace it. */
1843 if (la->la_ctime != 0)
1846 if (S_ISREG(la->la_mode)) {
1847 rc = dt_xattr_get(env, cobj, &LU_BUF_NULL, XATTR_NAME_LOV);
1848 /* If someone has created related OST-object(s),
1850 if ((rc > 0) || (rc < 0 && rc != -ENODATA))
1851 GOTO(log, rc = (rc > 0 ? 0 : rc));
1855 dt_read_lock(env, child, 0);
1856 rc = lfsck_links_read2(env, child, &ldata);
1857 dt_read_unlock(env, child);
1859 /* Someone changed the child, no need to replace. */
1866 rc = linkea_links_find(&ldata, cname, pfid);
1867 /* Someone moved the child, no need to replace. */
1871 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1874 th = dt_trans_create(env, dev);
1876 GOTO(log, rc = PTR_ERR(th));
1879 rc = dt_declare_destroy(env, cobj, th);
1884 rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
1888 rec->rec_type = S_IFDIR;
1889 rec->rec_fid = lfsck_dto2fid(child);
1890 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1891 (const struct dt_key *)name, th);
1895 rc = dt_trans_start_local(env, dev, th);
1900 rc = dt_destroy(env, cobj, th);
1905 /* The old name entry maybe not exist. */
1906 rc = dt_delete(env, parent, (const struct dt_key *)name, th);
1907 if (rc != 0 && rc != -ENOENT)
1910 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1911 (const struct dt_key *)name, th, 1);
1913 GOTO(stop, rc = (rc == 0 ? 1 : rc));
1916 dt_trans_stop(env, dev, th);
1919 lfsck_ibits_unlock(&clh, LCK_EX);
1922 if (cobj != NULL && !IS_ERR(cobj))
1923 lfsck_object_put(env, cobj);
1925 CDEBUG(D_LFSCK, "%s: namespace LFSCK conditionally destroy the "
1926 "object "DFID" because of conflict with the object "DFID
1927 " under the parent "DFID" with name %s: rc = %d\n",
1928 lfsck_lfsck2name(lfsck), PFID(cfid),
1929 PFID(lfsck_dto2fid(child)), PFID(pfid), name, rc);
1935 * Overwrite the linkEA for the object with the given ldata.
1937 * The caller should take the ldlm lock before the calling.
1939 * \param[in] env pointer to the thread context
1940 * \param[in] com pointer to the lfsck component
1941 * \param[in] obj pointer to the dt_object to be handled
1942 * \param[in] ldata pointer to the new linkEA data
1944 * \retval positive number for repaired cases
1945 * \retval 0 if nothing to be repaired
1946 * \retval negative error number on failure
1948 int lfsck_namespace_rebuild_linkea(const struct lu_env *env,
1949 struct lfsck_component *com,
1950 struct dt_object *obj,
1951 struct linkea_data *ldata)
1953 struct lfsck_instance *lfsck = com->lc_lfsck;
1954 struct dt_device *dev = lfsck_obj2dev(obj);
1955 struct thandle *th = NULL;
1956 struct lu_buf linkea_buf;
1960 th = dt_trans_create(env, dev);
1962 GOTO(log, rc = PTR_ERR(th));
1964 lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
1965 ldata->ld_leh->leh_len);
1966 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
1967 XATTR_NAME_LINK, 0, th);
1971 rc = dt_trans_start_local(env, dev, th);
1975 dt_write_lock(env, obj, 0);
1976 if (unlikely(lfsck_is_dead_obj(obj)))
1977 GOTO(unlock, rc = 0);
1979 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1980 GOTO(unlock, rc = 1);
1982 rc = dt_xattr_set(env, obj, &linkea_buf,
1983 XATTR_NAME_LINK, 0, th);
1985 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
1988 dt_write_unlock(env, obj);
1991 dt_trans_stop(env, dev, th);
1994 CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild linkEA for the "
1995 "object "DFID": rc = %d\n",
1996 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
1999 struct lfsck_namespace *ns = com->lc_file_ram;
2001 ns->ln_flags |= LF_INCONSISTENT;
2008 * Repair invalid name entry.
2010 * If the name entry contains invalid information, such as bad file type
2011 * or (and) corrupted object FID, then either remove the name entry or
2012 * udpate the name entry with the given (right) information.
2014 * \param[in] env pointer to the thread context
2015 * \param[in] com pointer to the lfsck component
2016 * \param[in] parent pointer to the parent directory
2017 * \param[in] child pointer to the object referenced by the name entry
2018 * \param[in] name the old name of the child under the parent directory
2019 * \param[in] name2 the new name of the child under the parent directory
2020 * \param[in] type the type claimed by the name entry
2021 * \param[in] update update the name entry if true; otherwise, remove it
2022 * \param[in] dec decrease the parent nlink count if true
2024 * \retval positive number for repaired successfully
2025 * \retval 0 if nothing to be repaired
2026 * \retval negative error number on failure
2028 int lfsck_namespace_repair_dirent(const struct lu_env *env,
2029 struct lfsck_component *com,
2030 struct dt_object *parent,
2031 struct dt_object *child,
2032 const char *name, const char *name2,
2033 __u16 type, bool update, bool dec)
2035 struct lfsck_thread_info *info = lfsck_env_info(env);
2036 struct dt_insert_rec *rec = &info->lti_dt_rec;
2037 const struct lu_fid *pfid = lfsck_dto2fid(parent);
2038 const struct lu_fid *cfid = lfsck_dto2fid(child);
2040 struct lfsck_instance *lfsck = com->lc_lfsck;
2041 struct dt_device *dev = lfsck->li_next;
2042 struct thandle *th = NULL;
2043 struct lfsck_lock_handle *llh = &info->lti_llh;
2044 struct lustre_handle lh = { 0 };
2048 parent = lfsck_object_locate(dev, parent);
2050 GOTO(log, rc = PTR_ERR(parent));
2052 if (unlikely(!dt_try_as_dir(env, parent)))
2053 GOTO(log, rc = -ENOTDIR);
2055 if (!update || strcmp(name, name2) == 0)
2056 rc = lfsck_lock(env, lfsck, parent, name, llh,
2057 MDS_INODELOCK_UPDATE, LCK_PW);
2059 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
2060 MDS_INODELOCK_UPDATE, LCK_PW);
2064 th = dt_trans_create(env, dev);
2066 GOTO(unlock1, rc = PTR_ERR(th));
2068 rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
2073 rec->rec_type = lfsck_object_type(child) & S_IFMT;
2074 rec->rec_fid = cfid;
2075 rc = dt_declare_insert(env, parent,
2076 (const struct dt_rec *)rec,
2077 (const struct dt_key *)name2, th);
2083 rc = dt_declare_ref_del(env, parent, th);
2088 rc = dt_trans_start_local(env, dev, th);
2093 dt_write_lock(env, parent, 0);
2094 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
2095 (const struct dt_key *)name);
2096 /* Someone has removed the bad name entry by race. */
2098 GOTO(unlock2, rc = 0);
2103 /* Someone has removed the bad name entry and reused it for other
2104 * object by race. */
2105 if (!lu_fid_eq(&tfid, cfid))
2106 GOTO(unlock2, rc = 0);
2108 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2109 GOTO(unlock2, rc = 1);
2111 rc = dt_delete(env, parent, (const struct dt_key *)name, th);
2116 rc = dt_insert(env, parent,
2117 (const struct dt_rec *)rec,
2118 (const struct dt_key *)name2, th, 1);
2124 rc = dt_ref_del(env, parent, th);
2129 GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
2132 dt_write_unlock(env, parent);
2135 dt_trans_stop(env, dev, th);
2137 /* We are not sure whether the child will become orphan or not.
2138 * Record it in the LFSCK trace file for further checking in
2139 * the second-stage scanning. */
2140 if (!update && !dec && rc == 0)
2141 lfsck_namespace_trace_update(env, com, cfid,
2142 LNTF_CHECK_LINKEA, true);
2145 /* It is harmless even if unlock the unused lock_handle */
2146 lfsck_ibits_unlock(&lh, LCK_PW);
2150 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found bad name "
2151 "entry for: parent "DFID", child "DFID", name %s, type "
2152 "in name entry %o, type claimed by child %o. repair it "
2153 "by %s with new name2 %s: rc = %d\n",
2154 lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
2155 name, type, update ? lfsck_object_type(child) : 0,
2156 update ? "updating" : "removing", name2, rc);
2159 struct lfsck_namespace *ns = com->lc_file_ram;
2161 ns->ln_flags |= LF_INCONSISTENT;
2168 * Update the ".." name entry for the given object.
2170 * The object's ".." is corrupted, this function will update the ".." name
2171 * entry with the given pfid, and the linkEA with the given ldata.
2173 * The caller should take the ldlm lock before the calling.
2175 * \param[in] env pointer to the thread context
2176 * \param[in] com pointer to the lfsck component
2177 * \param[in] obj pointer to the dt_object to be handled
2178 * \param[in] pfid the new fid for the object's ".." name entry
2179 * \param[in] cname the name for the @obj in the parent directory
2181 * \retval positive number for repaired cases
2182 * \retval 0 if nothing to be repaired
2183 * \retval negative error number on failure
2185 static int lfsck_namespace_repair_unmatched_pairs(const struct lu_env *env,
2186 struct lfsck_component *com,
2187 struct dt_object *obj,
2188 const struct lu_fid *pfid,
2189 struct lu_name *cname)
2191 struct lfsck_thread_info *info = lfsck_env_info(env);
2192 struct dt_insert_rec *rec = &info->lti_dt_rec;
2193 struct lfsck_instance *lfsck = com->lc_lfsck;
2194 struct dt_device *dev = lfsck_obj2dev(obj);
2195 struct thandle *th = NULL;
2196 struct linkea_data ldata = { NULL };
2197 struct lu_buf linkea_buf;
2201 LASSERT(!dt_object_remote(obj));
2202 LASSERT(S_ISDIR(lfsck_object_type(obj)));
2204 rc = linkea_data_new(&ldata, &info->lti_big_buf);
2208 rc = linkea_add_buf(&ldata, cname, pfid);
2212 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
2213 ldata.ld_leh->leh_len);
2215 th = dt_trans_create(env, dev);
2217 GOTO(log, rc = PTR_ERR(th));
2219 rc = dt_declare_delete(env, obj, (const struct dt_key *)dotdot, th);
2223 rec->rec_type = S_IFDIR;
2224 rec->rec_fid = pfid;
2225 rc = dt_declare_insert(env, obj, (const struct dt_rec *)rec,
2226 (const struct dt_key *)dotdot, th);
2230 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
2231 XATTR_NAME_LINK, 0, th);
2235 rc = dt_trans_start_local(env, dev, th);
2239 dt_write_lock(env, obj, 0);
2240 if (unlikely(lfsck_is_dead_obj(obj)))
2241 GOTO(unlock, rc = 0);
2243 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2244 GOTO(unlock, rc = 1);
2246 /* The old ".." name entry maybe not exist. */
2247 dt_delete(env, obj, (const struct dt_key *)dotdot, th);
2249 rc = dt_insert(env, obj, (const struct dt_rec *)rec,
2250 (const struct dt_key *)dotdot, th, 1);
2254 rc = dt_xattr_set(env, obj, &linkea_buf,
2255 XATTR_NAME_LINK, 0, th);
2257 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2260 dt_write_unlock(env, obj);
2263 dt_trans_stop(env, dev, th);
2266 CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild dotdot name entry for "
2267 "the object "DFID", new parent "DFID": rc = %d\n",
2268 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)),
2272 struct lfsck_namespace *ns = com->lc_file_ram;
2274 ns->ln_flags |= LF_INCONSISTENT;
2281 * Handle orphan @obj during Double Scan Directory.
2283 * Remove the @obj's current (invalid) linkEA entries, and insert
2284 * it in the directory .lustre/lost+found/MDTxxxx/ with the name:
2285 * ${FID}-${PFID}-D-${conflict_version}
2287 * The caller should take the ldlm lock before the calling.
2289 * \param[in] env pointer to the thread context
2290 * \param[in] com pointer to the lfsck component
2291 * \param[in] obj pointer to the orphan object to be handled
2292 * \param[in] pfid the new fid for the object's ".." name entry
2293 * \param[in,out] lh ldlm lock handler for the given @obj
2294 * \param[out] type to tell the caller what the inconsistency is
2296 * \retval positive number for repaired cases
2297 * \retval 0 if nothing to be repaired
2298 * \retval negative error number on failure
2301 lfsck_namespace_dsd_orphan(const struct lu_env *env,
2302 struct lfsck_component *com,
2303 struct dt_object *obj,
2304 const struct lu_fid *pfid,
2305 struct lustre_handle *lh,
2306 enum lfsck_namespace_inconsistency_type *type)
2308 struct lfsck_thread_info *info = lfsck_env_info(env);
2309 struct lfsck_namespace *ns = com->lc_file_ram;
2313 /* Remove the unrecognized linkEA. */
2314 rc = lfsck_namespace_links_remove(env, com, obj);
2315 lfsck_ibits_unlock(lh, LCK_EX);
2316 if (rc < 0 && rc != -ENODATA)
2319 *type = LNIT_MUL_REF;
2321 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
2322 * ever tried to verify some remote MDT-object that resides on this
2323 * MDT, but this MDT failed to respond such request. So means there
2324 * may be some remote name entry on other MDT that references this
2325 * object with another name, so we cannot know whether this linkEA
2326 * is valid or not. So keep it there and maybe resolved when next
2328 if (ns->ln_flags & LF_INCOMPLETE)
2331 /* The unique linkEA is invalid, even if the ".." name entry may be
2332 * valid, we still cannot know via which name entry this directory
2333 * will be referenced. Then handle it as pure orphan. */
2334 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2335 "-"DFID, PFID(pfid));
2336 rc = lfsck_namespace_insert_orphan(env, com, obj,
2337 info->lti_tmpbuf, "D", NULL);
2343 * Double Scan Directory object for single linkEA entry case.
2345 * The given @child has unique linkEA entry. If the linkEA entry is valid,
2346 * then check whether the name is in the namespace or not, if not, add the
2347 * missing name entry back to namespace. If the linkEA entry is invalid,
2348 * then remove it and insert the @child in the .lustre/lost+found/MDTxxxx/
2351 * \param[in] env pointer to the thread context
2352 * \param[in] com pointer to the lfsck component
2353 * \param[in] child pointer to the directory to be double scanned
2354 * \param[in] pfid the FID corresponding to the ".." entry
2355 * \param[in] ldata pointer to the linkEA data for the given @child
2356 * \param[in,out] lh ldlm lock handler for the given @child
2357 * \param[out] type to tell the caller what the inconsistency is
2358 * \param[in] retry if found inconsistency, but the caller does not hold
2359 * ldlm lock on the @child, then set @retry as true
2361 * \retval positive number for repaired cases
2362 * \retval 0 if nothing to be repaired
2363 * \retval negative error number on failure
2366 lfsck_namespace_dsd_single(const struct lu_env *env,
2367 struct lfsck_component *com,
2368 struct dt_object *child,
2369 const struct lu_fid *pfid,
2370 struct linkea_data *ldata,
2371 struct lustre_handle *lh,
2372 enum lfsck_namespace_inconsistency_type *type,
2375 struct lfsck_thread_info *info = lfsck_env_info(env);
2376 struct lu_name *cname = &info->lti_name;
2377 const struct lu_fid *cfid = lfsck_dto2fid(child);
2379 struct lfsck_namespace *ns = com->lc_file_ram;
2380 struct lfsck_instance *lfsck = com->lc_lfsck;
2381 struct dt_object *parent = NULL;
2382 struct lmv_mds_md_v1 *lmv;
2386 lfsck_namespace_unpack_linkea_entry(ldata, cname, &tfid, info->lti_key);
2387 /* The unique linkEA entry with bad parent will be handled as orphan. */
2388 if (!fid_is_sane(&tfid)) {
2389 if (!lustre_handle_is_used(lh) && retry != NULL)
2392 rc = lfsck_namespace_dsd_orphan(env, com, child,
2398 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
2400 GOTO(out, rc = PTR_ERR(parent));
2402 /* We trust the unique linkEA entry in spite of whether it matches the
2403 * ".." name entry or not. Because even if the linkEA entry is wrong
2404 * and the ".." name entry is right, we still cannot know via which
2405 * name entry the child will be referenced, since all known entries
2406 * have been verified during the first-stage scanning. */
2407 if (!dt_object_exists(parent)) {
2408 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2409 * has ever tried to verify some remote MDT-object that resides
2410 * on this MDT, but this MDT failed to respond such request. So
2411 * means there may be some remote name entry on other MDT that
2412 * references this object with another name, so we cannot know
2413 * whether this linkEA is valid or not. So keep it there and
2414 * maybe resolved when next LFSCK run. */
2415 if (ns->ln_flags & LF_INCOMPLETE)
2418 if (!lustre_handle_is_used(lh) && retry != NULL) {
2424 lfsck_ibits_unlock(lh, LCK_EX);
2427 lmv = &info->lti_lmv;
2428 rc = lfsck_read_stripe_lmv(env, child, lmv);
2429 if (rc != 0 && rc != -ENODATA)
2432 if (rc == -ENODATA || lmv->lmv_magic != LMV_MAGIC_STRIPE) {
2434 } else if (lfsck_shard_name_to_index(env,
2435 cname->ln_name, cname->ln_namelen,
2436 S_IFDIR, cfid) < 0) {
2437 /* It is an invalid name entry, we
2438 * cannot trust the parent also. */
2439 rc = lfsck_namespace_shrink_linkea(env, com, child,
2440 ldata, cname, &tfid, true);
2444 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2445 "-"DFID, PFID(pfid));
2446 rc = lfsck_namespace_insert_orphan(env, com, child,
2447 info->lti_tmpbuf, "S", NULL);
2452 /* Create the lost parent as an orphan. */
2453 rc = lfsck_namespace_create_orphan_dir(env, com, parent, lmv);
2455 /* Add the missing name entry to the parent. */
2456 rc = lfsck_namespace_insert_normal(env, com, parent,
2457 child, cname->ln_name);
2458 if (unlikely(rc == -EEXIST)) {
2459 /* Unfortunately, someone reused the name
2460 * under the parent by race. So we have
2461 * to remove the linkEA entry from
2462 * current child object. It means that the
2463 * LFSCK cannot recover the system
2464 * totally back to its original status,
2465 * but it is necessary to make the
2466 * current system to be consistent. */
2467 rc = lfsck_namespace_shrink_linkea(env,
2469 cname, &tfid, true);
2471 snprintf(info->lti_tmpbuf,
2472 sizeof(info->lti_tmpbuf),
2473 "-"DFID, PFID(pfid));
2474 rc = lfsck_namespace_insert_orphan(env,
2475 com, child, info->lti_tmpbuf,
2484 /* The unique linkEA entry with bad parent will be handled as orphan. */
2485 if (unlikely(!dt_try_as_dir(env, parent))) {
2486 if (!lustre_handle_is_used(lh) && retry != NULL)
2489 rc = lfsck_namespace_dsd_orphan(env, com, child,
2495 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
2496 (const struct dt_key *)cname->ln_name);
2497 if (rc == -ENOENT) {
2498 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2499 * has ever tried to verify some remote MDT-object that resides
2500 * on this MDT, but this MDT failed to respond such request. So
2501 * means there may be some remote name entry on other MDT that
2502 * references this object with another name, so we cannot know
2503 * whether this linkEA is valid or not. So keep it there and
2504 * maybe resolved when next LFSCK run. */
2505 if (ns->ln_flags & LF_INCOMPLETE)
2508 if (!lustre_handle_is_used(lh) && retry != NULL) {
2514 lfsck_ibits_unlock(lh, LCK_EX);
2515 rc = lfsck_namespace_check_name(env, parent, child, cname);
2522 /* It is an invalid name entry, drop it. */
2523 if (unlikely(rc > 0)) {
2524 rc = lfsck_namespace_shrink_linkea(env, com, child,
2525 ldata, cname, &tfid, true);
2527 snprintf(info->lti_tmpbuf,
2528 sizeof(info->lti_tmpbuf),
2529 "-"DFID, PFID(pfid));
2530 rc = lfsck_namespace_insert_orphan(env, com,
2531 child, info->lti_tmpbuf, "D", NULL);
2537 /* Add the missing name entry back to the namespace. */
2538 rc = lfsck_namespace_insert_normal(env, com, parent, child,
2540 if (unlikely(rc == -ESTALE))
2541 /* It may happen when the remote object has been
2542 * removed, but the local MDT is not aware of that. */
2545 if (unlikely(rc == -EEXIST)) {
2546 /* Unfortunately, someone reused the name under the
2547 * parent by race. So we have to remove the linkEA
2548 * entry from current child object. It means that the
2549 * LFSCK cannot recover the system totally back to
2550 * its original status, but it is necessary to make
2551 * the current system to be consistent.
2553 * It also may be because of the LFSCK found some
2554 * internal status of create operation. Under such
2555 * case, nothing to be done. */
2556 rc = lfsck_namespace_shrink_linkea_cond(env, com,
2557 parent, child, ldata, cname, &tfid);
2559 snprintf(info->lti_tmpbuf,
2560 sizeof(info->lti_tmpbuf),
2561 "-"DFID, PFID(pfid));
2562 rc = lfsck_namespace_insert_orphan(env, com,
2563 child, info->lti_tmpbuf, "D", NULL);
2573 if (!lu_fid_eq(&tfid, cfid)) {
2574 if (!lustre_handle_is_used(lh) && retry != NULL) {
2580 lfsck_ibits_unlock(lh, LCK_EX);
2581 /* The name entry references another MDT-object that
2582 * may be created by the LFSCK for repairing dangling
2583 * name entry. Try to replace it. */
2584 rc = lfsck_namespace_replace_cond(env, com, parent, child,
2587 rc = lfsck_namespace_dsd_orphan(env, com, child,
2593 if (fid_is_zero(pfid))
2596 /* The ".." name entry is wrong, update it. */
2597 if (!lu_fid_eq(pfid, lfsck_dto2fid(parent))) {
2598 if (!lustre_handle_is_used(lh) && retry != NULL) {
2604 *type = LNIT_UNMATCHED_PAIRS;
2605 rc = lfsck_namespace_repair_unmatched_pairs(env, com, child,
2606 lfsck_dto2fid(parent), cname);
2612 if (parent != NULL && !IS_ERR(parent))
2613 lfsck_object_put(env, parent);
2619 * Double Scan Directory object for multiple linkEA entries case.
2621 * The given @child has multiple linkEA entries. There is at most one linkEA
2622 * entry will be valid, all the others will be removed. Firstly, the function
2623 * will try to find out the linkEA entry for which the name entry exists under
2624 * the given parent (@pfid). If there is no linkEA entry that matches the given
2625 * ".." name entry, then tries to find out the first linkEA entry that both the
2626 * parent and the name entry exist to rebuild a new ".." name entry.
2628 * \param[in] env pointer to the thread context
2629 * \param[in] com pointer to the lfsck component
2630 * \param[in] child pointer to the directory to be double scanned
2631 * \param[in] pfid the FID corresponding to the ".." entry
2632 * \param[in] ldata pointer to the linkEA data for the given @child
2633 * \param[in,out] lh ldlm lock handler for the given @child
2634 * \param[out] type to tell the caller what the inconsistency is
2635 * \param[in] lpf true if the ".." entry is under lost+found/MDTxxxx/
2637 * \retval positive number for repaired cases
2638 * \retval 0 if nothing to be repaired
2639 * \retval negative error number on failure
2642 lfsck_namespace_dsd_multiple(const struct lu_env *env,
2643 struct lfsck_component *com,
2644 struct dt_object *child,
2645 const struct lu_fid *pfid,
2646 struct linkea_data *ldata,
2647 struct lustre_handle *lh,
2648 enum lfsck_namespace_inconsistency_type *type,
2651 struct lfsck_thread_info *info = lfsck_env_info(env);
2652 struct lu_name *cname = &info->lti_name;
2653 const struct lu_fid *cfid = lfsck_dto2fid(child);
2654 struct lu_fid *pfid2 = &info->lti_fid3;
2656 struct lfsck_namespace *ns = com->lc_file_ram;
2657 struct lfsck_instance *lfsck = com->lc_lfsck;
2658 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
2659 struct dt_object *parent = NULL;
2660 struct linkea_data ldata_new = { NULL };
2661 int dirent_count = 0;
2662 int linkea_count = 0;
2668 while (ldata->ld_lee != NULL) {
2669 lfsck_namespace_unpack_linkea_entry(ldata, cname, &tfid,
2671 /* Drop repeated linkEA entries. */
2672 lfsck_namespace_filter_linkea_entry(ldata, cname, &tfid, true);
2673 /* Drop invalid linkEA entry. */
2674 if (!fid_is_sane(&tfid)) {
2675 linkea_del_buf(ldata, cname);
2680 /* If current dotdot is the .lustre/lost+found/MDTxxxx/,
2681 * then it is possible that: the directry object has ever
2682 * been lost, but its name entry was there. In the former
2683 * LFSCK run, during the first-stage scanning, the LFSCK
2684 * found the dangling name entry, but it did not recreate
2685 * the lost object, and when moved to the second-stage
2686 * scanning, some children objects of the lost directory
2687 * object were found, then the LFSCK recreated such lost
2688 * directory object as an orphan.
2690 * When the LFSCK runs again, if the dangling name is still
2691 * there, the LFSCK should move the orphan directory object
2692 * back to the normal namespace. */
2693 if (!lpf && !lu_fid_eq(pfid, &tfid) && once) {
2694 linkea_next_entry(ldata);
2698 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
2700 RETURN(PTR_ERR(parent));
2702 if (!dt_object_exists(parent)) {
2703 lfsck_object_put(env, parent);
2704 if (ldata->ld_leh->leh_reccount > 1) {
2705 /* If it is NOT the last linkEA entry, then
2706 * there is still other chance to make the
2707 * child to be visible via other parent, then
2708 * remove this linkEA entry. */
2709 linkea_del_buf(ldata, cname);
2717 /* The linkEA entry with bad parent will be removed. */
2718 if (unlikely(!dt_try_as_dir(env, parent))) {
2719 lfsck_object_put(env, parent);
2720 linkea_del_buf(ldata, cname);
2725 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
2726 (const struct dt_key *)cname->ln_name);
2727 *pfid2 = *lfsck_dto2fid(parent);
2728 if (rc == -ENOENT) {
2729 lfsck_object_put(env, parent);
2730 linkea_next_entry(ldata);
2735 lfsck_object_put(env, parent);
2740 if (lu_fid_eq(&tfid, cfid)) {
2741 lfsck_object_put(env, parent);
2742 if (!lu_fid_eq(pfid, pfid2)) {
2743 *type = LNIT_UNMATCHED_PAIRS;
2744 rc = lfsck_namespace_repair_unmatched_pairs(env,
2745 com, child, pfid2, cname);
2751 /* It is the most common case that we find the
2752 * name entry corresponding to the linkEA entry
2753 * that matches the ".." name entry. */
2754 rc = linkea_data_new(&ldata_new, &info->lti_big_buf);
2758 rc = linkea_add_buf(&ldata_new, cname, pfid2);
2762 rc = lfsck_namespace_rebuild_linkea(env, com, child,
2767 linkea_del_buf(ldata, cname);
2769 linkea_first_entry(ldata);
2770 /* There may be some invalid dangling name entries under
2771 * other parent directories, remove all of them. */
2772 while (ldata->ld_lee != NULL) {
2773 lfsck_namespace_unpack_linkea_entry(ldata,
2774 cname, &tfid, info->lti_key);
2775 if (!fid_is_sane(&tfid))
2778 parent = lfsck_object_find_bottom(env, lfsck,
2780 if (IS_ERR(parent)) {
2781 rc = PTR_ERR(parent);
2782 if (rc != -ENOENT &&
2783 bk->lb_param & LPF_FAILOUT)
2789 if (!dt_object_exists(parent)) {
2790 lfsck_object_put(env, parent);
2794 rc = lfsck_namespace_repair_dirent(env, com,
2795 parent, child, cname->ln_name,
2796 cname->ln_name, S_IFDIR, false, true);
2797 lfsck_object_put(env, parent);
2799 if (bk->lb_param & LPF_FAILOUT)
2808 linkea_del_buf(ldata, cname);
2811 ns->ln_dirent_repaired += dirent_count;
2816 lfsck_ibits_unlock(lh, LCK_EX);
2817 /* The name entry references another MDT-object that may be
2818 * created by the LFSCK for repairing dangling name entry.
2819 * Try to replace it. */
2820 rc = lfsck_namespace_replace_cond(env, com, parent, child,
2822 lfsck_object_put(env, parent);
2829 linkea_del_buf(ldata, cname);
2832 linkea_first_entry(ldata);
2833 if (ldata->ld_leh->leh_reccount == 1) {
2834 rc = lfsck_namespace_dsd_single(env, com, child, pfid, ldata,
2837 if (rc == 0 && fid_is_zero(pfid) && linkea_count > 0)
2838 rc = lfsck_namespace_rebuild_linkea(env, com, child,
2844 /* All linkEA entries are invalid and removed, then handle the @child
2846 if (ldata->ld_leh->leh_reccount == 0) {
2847 rc = lfsck_namespace_dsd_orphan(env, com, child, pfid, lh,
2853 /* If the dangling name entry for the orphan directory object has
2854 * been remvoed, then just check whether the directory object is
2855 * still under the .lustre/lost+found/MDTxxxx/ or not. */
2861 /* There is no linkEA entry that matches the ".." name entry. Find
2862 * the first linkEA entry that both parent and name entry exist to
2863 * rebuild a new ".." name entry. */
2873 * Repair the object's nlink attribute.
2875 * If all the known name entries have been verified, then the object's hard
2876 * link attribute should match the object's linkEA entries count unless the
2877 * object's has too much hard link to be recorded in the linkEA. Such cases
2878 * should have been marked in the LFSCK trace file. Otherwise, trust the
2879 * linkEA to update the object's nlink attribute.
2881 * \param[in] env pointer to the thread context
2882 * \param[in] com pointer to the lfsck component
2883 * \param[in] obj pointer to the dt_object to be handled
2884 * \param[in,out] la pointer to buffer to object's attribute before
2885 * and after the repairing
2887 * \retval positive number for repaired cases
2888 * \retval 0 if nothing to be repaired
2889 * \retval negative error number on failure
2891 static int lfsck_namespace_repair_nlink(const struct lu_env *env,
2892 struct lfsck_component *com,
2893 struct dt_object *obj,
2896 struct lfsck_thread_info *info = lfsck_env_info(env);
2897 struct lu_fid *tfid = &info->lti_fid3;
2898 struct lfsck_namespace *ns = com->lc_file_ram;
2899 struct lfsck_instance *lfsck = com->lc_lfsck;
2900 struct dt_device *dev = lfsck_obj2dev(obj);
2901 const struct lu_fid *cfid = lfsck_dto2fid(obj);
2902 struct thandle *th = NULL;
2903 struct linkea_data ldata = { NULL };
2904 struct lustre_handle lh = { 0 };
2905 __u32 old = la->la_nlink;
2911 LASSERT(!dt_object_remote(obj));
2912 LASSERT(S_ISREG(lfsck_object_type(obj)));
2914 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
2915 MDS_INODELOCK_UPDATE, LCK_PW);
2919 th = dt_trans_create(env, dev);
2921 GOTO(log, rc = PTR_ERR(th));
2923 la->la_valid = LA_NLINK;
2924 rc = dt_declare_attr_set(env, obj, la, th);
2928 rc = dt_trans_start_local(env, dev, th);
2932 dt_write_lock(env, obj, 0);
2933 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
2934 * ever tried to verify some remote MDT-object that resides on this
2935 * MDT, but this MDT failed to respond such request. So means there
2936 * may be some remote name entry on other MDT that references this
2937 * object with another name, so we cannot know whether this linkEA
2938 * is valid or not. So keep it there and maybe resolved when next
2940 if (ns->ln_flags & LF_INCOMPLETE)
2941 GOTO(unlock, rc = 0);
2943 fid_cpu_to_be(tfid, cfid);
2944 idx = lfsck_sub_trace_file_fid2idx(cfid);
2945 rc = dt_lookup(env, com->lc_sub_trace_objs[idx].lsto_obj,
2946 (struct dt_rec *)&flags, (const struct dt_key *)tfid);
2950 if (flags & LNTF_SKIP_NLINK)
2951 GOTO(unlock, rc = 0);
2953 rc = dt_attr_get(env, obj, la);
2955 GOTO(unlock, rc = (rc == -ENOENT ? 0 : rc));
2957 rc = lfsck_links_read2(env, obj, &ldata);
2959 GOTO(unlock, rc = (rc == -ENODATA ? 0 : rc));
2961 if (la->la_nlink == ldata.ld_leh->leh_reccount ||
2962 unlikely(la->la_nlink == 0))
2963 GOTO(unlock, rc = 0);
2965 la->la_nlink = ldata.ld_leh->leh_reccount;
2966 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2967 GOTO(unlock, rc = 1);
2969 rc = dt_attr_set(env, obj, la, th);
2971 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2974 dt_write_unlock(env, obj);
2977 dt_trans_stop(env, dev, th);
2980 lfsck_ibits_unlock(&lh, LCK_PW);
2981 CDEBUG(D_LFSCK, "%s: namespace LFSCK repaired the object "DFID"'s "
2982 "nlink count from %u to %u: rc = %d\n",
2983 lfsck_lfsck2name(lfsck), PFID(cfid), old, la->la_nlink, rc);
2986 ns->ln_flags |= LF_INCONSISTENT;
2992 * Double scan the directory object for namespace LFSCK.
2994 * This function will verify the <parent, child> pairs in the namespace tree:
2995 * the parent references the child via some name entry that should be in the
2996 * child's linkEA entry, the child should back references the parent via its
2999 * The LFSCK will scan every linkEA entry in turn until find out the first
3000 * matched pairs. If found, then all other linkEA entries will be dropped.
3001 * If all the linkEA entries cannot match the ".." name entry, then there
3002 * are serveral possible cases:
3004 * 1) If there is only one linkEA entry, then trust it as long as the PFID
3005 * in the linkEA entry is valid.
3007 * 2) If there are multiple linkEA entries, then try to find the linkEA
3008 * that matches the ".." name entry. If found, then all other entries
3009 * are invalid; otherwise, it is quite possible that the ".." name entry
3010 * is corrupted. Under such case, the LFSCK will rebuild the ".." name
3011 * entry according to the first valid linkEA entry (both the parent and
3012 * the name entry should exist).
3014 * 3) If the directory object has no (valid) linkEA entry, then the
3015 * directory object will be handled as pure orphan and inserted
3016 * in the .lustre/lost+found/MDTxxxx/ with the name:
3017 * ${self_FID}-${PFID}-D-${conflict_version}
3019 * \param[in] env pointer to the thread context
3020 * \param[in] com pointer to the lfsck component
3021 * \param[in] child pointer to the directory object to be handled
3022 * \param[in] flags to indicate the specical checking on the @child
3024 * \retval positive number for repaired cases
3025 * \retval 0 if nothing to be repaired
3026 * \retval negative error number on failure
3028 static int lfsck_namespace_double_scan_dir(const struct lu_env *env,
3029 struct lfsck_component *com,
3030 struct dt_object *child, __u8 flags)
3032 struct lfsck_thread_info *info = lfsck_env_info(env);
3033 const struct lu_fid *cfid = lfsck_dto2fid(child);
3034 struct lu_fid *pfid = &info->lti_fid2;
3035 struct lfsck_namespace *ns = com->lc_file_ram;
3036 struct lfsck_instance *lfsck = com->lc_lfsck;
3037 struct lustre_handle lh = { 0 };
3038 struct linkea_data ldata = { NULL };
3039 bool unknown = false;
3042 enum lfsck_namespace_inconsistency_type type = LNIT_BAD_LINKEA;
3046 LASSERT(!dt_object_remote(child));
3048 if (flags & LNTF_UNCERTAIN_LMV) {
3049 if (flags & LNTF_RECHECK_NAME_HASH) {
3050 rc = lfsck_namespace_scan_shard(env, com, child);
3054 ns->ln_striped_shards_scanned++;
3056 ns->ln_striped_shards_skipped++;
3060 flags &= ~(LNTF_RECHECK_NAME_HASH | LNTF_UNCERTAIN_LMV);
3064 if (flags & (LNTF_CHECK_LINKEA | LNTF_CHECK_PARENT) &&
3065 !(lfsck->li_bookmark_ram.lb_param & LPF_ALL_TGT)) {
3066 CDEBUG(D_LFSCK, "%s: some MDT(s) maybe NOT take part in the"
3067 "the namespace LFSCK, then the LFSCK cannot guarantee"
3068 "all the name entries have been verified in first-stage"
3069 "scanning. So have to skip orphan related handling for"
3070 "the directory object "DFID" with remote name entry\n",
3071 lfsck_lfsck2name(lfsck), PFID(cfid));
3076 if (unlikely(!dt_try_as_dir(env, child)))
3077 GOTO(out, rc = -ENOTDIR);
3079 /* We only take ldlm lock on the @child when required. When the
3080 * logic comes here for the first time, it is always false. */
3084 rc = lfsck_ibits_lock(env, lfsck, child, &lh,
3085 MDS_INODELOCK_UPDATE |
3086 MDS_INODELOCK_XATTR, LCK_EX);
3091 dt_read_lock(env, child, 0);
3092 if (unlikely(lfsck_is_dead_obj(child))) {
3093 dt_read_unlock(env, child);
3098 rc = dt_lookup(env, child, (struct dt_rec *)pfid,
3099 (const struct dt_key *)dotdot);
3101 if (rc != -ENOENT && rc != -ENODATA && rc != -EINVAL) {
3102 dt_read_unlock(env, child);
3107 if (!lustre_handle_is_used(&lh)) {
3108 dt_read_unlock(env, child);
3113 } else if (lfsck->li_lpf_obj != NULL &&
3114 lu_fid_eq(pfid, lfsck_dto2fid(lfsck->li_lpf_obj))) {
3116 } else if (unlikely(!fid_is_sane(pfid))) {
3120 rc = lfsck_links_read(env, child, &ldata);
3121 dt_read_unlock(env, child);
3123 if (rc != -ENODATA && rc != -EINVAL)
3126 if (!lustre_handle_is_used(&lh))
3129 if (rc == -EINVAL && !fid_is_zero(pfid)) {
3130 /* Remove the corrupted linkEA. */
3131 rc = lfsck_namespace_links_remove(env, com, child);
3133 /* Here, because of the crashed linkEA, we
3134 * cannot know whether there is some parent
3135 * that references the child directory via
3136 * some name entry or not. So keep it there,
3137 * when the LFSCK run next time, if there is
3138 * some parent that references this object,
3139 * then the LFSCK can rebuild the linkEA;
3140 * otherwise, this object will be handled
3141 * as orphan as above. */
3144 /* 1. If we have neither ".." nor linkEA,
3145 * then it is an orphan.
3147 * 2. If we only have the ".." name entry,
3148 * but no parent references this child
3149 * directory, then handle it as orphan. */
3150 lfsck_ibits_unlock(&lh, LCK_EX);
3151 type = LNIT_MUL_REF;
3153 /* If the LFSCK is marked as LF_INCOMPLETE,
3154 * then means some MDT has ever tried to
3155 * verify some remote MDT-object that resides
3156 * on this MDT, but this MDT failed to respond
3157 * such request. So means there may be some
3158 * remote name entry on other MDT that
3159 * references this object with another name,
3160 * so we cannot know whether this linkEA is
3161 * valid or not. So keep it there and maybe
3162 * resolved when next LFSCK run. */
3163 if (ns->ln_flags & LF_INCOMPLETE)
3166 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
3167 "-"DFID, PFID(pfid));
3168 rc = lfsck_namespace_insert_orphan(env, com, child,
3169 info->lti_tmpbuf, "D", NULL);
3175 linkea_first_entry(&ldata);
3176 /* This is the most common case: the object has unique linkEA entry. */
3177 if (ldata.ld_leh->leh_reccount == 1) {
3178 rc = lfsck_namespace_dsd_single(env, com, child, pfid, &ldata,
3179 &lh, &type, &retry);
3181 LASSERT(!lustre_handle_is_used(&lh));
3190 if (!lustre_handle_is_used(&lh))
3193 if (unlikely(ldata.ld_leh->leh_reccount == 0)) {
3194 rc = lfsck_namespace_dsd_orphan(env, com, child, pfid, &lh,
3200 /* When we come here, the cases usually like that:
3201 * 1) The directory object has a corrupted linkEA entry. During the
3202 * first-stage scanning, the LFSCK cannot know such corruption,
3203 * then it appends the right linkEA entry according to the found
3204 * name entry after the bad one.
3206 * 2) The directory object has a right linkEA entry. During the
3207 * first-stage scanning, the LFSCK finds some bad name entry,
3208 * but the LFSCK cannot aware that at that time, then it adds
3209 * the bad linkEA entry for further processing. */
3210 rc = lfsck_namespace_dsd_multiple(env, com, child, pfid, &ldata,
3216 lfsck_ibits_unlock(&lh, LCK_EX);
3219 case LNIT_BAD_LINKEA:
3220 ns->ln_linkea_repaired++;
3222 case LNIT_UNMATCHED_PAIRS:
3223 ns->ln_unmatched_pairs_repaired++;
3226 ns->ln_mul_ref_repaired++;
3234 ns->ln_unknown_inconsistency++;
3240 * Double scan the MDT-object for namespace LFSCK.
3242 * If the MDT-object contains invalid or repeated linkEA entries, then drop
3243 * those entries from the linkEA; if the linkEA becomes empty or the object
3244 * has no linkEA, then it is an orphan and will be added into the directory
3245 * .lustre/lost+found/MDTxxxx/; if the remote parent is lost, then recreate
3246 * the remote parent; if the name entry corresponding to some linkEA entry
3247 * is lost, then add the name entry back to the namespace.
3249 * \param[in] env pointer to the thread context
3250 * \param[in] com pointer to the lfsck component
3251 * \param[in] child pointer to the dt_object to be handled
3252 * \param[in] flags some hints to indicate how the @child should be handled
3254 * \retval positive number for repaired cases
3255 * \retval 0 if nothing to be repaired
3256 * \retval negative error number on failure
3258 static int lfsck_namespace_double_scan_one(const struct lu_env *env,
3259 struct lfsck_component *com,
3260 struct dt_object *child, __u8 flags)
3262 struct lfsck_thread_info *info = lfsck_env_info(env);
3263 struct lu_attr *la = &info->lti_la;
3264 struct lu_name *cname = &info->lti_name;
3265 struct lu_fid *pfid = &info->lti_fid;
3266 struct lu_fid *cfid = &info->lti_fid2;
3267 struct lfsck_instance *lfsck = com->lc_lfsck;
3268 struct lfsck_namespace *ns = com->lc_file_ram;
3269 struct dt_object *parent = NULL;
3270 struct linkea_data ldata = { NULL };
3271 bool repaired = false;
3276 dt_read_lock(env, child, 0);
3277 if (unlikely(lfsck_is_dead_obj(child))) {
3278 dt_read_unlock(env, child);
3283 if (S_ISDIR(lfsck_object_type(child))) {
3284 dt_read_unlock(env, child);
3285 rc = lfsck_namespace_double_scan_dir(env, com, child, flags);
3290 rc = lfsck_links_read(env, child, &ldata);
3291 dt_read_unlock(env, child);
3295 linkea_first_entry(&ldata);
3296 while (ldata.ld_lee != NULL) {
3297 lfsck_namespace_unpack_linkea_entry(&ldata, cname, pfid,
3299 rc = lfsck_namespace_filter_linkea_entry(&ldata, cname, pfid,
3301 /* Found repeated linkEA entries */
3303 rc = lfsck_namespace_shrink_linkea(env, com, child,
3304 &ldata, cname, pfid, false);
3316 /* Invalid PFID in the linkEA entry. */
3317 if (!fid_is_sane(pfid)) {
3318 rc = lfsck_namespace_shrink_linkea(env, com, child,
3319 &ldata, cname, pfid, true);
3329 parent = lfsck_object_find_bottom(env, lfsck, pfid);
3331 GOTO(out, rc = PTR_ERR(parent));
3333 if (!dt_object_exists(parent)) {
3336 if (ldata.ld_leh->leh_reccount > 1) {
3337 /* If it is NOT the last linkEA entry, then
3338 * there is still other chance to make the
3339 * child to be visible via other parent, then
3340 * remove this linkEA entry. */
3341 rc = lfsck_namespace_shrink_linkea(env, com,
3342 child, &ldata, cname, pfid, true);
3344 /* If the LFSCK is marked as LF_INCOMPLETE,
3345 * then means some MDT has ever tried to
3346 * verify some remote MDT-object that resides
3347 * on this MDT, but this MDT failed to respond
3348 * such request. So means there may be some
3349 * remote name entry on other MDT that
3350 * references this object with another name,
3351 * so we cannot know whether this linkEA is
3352 * valid or not. So keep it there and maybe
3353 * resolved when next LFSCK run. */
3354 if (ns->ln_flags & LF_INCOMPLETE) {
3355 lfsck_object_put(env, parent);
3360 /* Create the lost parent as an orphan. */
3361 rc = lfsck_namespace_create_orphan_dir(env, com,
3364 lfsck_object_put(env, parent);
3372 /* Add the missing name entry to the parent. */
3373 rc = lfsck_namespace_insert_normal(env, com,
3374 parent, child, cname->ln_name);
3375 if (unlikely(rc == -EEXIST))
3376 /* Unfortunately, someone reused the
3377 * name under the parent by race. So we
3378 * have to remove the linkEA entry from
3379 * current child object. It means that
3380 * the LFSCK cannot recover the system
3381 * totally back to its original status,
3382 * but it is necessary to make the
3383 * current system to be consistent. */
3384 rc = lfsck_namespace_shrink_linkea(env,
3388 linkea_next_entry(&ldata);
3391 lfsck_object_put(env, parent);
3401 /* The linkEA entry with bad parent will be removed. */
3402 if (unlikely(!dt_try_as_dir(env, parent))) {
3403 lfsck_object_put(env, parent);
3404 rc = lfsck_namespace_shrink_linkea(env, com, child,
3405 &ldata, cname, pfid, true);
3415 rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
3416 (const struct dt_key *)cname->ln_name);
3417 if (rc != 0 && rc != -ENOENT) {
3418 lfsck_object_put(env, parent);
3424 if (lu_fid_eq(cfid, lfsck_dto2fid(child))) {
3425 /* It is the most common case that we
3426 * find the name entry corresponding
3427 * to the linkEA entry. */
3428 lfsck_object_put(env, parent);
3429 linkea_next_entry(&ldata);
3431 /* The name entry references another
3432 * MDT-object that may be created by
3433 * the LFSCK for repairing dangling
3434 * name entry. Try to replace it. */
3435 rc = lfsck_namespace_replace_cond(env, com,
3436 parent, child, cfid, cname);
3437 lfsck_object_put(env, parent);
3443 linkea_next_entry(&ldata);
3445 rc = lfsck_namespace_shrink_linkea(env,
3459 rc = dt_attr_get(env, child, la);
3463 /* If there is no name entry in the parent dir and the object
3464 * link count is less than the linkea entries count, then the
3465 * linkea entry should be removed. */
3466 if (ldata.ld_leh->leh_reccount > la->la_nlink) {
3467 rc = lfsck_namespace_shrink_linkea_cond(env, com,
3468 parent, child, &ldata, cname, pfid);
3469 lfsck_object_put(env, parent);
3479 /* If the LFSCK is marked as LF_INCOMPLETE, then means some
3480 * MDT has ever tried to verify some remote MDT-object that
3481 * resides on this MDT, but this MDT failed to respond such
3482 * request. So means there may be some remote name entry on
3483 * other MDT that references this object with another name,
3484 * so we cannot know whether this linkEA is valid or not.
3485 * So keep it there and maybe resolved when next LFSCK run. */
3486 if (ns->ln_flags & LF_INCOMPLETE) {
3487 lfsck_object_put(env, parent);
3492 rc = lfsck_namespace_check_name(env, parent, child, cname);
3497 lfsck_object_put(env, parent);
3502 /* It is an invalid name entry, drop it. */
3503 if (unlikely(rc > 0)) {
3504 lfsck_object_put(env, parent);
3505 rc = lfsck_namespace_shrink_linkea(env, com, child,
3506 &ldata, cname, pfid, true);
3516 /* Add the missing name entry back to the namespace. */
3517 rc = lfsck_namespace_insert_normal(env, com, parent, child,
3519 if (unlikely(rc == -ESTALE))
3520 /* It may happen when the remote object has been
3521 * removed, but the local MDT is not aware of that. */
3524 if (unlikely(rc == -EEXIST))
3525 /* Unfortunately, someone reused the name under the
3526 * parent by race. So we have to remove the linkEA
3527 * entry from current child object. It means that the
3528 * LFSCK cannot recover the system totally back to
3529 * its original status, but it is necessary to make
3530 * the current system to be consistent.
3532 * It also may be because of the LFSCK found some
3533 * internal status of create operation. Under such
3534 * case, nothing to be done. */
3535 rc = lfsck_namespace_shrink_linkea_cond(env, com,
3536 parent, child, &ldata, cname, pfid);
3538 linkea_next_entry(&ldata);
3540 lfsck_object_put(env, parent);
3551 if (rc < 0 && rc != -ENODATA)
3555 LASSERT(ldata.ld_leh != NULL);
3557 count = ldata.ld_leh->leh_reccount;
3561 /* If the LFSCK is marked as LF_INCOMPLETE, then means some
3562 * MDT has ever tried to verify some remote MDT-object that
3563 * resides on this MDT, but this MDT failed to respond such
3564 * request. So means there may be some remote name entry on
3565 * other MDT that references this object with another name,
3566 * so we cannot know whether this linkEA is valid or not.
3567 * So keep it there and maybe resolved when next LFSCK run. */
3568 if (!(ns->ln_flags & LF_INCOMPLETE)) {
3569 /* If the child becomes orphan, then insert it into
3570 * the global .lustre/lost+found/MDTxxxx directory. */
3571 rc = lfsck_namespace_insert_orphan(env, com, child,
3577 ns->ln_mul_ref_repaired++;
3582 rc = dt_attr_get(env, child, la);
3586 if (la->la_nlink != 0 && la->la_nlink != count) {
3587 rc = lfsck_namespace_repair_nlink(env, com, child, la);
3589 ns->ln_objs_nlink_repaired++;
3596 if (la->la_nlink > 1)
3597 ns->ln_mul_linked_repaired++;
3606 static void lfsck_namespace_dump_statistics(struct seq_file *m,
3607 struct lfsck_namespace *ns,
3608 __u64 checked_phase1,
3609 __u64 checked_phase2,
3613 seq_printf(m, "checked_phase1: "LPU64"\n"
3614 "checked_phase2: "LPU64"\n"
3615 "updated_phase1: "LPU64"\n"
3616 "updated_phase2: "LPU64"\n"
3617 "failed_phase1: "LPU64"\n"
3618 "failed_phase2: "LPU64"\n"
3619 "directories: "LPU64"\n"
3620 "dirent_repaired: "LPU64"\n"
3621 "linkea_repaired: "LPU64"\n"
3622 "nlinks_repaired: "LPU64"\n"
3623 "multiple_linked_checked: "LPU64"\n"
3624 "multiple_linked_repaired: "LPU64"\n"
3625 "unknown_inconsistency: "LPU64"\n"
3626 "unmatched_pairs_repaired: "LPU64"\n"
3627 "dangling_repaired: "LPU64"\n"
3628 "multiple_referenced_repaired: "LPU64"\n"
3629 "bad_file_type_repaired: "LPU64"\n"
3630 "lost_dirent_repaired: "LPU64"\n"
3631 "local_lost_found_scanned: "LPU64"\n"
3632 "local_lost_found_moved: "LPU64"\n"
3633 "local_lost_found_skipped: "LPU64"\n"
3634 "local_lost_found_failed: "LPU64"\n"
3635 "striped_dirs_scanned: "LPU64"\n"
3636 "striped_dirs_repaired: "LPU64"\n"
3637 "striped_dirs_failed: "LPU64"\n"
3638 "striped_dirs_disabled: "LPU64"\n"
3639 "striped_dirs_skipped: "LPU64"\n"
3640 "striped_shards_scanned: "LPU64"\n"
3641 "striped_shards_repaired: "LPU64"\n"
3642 "striped_shards_failed: "LPU64"\n"
3643 "striped_shards_skipped: "LPU64"\n"
3644 "name_hash_repaired: "LPU64"\n"
3645 "success_count: %u\n"
3646 "run_time_phase1: %u seconds\n"
3647 "run_time_phase2: %u seconds\n",
3650 ns->ln_items_repaired,
3651 ns->ln_objs_repaired_phase2,
3652 ns->ln_items_failed,
3653 ns->ln_objs_failed_phase2,
3654 ns->ln_dirs_checked,
3655 ns->ln_dirent_repaired,
3656 ns->ln_linkea_repaired,
3657 ns->ln_objs_nlink_repaired,
3658 ns->ln_mul_linked_checked,
3659 ns->ln_mul_linked_repaired,
3660 ns->ln_unknown_inconsistency,
3661 ns->ln_unmatched_pairs_repaired,
3662 ns->ln_dangling_repaired,
3663 ns->ln_mul_ref_repaired,
3664 ns->ln_bad_type_repaired,
3665 ns->ln_lost_dirent_repaired,
3666 ns->ln_local_lpf_scanned,
3667 ns->ln_local_lpf_moved,
3668 ns->ln_local_lpf_skipped,
3669 ns->ln_local_lpf_failed,
3670 ns->ln_striped_dirs_scanned,
3671 ns->ln_striped_dirs_repaired,
3672 ns->ln_striped_dirs_failed,
3673 ns->ln_striped_dirs_disabled,
3674 ns->ln_striped_dirs_skipped,
3675 ns->ln_striped_shards_scanned,
3676 ns->ln_striped_shards_repaired,
3677 ns->ln_striped_shards_failed,
3678 ns->ln_striped_shards_skipped,
3679 ns->ln_name_hash_repaired,
3680 ns->ln_success_count,
3685 static void lfsck_namespace_release_lmv(const struct lu_env *env,
3686 struct lfsck_component *com)
3688 struct lfsck_instance *lfsck = com->lc_lfsck;
3689 struct lfsck_namespace *ns = com->lc_file_ram;
3691 while (!list_empty(&lfsck->li_list_lmv)) {
3692 struct lfsck_lmv_unit *llu;
3693 struct lfsck_lmv *llmv;
3695 llu = list_entry(lfsck->li_list_lmv.next,
3696 struct lfsck_lmv_unit, llu_link);
3697 llmv = &llu->llu_lmv;
3699 LASSERTF(atomic_read(&llmv->ll_ref) == 1,
3700 "still in using: %u\n",
3701 atomic_read(&llmv->ll_ref));
3703 ns->ln_striped_dirs_skipped++;
3704 lfsck_lmv_put(env, llmv);
3708 static int lfsck_namespace_check_for_double_scan(const struct lu_env *env,
3709 struct lfsck_component *com,
3710 struct dt_object *obj)
3712 struct lu_attr *la = &lfsck_env_info(env)->lti_la;
3715 rc = dt_attr_get(env, obj, la);
3719 /* zero-linkEA object may be orphan, but it also maybe because
3720 * of upgrading. Currently, we cannot record it for double scan.
3721 * Because it may cause the LFSCK trace file to be too large. */
3723 /* "la_ctime" == 1 means that it has ever been removed from
3724 * backend /lost+found directory but not been added back to
3725 * the normal namespace yet. */
3727 if ((S_ISREG(lfsck_object_type(obj)) && la->la_nlink > 1) ||
3728 unlikely(la->la_ctime == 1))
3729 rc = lfsck_namespace_trace_update(env, com, lfsck_dto2fid(obj),
3730 LNTF_CHECK_LINKEA, true);
3735 /* namespace APIs */
3737 static int lfsck_namespace_reset(const struct lu_env *env,
3738 struct lfsck_component *com, bool init)
3740 struct lfsck_instance *lfsck = com->lc_lfsck;
3741 struct lfsck_namespace *ns = com->lc_file_ram;
3742 struct lfsck_assistant_data *lad = com->lc_data;
3743 struct dt_object *root;
3744 struct dt_object *dto;
3748 root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
3750 GOTO(log, rc = PTR_ERR(root));
3752 if (unlikely(!dt_try_as_dir(env, root)))
3753 GOTO(put, rc = -ENOTDIR);
3755 down_write(&com->lc_sem);
3757 memset(ns, 0, sizeof(*ns));
3759 __u32 count = ns->ln_success_count;
3760 __u64 last_time = ns->ln_time_last_complete;
3762 memset(ns, 0, sizeof(*ns));
3763 ns->ln_success_count = count;
3764 ns->ln_time_last_complete = last_time;
3766 ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
3767 ns->ln_status = LS_INIT;
3769 lfsck_object_put(env, com->lc_obj);
3771 dto = lfsck_namespace_load_one_trace_file(env, com, root,
3772 LFSCK_NAMESPACE, true);
3774 GOTO(out, rc = PTR_ERR(dto));
3777 rc = lfsck_namespace_load_sub_trace_files(env, com, true);
3781 lad->lad_incomplete = 0;
3782 CFS_RESET_BITMAP(lad->lad_bitmap);
3784 rc = lfsck_namespace_store(env, com, true);
3789 up_write(&com->lc_sem);
3792 lfsck_object_put(env, root);
3794 CDEBUG(D_LFSCK, "%s: namespace LFSCK reset: rc = %d\n",
3795 lfsck_lfsck2name(lfsck), rc);
3800 lfsck_namespace_fail(const struct lu_env *env, struct lfsck_component *com,
3803 struct lfsck_namespace *ns = com->lc_file_ram;
3805 down_write(&com->lc_sem);
3807 com->lc_new_checked++;
3808 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
3809 up_write(&com->lc_sem);
3812 static void lfsck_namespace_close_dir(const struct lu_env *env,
3813 struct lfsck_component *com)
3815 struct lfsck_namespace *ns = com->lc_file_ram;
3816 struct lfsck_assistant_data *lad = com->lc_data;
3817 struct lfsck_assistant_object *lso = NULL;
3818 struct lfsck_instance *lfsck = com->lc_lfsck;
3819 struct lfsck_lmv *llmv = lfsck->li_lmv;
3820 struct lfsck_namespace_req *lnr;
3822 sizeof(*lnr) + LFSCK_TMPBUF_LEN;
3823 bool wakeup = false;
3829 OBD_ALLOC(lnr, size);
3831 ns->ln_striped_dirs_skipped++;
3836 lso = lfsck_assistant_object_init(env, lfsck_dto2fid(lfsck->li_obj_dir),
3837 NULL, lfsck->li_pos_current.lp_oit_cookie, true);
3839 OBD_FREE(lnr, size);
3840 ns->ln_striped_dirs_skipped++;
3845 /* Generate a dummy request to indicate that all shards' name entry
3846 * in this striped directory has been scanned for the first time. */
3847 INIT_LIST_HEAD(&lnr->lnr_lar.lar_list);
3848 lnr->lnr_lar.lar_parent = lso;
3849 lnr->lnr_lmv = lfsck_lmv_get(llmv);
3850 lnr->lnr_fid = *lfsck_dto2fid(lfsck->li_obj_dir);
3851 lnr->lnr_dir_cookie = MDS_DIR_END_OFF;
3852 lnr->lnr_size = size;
3854 spin_lock(&lad->lad_lock);
3855 if (lad->lad_assistant_status < 0) {
3856 spin_unlock(&lad->lad_lock);
3857 lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar);
3858 ns->ln_striped_dirs_skipped++;
3863 list_add_tail(&lnr->lnr_lar.lar_list, &lad->lad_req_list);
3864 if (lad->lad_prefetched == 0)
3867 lad->lad_prefetched++;
3868 spin_unlock(&lad->lad_lock);
3870 wake_up_all(&lad->lad_thread.t_ctl_waitq);
3875 static int lfsck_namespace_open_dir(const struct lu_env *env,
3876 struct lfsck_component *com)
3878 struct lfsck_instance *lfsck = com->lc_lfsck;
3879 struct lfsck_namespace *ns = com->lc_file_ram;
3880 struct lfsck_lmv *llmv = lfsck->li_lmv;
3887 if (llmv->ll_lmv_master) {
3888 struct lmv_mds_md_v1 *lmv = &llmv->ll_lmv;
3890 if (lmv->lmv_master_mdt_index != lfsck_dev_idx(lfsck)) {
3891 lmv->lmv_master_mdt_index =
3892 lfsck_dev_idx(lfsck);
3893 ns->ln_flags |= LF_INCONSISTENT;
3894 llmv->ll_lmv_updated = 1;
3897 rc = lfsck_namespace_verify_stripe_slave(env, com,
3898 lfsck->li_obj_dir, llmv);
3901 RETURN(rc > 0 ? 0 : rc);
3904 static int lfsck_namespace_checkpoint(const struct lu_env *env,
3905 struct lfsck_component *com, bool init)
3907 struct lfsck_instance *lfsck = com->lc_lfsck;
3908 struct lfsck_namespace *ns = com->lc_file_ram;
3912 rc = lfsck_checkpoint_generic(env, com);
3917 down_write(&com->lc_sem);
3919 ns->ln_pos_latest_start = lfsck->li_pos_checkpoint;
3921 ns->ln_pos_last_checkpoint = lfsck->li_pos_checkpoint;
3922 ns->ln_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
3923 HALF_SEC - lfsck->li_time_last_checkpoint);
3924 ns->ln_time_last_checkpoint = cfs_time_current_sec();
3925 ns->ln_items_checked += com->lc_new_checked;
3926 com->lc_new_checked = 0;
3929 rc = lfsck_namespace_store(env, com, false);
3930 up_write(&com->lc_sem);
3933 CDEBUG(D_LFSCK, "%s: namespace LFSCK checkpoint at the pos ["LPU64
3934 ", "DFID", "LPX64"]: rc = %d\n", lfsck_lfsck2name(lfsck),
3935 lfsck->li_pos_current.lp_oit_cookie,
3936 PFID(&lfsck->li_pos_current.lp_dir_parent),
3937 lfsck->li_pos_current.lp_dir_cookie, rc);
3939 return rc > 0 ? 0 : rc;
3942 static int lfsck_namespace_prep(const struct lu_env *env,
3943 struct lfsck_component *com,
3944 struct lfsck_start_param *lsp)
3946 struct lfsck_instance *lfsck = com->lc_lfsck;
3947 struct lfsck_namespace *ns = com->lc_file_ram;
3948 struct lfsck_position *pos = &com->lc_pos_start;
3951 rc = lfsck_namespace_load_bitmap(env, com);
3952 if (rc != 0 || ns->ln_status == LS_COMPLETED) {
3953 rc = lfsck_namespace_reset(env, com, false);
3955 rc = lfsck_set_param(env, lfsck, lsp->lsp_start, true);
3958 CDEBUG(D_LFSCK, "%s: namespace LFSCK prep failed: "
3959 "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
3965 down_write(&com->lc_sem);
3966 ns->ln_time_latest_start = cfs_time_current_sec();
3967 spin_lock(&lfsck->li_lock);
3969 if (ns->ln_flags & LF_SCANNED_ONCE) {
3970 if (!lfsck->li_drop_dryrun ||
3971 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent)) {
3972 ns->ln_status = LS_SCANNING_PHASE2;
3973 list_move_tail(&com->lc_link,
3974 &lfsck->li_list_double_scan);
3975 if (!list_empty(&com->lc_link_dir))
3976 list_del_init(&com->lc_link_dir);
3977 lfsck_pos_set_zero(pos);
3979 ns->ln_status = LS_SCANNING_PHASE1;
3980 ns->ln_run_time_phase1 = 0;
3981 ns->ln_run_time_phase2 = 0;
3982 ns->ln_items_checked = 0;
3983 ns->ln_items_repaired = 0;
3984 ns->ln_items_failed = 0;
3985 ns->ln_dirs_checked = 0;
3986 ns->ln_objs_checked_phase2 = 0;
3987 ns->ln_objs_repaired_phase2 = 0;
3988 ns->ln_objs_failed_phase2 = 0;
3989 ns->ln_objs_nlink_repaired = 0;
3990 ns->ln_dirent_repaired = 0;
3991 ns->ln_linkea_repaired = 0;
3992 ns->ln_mul_linked_checked = 0;
3993 ns->ln_mul_linked_repaired = 0;
3994 ns->ln_unknown_inconsistency = 0;
3995 ns->ln_unmatched_pairs_repaired = 0;
3996 ns->ln_dangling_repaired = 0;
3997 ns->ln_mul_ref_repaired = 0;
3998 ns->ln_bad_type_repaired = 0;
3999 ns->ln_lost_dirent_repaired = 0;
4000 ns->ln_striped_dirs_scanned = 0;
4001 ns->ln_striped_dirs_repaired = 0;
4002 ns->ln_striped_dirs_failed = 0;
4003 ns->ln_striped_dirs_disabled = 0;
4004 ns->ln_striped_dirs_skipped = 0;
4005 ns->ln_striped_shards_scanned = 0;
4006 ns->ln_striped_shards_repaired = 0;
4007 ns->ln_striped_shards_failed = 0;
4008 ns->ln_striped_shards_skipped = 0;
4009 ns->ln_name_hash_repaired = 0;
4010 fid_zero(&ns->ln_fid_latest_scanned_phase2);
4011 if (list_empty(&com->lc_link_dir))
4012 list_add_tail(&com->lc_link_dir,
4013 &lfsck->li_list_dir);
4014 *pos = ns->ln_pos_first_inconsistent;
4017 ns->ln_status = LS_SCANNING_PHASE1;
4018 if (list_empty(&com->lc_link_dir))
4019 list_add_tail(&com->lc_link_dir,
4020 &lfsck->li_list_dir);
4021 if (!lfsck->li_drop_dryrun ||
4022 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent)) {
4023 *pos = ns->ln_pos_last_checkpoint;
4024 pos->lp_oit_cookie++;
4026 *pos = ns->ln_pos_first_inconsistent;
4030 spin_unlock(&lfsck->li_lock);
4031 up_write(&com->lc_sem);
4033 rc = lfsck_start_assistant(env, com, lsp);
4035 CDEBUG(D_LFSCK, "%s: namespace LFSCK prep done, start pos ["LPU64", "
4036 DFID", "LPX64"]: rc = %d\n",
4037 lfsck_lfsck2name(lfsck), pos->lp_oit_cookie,
4038 PFID(&pos->lp_dir_parent), pos->lp_dir_cookie, rc);
4043 static int lfsck_namespace_exec_oit(const struct lu_env *env,
4044 struct lfsck_component *com,
4045 struct dt_object *obj)
4047 struct lfsck_thread_info *info = lfsck_env_info(env);
4048 struct lfsck_namespace *ns = com->lc_file_ram;
4049 struct lfsck_instance *lfsck = com->lc_lfsck;
4050 const struct lu_fid *fid = lfsck_dto2fid(obj);
4051 struct lu_fid *pfid = &info->lti_fid2;
4052 struct lu_name *cname = &info->lti_name;
4053 struct lu_seq_range *range = &info->lti_range;
4054 struct seq_server_site *ss = lfsck_dev_site(lfsck);
4055 struct linkea_data ldata = { NULL };
4056 __u32 idx = lfsck_dev_idx(lfsck);
4060 rc = lfsck_links_read(env, obj, &ldata);
4064 /* -EINVAL means crashed linkEA, should be verified. */
4065 if (rc == -EINVAL) {
4066 rc = lfsck_namespace_trace_update(env, com, fid,
4067 LNTF_CHECK_LINKEA, true);
4069 struct lustre_handle lh = { 0 };
4071 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
4072 MDS_INODELOCK_UPDATE |
4073 MDS_INODELOCK_XATTR, LCK_EX);
4075 rc = lfsck_namespace_links_remove(env, com,
4077 lfsck_ibits_unlock(&lh, LCK_EX);
4081 GOTO(out, rc = (rc == -ENOENT ? 0 : rc));
4084 if (rc == -ENODATA) {
4085 rc = lfsck_namespace_check_for_double_scan(env, com, obj);
4093 /* Record multiple-linked object. */
4094 if (ldata.ld_leh->leh_reccount > 1) {
4095 rc = lfsck_namespace_trace_update(env, com, fid,
4096 LNTF_CHECK_LINKEA, true);
4101 linkea_first_entry(&ldata);
4102 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, cname, pfid);
4103 if (!fid_is_sane(pfid)) {
4104 rc = lfsck_namespace_trace_update(env, com, fid,
4105 LNTF_CHECK_PARENT, true);
4107 fld_range_set_mdt(range);
4108 rc = fld_local_lookup(env, ss->ss_server_fld,
4109 fid_seq(pfid), range);
4110 if ((rc == -ENOENT) ||
4111 (rc == 0 && range->lsr_index != idx))
4112 rc = lfsck_namespace_trace_update(env, com, fid,
4113 LNTF_CHECK_LINKEA, true);
4115 rc = lfsck_namespace_check_for_double_scan(env, com,
4122 down_write(&com->lc_sem);
4123 if (S_ISDIR(lfsck_object_type(obj)))
4124 ns->ln_dirs_checked++;
4126 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
4127 up_write(&com->lc_sem);
4132 static int lfsck_namespace_exec_dir(const struct lu_env *env,
4133 struct lfsck_component *com,
4134 struct lfsck_assistant_object *lso,
4135 struct lu_dirent *ent, __u16 type)
4137 struct lfsck_assistant_data *lad = com->lc_data;
4138 struct lfsck_instance *lfsck = com->lc_lfsck;
4139 struct lfsck_namespace_req *lnr;
4140 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4141 struct ptlrpc_thread *mthread = &lfsck->li_thread;
4142 struct ptlrpc_thread *athread = &lad->lad_thread;
4143 struct l_wait_info lwi = { 0 };
4144 bool wakeup = false;
4146 l_wait_event(mthread->t_ctl_waitq,
4147 lad->lad_prefetched < bk->lb_async_windows ||
4148 !thread_is_running(mthread) ||
4149 thread_is_stopped(athread),
4152 if (unlikely(!thread_is_running(mthread)) ||
4153 thread_is_stopped(athread))
4156 if (unlikely(lfsck_is_dead_obj(lfsck->li_obj_dir)))
4159 lnr = lfsck_namespace_assistant_req_init(com->lc_lfsck, lso, ent, type);
4161 struct lfsck_namespace *ns = com->lc_file_ram;
4163 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
4164 return PTR_ERR(lnr);
4167 spin_lock(&lad->lad_lock);
4168 if (lad->lad_assistant_status < 0) {
4169 spin_unlock(&lad->lad_lock);
4170 lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar);
4171 return lad->lad_assistant_status;
4174 list_add_tail(&lnr->lnr_lar.lar_list, &lad->lad_req_list);
4175 if (lad->lad_prefetched == 0)
4178 lad->lad_prefetched++;
4179 spin_unlock(&lad->lad_lock);
4181 wake_up_all(&lad->lad_thread.t_ctl_waitq);
4183 down_write(&com->lc_sem);
4184 com->lc_new_checked++;
4185 up_write(&com->lc_sem);
4190 static int lfsck_namespace_post(const struct lu_env *env,
4191 struct lfsck_component *com,
4192 int result, bool init)
4194 struct lfsck_instance *lfsck = com->lc_lfsck;
4195 struct lfsck_namespace *ns = com->lc_file_ram;
4199 lfsck_post_generic(env, com, &result);
4201 down_write(&com->lc_sem);
4202 lfsck_namespace_release_lmv(env, com);
4204 spin_lock(&lfsck->li_lock);
4206 ns->ln_pos_last_checkpoint = lfsck->li_pos_checkpoint;
4208 ns->ln_status = LS_SCANNING_PHASE2;
4209 ns->ln_flags |= LF_SCANNED_ONCE;
4210 ns->ln_flags &= ~LF_UPGRADE;
4211 list_del_init(&com->lc_link_dir);
4212 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
4213 } else if (result == 0) {
4214 if (lfsck->li_status != 0)
4215 ns->ln_status = lfsck->li_status;
4217 ns->ln_status = LS_STOPPED;
4218 if (ns->ln_status != LS_PAUSED) {
4219 list_del_init(&com->lc_link_dir);
4220 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4223 ns->ln_status = LS_FAILED;
4224 list_del_init(&com->lc_link_dir);
4225 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4227 spin_unlock(&lfsck->li_lock);
4230 ns->ln_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
4231 HALF_SEC - lfsck->li_time_last_checkpoint);
4232 ns->ln_time_last_checkpoint = cfs_time_current_sec();
4233 ns->ln_items_checked += com->lc_new_checked;
4234 com->lc_new_checked = 0;
4237 rc = lfsck_namespace_store(env, com, false);
4238 up_write(&com->lc_sem);
4240 CDEBUG(D_LFSCK, "%s: namespace LFSCK post done: rc = %d\n",
4241 lfsck_lfsck2name(lfsck), rc);
4247 lfsck_namespace_dump(const struct lu_env *env, struct lfsck_component *com,
4250 struct lfsck_instance *lfsck = com->lc_lfsck;
4251 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4252 struct lfsck_namespace *ns = com->lc_file_ram;
4255 down_read(&com->lc_sem);
4256 seq_printf(m, "name: lfsck_namespace\n"
4262 lfsck_status2names(ns->ln_status));
4264 rc = lfsck_bits_dump(m, ns->ln_flags, lfsck_flags_names, "flags");
4268 rc = lfsck_bits_dump(m, bk->lb_param, lfsck_param_names, "param");
4272 rc = lfsck_time_dump(m, ns->ln_time_last_complete,
4277 rc = lfsck_time_dump(m, ns->ln_time_latest_start,
4282 rc = lfsck_time_dump(m, ns->ln_time_last_checkpoint,
4287 rc = lfsck_pos_dump(m, &ns->ln_pos_latest_start,
4288 "latest_start_position");
4292 rc = lfsck_pos_dump(m, &ns->ln_pos_last_checkpoint,
4293 "last_checkpoint_position");
4297 rc = lfsck_pos_dump(m, &ns->ln_pos_first_inconsistent,
4298 "first_failure_position");
4302 if (ns->ln_status == LS_SCANNING_PHASE1) {
4303 struct lfsck_position pos;
4304 const struct dt_it_ops *iops;
4305 cfs_duration_t duration = cfs_time_current() -
4306 lfsck->li_time_last_checkpoint;
4307 __u64 checked = ns->ln_items_checked + com->lc_new_checked;
4308 __u64 speed = checked;
4309 __u64 new_checked = com->lc_new_checked *
4310 msecs_to_jiffies(MSEC_PER_SEC);
4311 __u32 rtime = ns->ln_run_time_phase1 +
4312 cfs_duration_sec(duration + HALF_SEC);
4315 do_div(new_checked, duration);
4318 do_div(speed, rtime);
4320 lfsck_namespace_dump_statistics(m, ns, checked, 0, rtime, 0);
4321 seq_printf(m, "average_speed_phase1: "LPU64" items/sec\n"
4322 "average_speed_phase2: N/A\n"
4323 "average_speed_total: "LPU64" items/sec\n"
4324 "real_time_speed_phase1: "LPU64" items/sec\n"
4325 "real_time_speed_phase2: N/A\n",
4330 LASSERT(lfsck->li_di_oit != NULL);
4332 iops = &lfsck->li_obj_oit->do_index_ops->dio_it;
4334 /* The low layer otable-based iteration position may NOT
4335 * exactly match the namespace-based directory traversal
4336 * cookie. Generally, it is not a serious issue. But the
4337 * caller should NOT make assumption on that. */
4338 pos.lp_oit_cookie = iops->store(env, lfsck->li_di_oit);
4339 if (!lfsck->li_current_oit_processed)
4340 pos.lp_oit_cookie--;
4342 spin_lock(&lfsck->li_lock);
4343 if (lfsck->li_di_dir != NULL) {
4344 pos.lp_dir_cookie = lfsck->li_cookie_dir;
4345 if (pos.lp_dir_cookie >= MDS_DIR_END_OFF) {
4346 fid_zero(&pos.lp_dir_parent);
4347 pos.lp_dir_cookie = 0;
4350 *lfsck_dto2fid(lfsck->li_obj_dir);
4353 fid_zero(&pos.lp_dir_parent);
4354 pos.lp_dir_cookie = 0;
4356 spin_unlock(&lfsck->li_lock);
4357 lfsck_pos_dump(m, &pos, "current_position");
4358 } else if (ns->ln_status == LS_SCANNING_PHASE2) {
4359 cfs_duration_t duration = cfs_time_current() -
4360 com->lc_time_last_checkpoint;
4361 __u64 checked = ns->ln_objs_checked_phase2 +
4362 com->lc_new_checked;
4363 __u64 speed1 = ns->ln_items_checked;
4364 __u64 speed2 = checked;
4365 __u64 speed0 = speed1 + speed2;
4366 __u64 new_checked = com->lc_new_checked *
4367 msecs_to_jiffies(MSEC_PER_SEC);
4368 __u32 rtime = ns->ln_run_time_phase2 +
4369 cfs_duration_sec(duration + HALF_SEC);
4370 __u32 time0 = ns->ln_run_time_phase1 + rtime;
4373 do_div(new_checked, duration);
4375 if (ns->ln_run_time_phase1 != 0)
4376 do_div(speed1, ns->ln_run_time_phase1);
4377 else if (ns->ln_items_checked != 0)
4381 do_div(speed2, rtime);
4382 else if (checked != 0)
4386 do_div(speed0, time0);
4388 lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked,
4390 ns->ln_run_time_phase1, rtime);
4391 seq_printf(m, "average_speed_phase1: "LPU64" items/sec\n"
4392 "average_speed_phase2: "LPU64" objs/sec\n"
4393 "average_speed_total: "LPU64" items/sec\n"
4394 "real_time_speed_phase1: N/A\n"
4395 "real_time_speed_phase2: "LPU64" objs/sec\n"
4396 "current_position: "DFID"\n",
4401 PFID(&ns->ln_fid_latest_scanned_phase2));
4403 __u64 speed1 = ns->ln_items_checked;
4404 __u64 speed2 = ns->ln_objs_checked_phase2;
4405 __u64 speed0 = speed1 + speed2;
4406 __u32 time0 = ns->ln_run_time_phase1 + ns->ln_run_time_phase2;
4408 if (ns->ln_run_time_phase1 != 0)
4409 do_div(speed1, ns->ln_run_time_phase1);
4410 else if (ns->ln_items_checked != 0)
4413 if (ns->ln_run_time_phase2 != 0)
4414 do_div(speed2, ns->ln_run_time_phase2);
4415 else if (ns->ln_objs_checked_phase2 != 0)
4419 do_div(speed0, time0);
4421 lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked,
4422 ns->ln_objs_checked_phase2,
4423 ns->ln_run_time_phase1,
4424 ns->ln_run_time_phase2);
4425 seq_printf(m, "average_speed_phase1: "LPU64" items/sec\n"
4426 "average_speed_phase2: "LPU64" objs/sec\n"
4427 "average_speed_total: "LPU64" items/sec\n"
4428 "real_time_speed_phase1: N/A\n"
4429 "real_time_speed_phase2: N/A\n"
4430 "current_position: N/A\n",
4436 up_read(&com->lc_sem);
4440 static int lfsck_namespace_double_scan(const struct lu_env *env,
4441 struct lfsck_component *com)
4443 struct lfsck_namespace *ns = com->lc_file_ram;
4444 struct lfsck_assistant_data *lad = com->lc_data;
4445 struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
4446 struct lfsck_tgt_desc *ltd;
4447 struct lfsck_tgt_desc *next;
4450 rc = lfsck_double_scan_generic(env, com, ns->ln_status);
4451 if (thread_is_stopped(&lad->lad_thread)) {
4452 LASSERT(list_empty(&lad->lad_req_list));
4453 LASSERT(list_empty(&lad->lad_mdt_phase1_list));
4455 spin_lock(<ds->ltd_lock);
4456 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4457 ltd_namespace_phase_list) {
4458 list_del_init(<d->ltd_namespace_phase_list);
4460 spin_unlock(<ds->ltd_lock);
4466 static void lfsck_namespace_data_release(const struct lu_env *env,
4467 struct lfsck_component *com)
4469 struct lfsck_assistant_data *lad = com->lc_data;
4470 struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
4471 struct lfsck_tgt_desc *ltd;
4472 struct lfsck_tgt_desc *next;
4474 LASSERT(lad != NULL);
4475 LASSERT(thread_is_init(&lad->lad_thread) ||
4476 thread_is_stopped(&lad->lad_thread));
4477 LASSERT(list_empty(&lad->lad_req_list));
4479 com->lc_data = NULL;
4480 lfsck_namespace_release_lmv(env, com);
4482 spin_lock(<ds->ltd_lock);
4483 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
4484 ltd_namespace_phase_list) {
4485 list_del_init(<d->ltd_namespace_phase_list);
4487 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4488 ltd_namespace_phase_list) {
4489 list_del_init(<d->ltd_namespace_phase_list);
4491 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_list,
4492 ltd_namespace_list) {
4493 list_del_init(<d->ltd_namespace_list);
4495 spin_unlock(<ds->ltd_lock);
4497 if (likely(lad->lad_bitmap != NULL))
4498 CFS_FREE_BITMAP(lad->lad_bitmap);
4503 static void lfsck_namespace_quit(const struct lu_env *env,
4504 struct lfsck_component *com)
4506 struct lfsck_assistant_data *lad = com->lc_data;
4507 struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
4508 struct lfsck_tgt_desc *ltd;
4509 struct lfsck_tgt_desc *next;
4511 LASSERT(lad != NULL);
4513 lfsck_quit_generic(env, com);
4515 LASSERT(thread_is_init(&lad->lad_thread) ||
4516 thread_is_stopped(&lad->lad_thread));
4517 LASSERT(list_empty(&lad->lad_req_list));
4519 lfsck_namespace_release_lmv(env, com);
4521 spin_lock(<ds->ltd_lock);
4522 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
4523 ltd_namespace_phase_list) {
4524 list_del_init(<d->ltd_namespace_phase_list);
4526 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4527 ltd_namespace_phase_list) {
4528 list_del_init(<d->ltd_namespace_phase_list);
4530 spin_unlock(<ds->ltd_lock);
4533 static int lfsck_namespace_in_notify(const struct lu_env *env,
4534 struct lfsck_component *com,
4535 struct lfsck_request *lr,
4538 struct lfsck_instance *lfsck = com->lc_lfsck;
4539 struct lfsck_namespace *ns = com->lc_file_ram;
4540 struct lfsck_assistant_data *lad = com->lc_data;
4541 struct lfsck_tgt_descs *ltds = &lfsck->li_mdt_descs;
4542 struct lfsck_tgt_desc *ltd;
4547 switch (lr->lr_event) {
4548 case LE_SKIP_NLINK_DECLARE: {
4549 struct dt_object *obj;
4550 struct lu_fid *key = &lfsck_env_info(env)->lti_fid3;
4554 LASSERT(th != NULL);
4556 idx = lfsck_sub_trace_file_fid2idx(&lr->lr_fid);
4557 obj = com->lc_sub_trace_objs[idx].lsto_obj;
4558 fid_cpu_to_be(key, &lr->lr_fid);
4559 mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
4560 rc = dt_declare_delete(env, obj,
4561 (const struct dt_key *)key, th);
4563 rc = dt_declare_insert(env, obj,
4564 (const struct dt_rec *)&flags,
4565 (const struct dt_key *)key, th);
4566 mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
4570 case LE_SKIP_NLINK: {
4571 struct dt_object *obj;
4572 struct lu_fid *key = &lfsck_env_info(env)->lti_fid3;
4578 LASSERT(th != NULL);
4580 idx = lfsck_sub_trace_file_fid2idx(&lr->lr_fid);
4581 obj = com->lc_sub_trace_objs[idx].lsto_obj;
4582 fid_cpu_to_be(key, &lr->lr_fid);
4583 mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
4584 rc = dt_lookup(env, obj, (struct dt_rec *)&flags,
4585 (const struct dt_key *)key);
4587 if (flags & LNTF_SKIP_NLINK) {
4589 &com->lc_sub_trace_objs[idx].lsto_mutex);
4595 } else if (rc != -ENOENT) {
4599 flags |= LNTF_SKIP_NLINK;
4601 rc = dt_delete(env, obj, (const struct dt_key *)key,
4607 rc = dt_insert(env, obj, (const struct dt_rec *)&flags,
4608 (const struct dt_key *)key, th, 1);
4613 mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
4614 CDEBUG(D_LFSCK, "%s: RPC service thread mark the "DFID
4615 " to be skipped for namespace double scan: rc = %d\n",
4616 lfsck_lfsck2name(com->lc_lfsck), PFID(&lr->lr_fid), rc);
4619 /* If we cannot record this object in the LFSCK tracing,
4620 * we have to mark the LFSC as LF_INCOMPLETE, then the
4621 * LFSCK will skip nlink attribute verification for
4623 ns->ln_flags |= LF_INCOMPLETE;
4627 case LE_SET_LMV_MASTER: {
4628 struct dt_object *obj;
4630 obj = lfsck_object_find_bottom(env, lfsck, &lr->lr_fid);
4632 RETURN(PTR_ERR(obj));
4634 if (likely(dt_object_exists(obj)))
4635 rc = lfsck_namespace_notify_lmv_master_local(env, com,
4638 lfsck_object_put(env, obj);
4640 RETURN(rc > 0 ? 0 : rc);
4642 case LE_SET_LMV_SLAVE: {
4643 if (!(lr->lr_flags & LEF_RECHECK_NAME_HASH))
4644 ns->ln_striped_shards_repaired++;
4646 rc = lfsck_namespace_trace_update(env, com, &lr->lr_fid,
4647 LNTF_RECHECK_NAME_HASH, true);
4649 RETURN(rc > 0 ? 0 : rc);
4651 case LE_PHASE1_DONE:
4652 case LE_PHASE2_DONE:
4659 CDEBUG(D_LFSCK, "%s: namespace LFSCK handles notify %u from MDT %x, "
4660 "status %d, flags %x\n", lfsck_lfsck2name(lfsck), lr->lr_event,
4661 lr->lr_index, lr->lr_status, lr->lr_flags2);
4663 spin_lock(<ds->ltd_lock);
4664 ltd = lfsck_ltd2tgt(ltds, lr->lr_index);
4666 spin_unlock(<ds->ltd_lock);
4671 list_del_init(<d->ltd_namespace_phase_list);
4672 switch (lr->lr_event) {
4673 case LE_PHASE1_DONE:
4674 if (lr->lr_status <= 0) {
4675 ltd->ltd_namespace_done = 1;
4676 list_del_init(<d->ltd_namespace_list);
4677 CDEBUG(D_LFSCK, "%s: MDT %x failed/stopped at "
4678 "phase1 for namespace LFSCK: rc = %d.\n",
4679 lfsck_lfsck2name(lfsck),
4680 ltd->ltd_index, lr->lr_status);
4681 ns->ln_flags |= LF_INCOMPLETE;
4686 if (lr->lr_flags2 & LF_INCOMPLETE)
4687 ns->ln_flags |= LF_INCOMPLETE;
4689 if (list_empty(<d->ltd_namespace_list))
4690 list_add_tail(<d->ltd_namespace_list,
4691 &lad->lad_mdt_list);
4692 list_add_tail(<d->ltd_namespace_phase_list,
4693 &lad->lad_mdt_phase2_list);
4695 case LE_PHASE2_DONE:
4696 ltd->ltd_namespace_done = 1;
4697 list_del_init(<d->ltd_namespace_list);
4701 ltd->ltd_namespace_done = 1;
4702 list_del_init(<d->ltd_namespace_list);
4703 if (!(lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT)) {
4705 "%s: the peer MDT %x exit namespace LFSCK\n",
4706 lfsck_lfsck2name(lfsck), ltd->ltd_index);
4707 ns->ln_flags |= LF_INCOMPLETE;
4713 spin_unlock(<ds->ltd_lock);
4715 if (fail && lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) {
4716 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
4718 memset(stop, 0, sizeof(*stop));
4719 stop->ls_status = lr->lr_status;
4720 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
4721 lfsck_stop(env, lfsck->li_bottom, stop);
4722 } else if (lfsck_phase2_next_ready(lad)) {
4723 wake_up_all(&lad->lad_thread.t_ctl_waitq);
4729 static int lfsck_namespace_query(const struct lu_env *env,
4730 struct lfsck_component *com)
4732 struct lfsck_namespace *ns = com->lc_file_ram;
4734 return ns->ln_status;
4737 static struct lfsck_operations lfsck_namespace_ops = {
4738 .lfsck_reset = lfsck_namespace_reset,
4739 .lfsck_fail = lfsck_namespace_fail,
4740 .lfsck_close_dir = lfsck_namespace_close_dir,
4741 .lfsck_open_dir = lfsck_namespace_open_dir,
4742 .lfsck_checkpoint = lfsck_namespace_checkpoint,
4743 .lfsck_prep = lfsck_namespace_prep,
4744 .lfsck_exec_oit = lfsck_namespace_exec_oit,
4745 .lfsck_exec_dir = lfsck_namespace_exec_dir,
4746 .lfsck_post = lfsck_namespace_post,
4747 .lfsck_dump = lfsck_namespace_dump,
4748 .lfsck_double_scan = lfsck_namespace_double_scan,
4749 .lfsck_data_release = lfsck_namespace_data_release,
4750 .lfsck_quit = lfsck_namespace_quit,
4751 .lfsck_in_notify = lfsck_namespace_in_notify,
4752 .lfsck_query = lfsck_namespace_query,
4756 * Repair dangling name entry.
4758 * For the name entry with dangling reference, we need to repare the
4759 * inconsistency according to the LFSCK sponsor's requirement:
4761 * 1) Keep the inconsistency there and report the inconsistency case,
4762 * then give the chance to the application to find related issues,
4763 * and the users can make the decision about how to handle it with
4764 * more human knownledge. (by default)
4766 * 2) Re-create the missing MDT-object with the FID information.
4768 * \param[in] env pointer to the thread context
4769 * \param[in] com pointer to the lfsck component
4770 * \param[in] parent pointer to the dir object that contains the dangling
4772 * \param[in] child pointer to the object corresponding to the dangling
4774 * \param[in] lnr pointer to the namespace request that contains the
4775 * name's name, parent object, parent's LMV, and ect.
4777 * \retval positive number if no need to repair
4778 * \retval zero for repaired successfully
4779 * \retval negative error number on failure
4781 int lfsck_namespace_repair_dangling(const struct lu_env *env,
4782 struct lfsck_component *com,
4783 struct dt_object *parent,
4784 struct dt_object *child,
4785 struct lfsck_namespace_req *lnr)
4787 struct lfsck_thread_info *info = lfsck_env_info(env);
4788 struct lu_attr *la = &info->lti_la;
4789 struct dt_allocation_hint *hint = &info->lti_hint;
4790 struct dt_object_format *dof = &info->lti_dof;
4791 struct dt_insert_rec *rec = &info->lti_dt_rec;
4792 struct lmv_mds_md_v1 *lmv2 = &info->lti_lmv2;
4793 const struct lu_name *cname;
4794 const struct lu_fid *pfid = lfsck_dto2fid(parent);
4795 const struct lu_fid *cfid = lfsck_dto2fid(child);
4796 struct linkea_data ldata = { NULL };
4797 struct lfsck_lock_handle *llh = &info->lti_llh;
4798 struct lu_buf linkea_buf;
4799 struct lu_buf lmv_buf;
4800 struct lfsck_instance *lfsck = com->lc_lfsck;
4801 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4802 struct dt_device *dev = lfsck->li_next;
4803 struct thandle *th = NULL;
4805 __u16 type = lnr->lnr_type;
4809 cname = lfsck_name_get_const(env, lnr->lnr_name, lnr->lnr_namelen);
4810 if (bk->lb_param & LPF_CREATE_MDTOBJ)
4815 if (!create || bk->lb_param & LPF_DRYRUN)
4818 /* We may need to create the sub-objects of the @child via LOD,
4819 * so make the modification based on lfsck->li_next. */
4821 parent = lfsck_object_locate(dev, parent);
4823 GOTO(log, rc = PTR_ERR(parent));
4825 if (unlikely(!dt_try_as_dir(env, parent)))
4826 GOTO(log, rc = -ENOTDIR);
4828 child = lfsck_object_locate(dev, child);
4830 GOTO(log, rc = PTR_ERR(child));
4832 rc = linkea_data_new(&ldata, &info->lti_linkea_buf2);
4836 rc = linkea_add_buf(&ldata, cname, pfid);
4840 rc = lfsck_lock(env, lfsck, parent, lnr->lnr_name, llh,
4841 MDS_INODELOCK_UPDATE, LCK_PR);
4845 rc = lfsck_namespace_check_exist(env, parent, child, lnr->lnr_name);
4849 /* Set the ctime as zero, then others can know it is created for
4850 * repairing dangling name entry by LFSCK. And if the LFSCK made
4851 * wrong decision and the real MDT-object has been found later,
4852 * then the LFSCK has chance to fix the incosistency properly. */
4853 memset(la, 0, sizeof(*la));
4854 la->la_mode = (type & S_IFMT) | 0600;
4855 la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
4856 LA_ATIME | LA_MTIME | LA_CTIME;
4858 child->do_ops->do_ah_init(env, hint, parent, child,
4859 la->la_mode & S_IFMT);
4861 memset(dof, 0, sizeof(*dof));
4862 dof->dof_type = dt_mode_to_dft(type);
4863 /* If the target is a regular file, then the LFSCK will only create
4864 * the MDT-object without stripes (dof->dof_reg.striped = 0). related
4865 * OST-objects will be created when write open. */
4867 th = dt_trans_create(env, dev);
4869 GOTO(log, rc = PTR_ERR(th));
4871 /* 1a. create child. */
4872 rc = dt_declare_create(env, child, la, hint, dof, th);
4876 if (S_ISDIR(type)) {
4877 if (unlikely(!dt_try_as_dir(env, child)))
4878 GOTO(stop, rc = -ENOTDIR);
4880 /* 2a. increase child nlink */
4881 rc = dt_declare_ref_add(env, child, th);
4885 /* 3a. insert dot into child dir */
4886 rec->rec_type = S_IFDIR;
4887 rec->rec_fid = cfid;
4888 rc = dt_declare_insert(env, child,
4889 (const struct dt_rec *)rec,
4890 (const struct dt_key *)dot, th);
4894 /* 4a. insert dotdot into child dir */
4895 rec->rec_fid = pfid;
4896 rc = dt_declare_insert(env, child,
4897 (const struct dt_rec *)rec,
4898 (const struct dt_key *)dotdot, th);
4902 /* 5a. generate slave LMV EA. */
4903 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
4906 idx = lfsck_shard_name_to_index(env,
4907 lnr->lnr_name, lnr->lnr_namelen,
4909 if (unlikely(idx < 0))
4910 GOTO(stop, rc = idx);
4912 *lmv2 = lnr->lnr_lmv->ll_lmv;
4913 lmv2->lmv_magic = LMV_MAGIC_STRIPE;
4914 lmv2->lmv_master_mdt_index = idx;
4916 lfsck_lmv_header_cpu_to_le(lmv2, lmv2);
4917 lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
4918 rc = dt_declare_xattr_set(env, child, &lmv_buf,
4919 XATTR_NAME_LMV, 0, th);
4925 /* 6a. insert linkEA for child */
4926 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
4927 ldata.ld_leh->leh_len);
4928 rc = dt_declare_xattr_set(env, child, &linkea_buf,
4929 XATTR_NAME_LINK, 0, th);
4933 rc = dt_trans_start_local(env, dev, th);
4935 GOTO(stop, rc = (rc == -EEXIST ? 1 : rc));
4937 dt_write_lock(env, child, 0);
4938 /* 1b. create child */
4939 rc = dt_create(env, child, la, hint, dof, th);
4941 GOTO(unlock, rc = (rc == -EEXIST ? 1 : rc));
4943 if (S_ISDIR(type)) {
4944 /* 2b. increase child nlink */
4945 rc = dt_ref_add(env, child, th);
4949 /* 3b. insert dot into child dir */
4950 rec->rec_type = S_IFDIR;
4951 rec->rec_fid = cfid;
4952 rc = dt_insert(env, child, (const struct dt_rec *)rec,
4953 (const struct dt_key *)dot, th, 1);
4957 /* 4b. insert dotdot into child dir */
4958 rec->rec_fid = pfid;
4959 rc = dt_insert(env, child, (const struct dt_rec *)rec,
4960 (const struct dt_key *)dotdot, th, 1);
4964 /* 5b. generate slave LMV EA. */
4965 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
4966 rc = dt_xattr_set(env, child, &lmv_buf, XATTR_NAME_LMV,
4973 /* 6b. insert linkEA for child. */
4974 rc = dt_xattr_set(env, child, &linkea_buf,
4975 XATTR_NAME_LINK, 0, th);
4980 dt_write_unlock(env, child);
4983 dt_trans_stop(env, dev, th);
4987 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found dangling "
4988 "reference for: parent "DFID", child "DFID", type %u, "
4989 "name %s. %s: rc = %d\n", lfsck_lfsck2name(lfsck),
4990 PFID(pfid), PFID(cfid), type, cname->ln_name,
4991 create ? "Create the lost MDT-object as required" :
4992 "Keep the MDT-object there by default", rc);
4995 struct lfsck_namespace *ns = com->lc_file_ram;
4997 ns->ln_flags |= LF_INCONSISTENT;
5003 static int lfsck_namespace_assistant_handler_p1(const struct lu_env *env,
5004 struct lfsck_component *com,
5005 struct lfsck_assistant_req *lar)
5007 struct lfsck_thread_info *info = lfsck_env_info(env);
5008 struct lu_attr *la = &info->lti_la;
5009 struct lfsck_instance *lfsck = com->lc_lfsck;
5010 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5011 struct lfsck_namespace *ns = com->lc_file_ram;
5012 struct lfsck_assistant_data *lad = com->lc_data;
5013 struct linkea_data ldata = { NULL };
5014 const struct lu_name *cname;
5015 struct thandle *handle = NULL;
5016 struct lfsck_namespace_req *lnr =
5017 container_of0(lar, struct lfsck_namespace_req, lnr_lar);
5018 struct dt_object *dir = NULL;
5019 struct dt_object *obj = NULL;
5020 struct lfsck_assistant_object *lso = lar->lar_parent;
5021 const struct lu_fid *pfid = &lso->lso_fid;
5022 struct dt_device *dev = NULL;
5023 struct lustre_handle lh = { 0 };
5024 bool repaired = false;
5025 bool dtlocked = false;
5029 bool bad_hash = false;
5030 bool bad_linkea = false;
5034 enum lfsck_namespace_inconsistency_type type = LNIT_NONE;
5041 if (lnr->lnr_attr & LUDA_UPGRADE) {
5042 ns->ln_flags |= LF_UPGRADE;
5043 ns->ln_dirent_repaired++;
5045 } else if (lnr->lnr_attr & LUDA_REPAIR) {
5046 ns->ln_flags |= LF_INCONSISTENT;
5047 ns->ln_dirent_repaired++;
5051 if (unlikely(fid_is_zero(&lnr->lnr_fid))) {
5052 if (strcmp(lnr->lnr_name, dotdot) != 0)
5055 rc = lfsck_namespace_trace_update(env, com, pfid,
5056 LNTF_CHECK_PARENT, true);
5061 if (unlikely(!fid_is_sane(&lnr->lnr_fid))) {
5062 CDEBUG(D_LFSCK, "%s: dir scan find invalid FID "DFID
5063 " for the name entry %.*s under "DFID"\n",
5064 lfsck_lfsck2name(lfsck), PFID(&lnr->lnr_fid),
5065 lnr->lnr_namelen, lnr->lnr_name, PFID(pfid));
5067 if (strcmp(lnr->lnr_name, dotdot) != 0)
5068 /* invalid FID means bad name entry, remove it. */
5069 type = LNIT_BAD_DIRENT;
5071 /* If the parent FID is invalid, we cannot remove
5072 * the ".." entry directly. */
5073 rc = lfsck_namespace_trace_update(env, com, pfid,
5074 LNTF_CHECK_PARENT, true);
5079 if (unlikely(lnr->lnr_dir_cookie == MDS_DIR_END_OFF)) {
5080 rc = lfsck_namespace_striped_dir_rescan(env, com, lnr);
5085 if (fid_seq_is_dot(fid_seq(&lnr->lnr_fid)))
5088 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
5089 rc = lfsck_namespace_handle_striped_master(env, com, lnr);
5094 idx = lfsck_find_mdt_idx_by_fid(env, lfsck, &lnr->lnr_fid);
5096 GOTO(out, rc = idx);
5098 if (idx == lfsck_dev_idx(lfsck)) {
5099 if (unlikely(strcmp(lnr->lnr_name, dotdot) == 0))
5102 dev = lfsck->li_bottom;
5104 struct lfsck_tgt_desc *ltd;
5106 /* Usually, some local filesystem consistency verification
5107 * tools can guarantee the local namespace tree consistenct.
5108 * So the LFSCK will only verify the remote directory. */
5109 if (unlikely(strcmp(lnr->lnr_name, dotdot) == 0)) {
5110 rc = lfsck_namespace_trace_update(env, com, pfid,
5111 LNTF_CHECK_PARENT, true);
5116 ltd = lfsck_ltd2tgt(&lfsck->li_mdt_descs, idx);
5117 if (unlikely(ltd == NULL)) {
5118 CDEBUG(D_LFSCK, "%s: cannot talk with MDT %x which "
5119 "did not join the namespace LFSCK\n",
5120 lfsck_lfsck2name(lfsck), idx);
5121 lfsck_lad_set_bitmap(env, com, idx);
5123 GOTO(out, rc = -ENODEV);
5129 obj = lfsck_object_find_by_dev(env, dev, &lnr->lnr_fid);
5131 GOTO(out, rc = PTR_ERR(obj));
5133 cname = lfsck_name_get_const(env, lnr->lnr_name, lnr->lnr_namelen);
5134 if (dt_object_exists(obj) == 0) {
5138 dir = lfsck_assistant_object_load(env, lfsck, lso);
5142 GOTO(trace, rc == -ENOENT ? 0 : rc);
5146 rc = lfsck_namespace_check_exist(env, dir, obj, lnr->lnr_name);
5148 if (!lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5149 lnr->lnr_name, lnr->lnr_namelen)) {
5150 type = LNIT_BAD_DIRENT;
5155 type = LNIT_DANGLING;
5156 rc = lfsck_namespace_repair_dangling(env, com, dir,
5165 if (!(bk->lb_param & LPF_DRYRUN) && lad->lad_advance_lock) {
5168 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
5169 MDS_INODELOCK_UPDATE |
5170 MDS_INODELOCK_XATTR, LCK_EX);
5174 handle = dt_trans_create(env, dev);
5176 GOTO(out, rc = PTR_ERR(handle));
5178 rc = lfsck_declare_namespace_exec_dir(env, obj, handle);
5182 rc = dt_trans_start_local(env, dev, handle);
5186 dt_write_lock(env, obj, 0);
5190 rc = lfsck_links_read(env, obj, &ldata);
5191 if (unlikely(rc == -ENOENT)) {
5192 if (handle != NULL) {
5193 dt_write_unlock(env, obj);
5196 dt_trans_stop(env, dev, handle);
5199 lfsck_ibits_unlock(&lh, LCK_EX);
5202 /* It may happen when the remote object has been removed,
5203 * but the local MDT is not aware of that. */
5205 } else if (rc == 0) {
5206 count = ldata.ld_leh->leh_reccount;
5207 rc = linkea_links_find(&ldata, cname, pfid);
5209 (count == 1 || !S_ISDIR(lfsck_object_type(obj)))) {
5210 if ((lfsck_object_type(obj) & S_IFMT) !=
5212 ns->ln_flags |= LF_INCONSISTENT;
5213 type = LNIT_BAD_TYPE;
5219 /* If the name entry hash does not match the slave striped
5220 * directory, and the name entry does not match also, then
5221 * it is quite possible that name entry is corrupted. */
5222 if (!lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5223 lnr->lnr_name, lnr->lnr_namelen)) {
5224 ns->ln_flags |= LF_INCONSISTENT;
5225 type = LNIT_BAD_DIRENT;
5230 /* If the file type stored in the name entry does not match
5231 * the file type claimed by the object, and the object does
5232 * not recognize the name entry, then it is quite possible
5233 * that the name entry is corrupted. */
5234 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type) {
5235 ns->ln_flags |= LF_INCONSISTENT;
5236 type = LNIT_BAD_DIRENT;
5241 /* For sub-dir object, we cannot make sure whether the sub-dir
5242 * back references the parent via ".." name entry correctly or
5243 * not in the LFSCK first-stage scanning. It may be that the
5244 * (remote) sub-dir ".." name entry has no parent FID after
5245 * file-level backup/restore and its linkEA may be wrong.
5246 * So under such case, we should replace the linkEA according
5247 * to current name entry. But this needs to be done during the
5248 * LFSCK second-stage scanning. The LFSCK will record the name
5249 * entry for further possible using. */
5253 } else if (unlikely(rc == -EINVAL)) {
5254 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type)
5255 type = LNIT_BAD_TYPE;
5258 /* The magic crashed, we are not sure whether there are more
5259 * corrupt data in the linkea, so remove all linkea entries. */
5263 } else if (rc == -ENODATA) {
5264 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type)
5265 type = LNIT_BAD_TYPE;
5272 if (bk->lb_param & LPF_DRYRUN) {
5274 ns->ln_flags |= LF_UPGRADE;
5276 ns->ln_flags |= LF_INCONSISTENT;
5277 ns->ln_linkea_repaired++;
5283 if (!lustre_handle_is_used(&lh)) {
5292 dir = lfsck_assistant_object_load(env, lfsck, lso);
5296 GOTO(stop, rc == -ENOENT ? 0 : rc);
5300 rc = lfsck_namespace_check_exist(env, dir, obj, lnr->lnr_name);
5305 if (!remove && newdata)
5306 ns->ln_flags |= LF_UPGRADE;
5307 else if (remove || !(ns->ln_flags & LF_UPGRADE))
5308 ns->ln_flags |= LF_INCONSISTENT;
5313 rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, handle);
5319 rc = linkea_data_new(&ldata,
5320 &lfsck_env_info(env)->lti_linkea_buf);
5325 rc = linkea_add_buf(&ldata, cname, pfid);
5329 rc = lfsck_links_write(env, obj, &ldata, handle);
5330 if (unlikely(rc == -ENOSPC) &&
5331 S_ISREG(lfsck_object_type(obj)) && !dt_object_remote(obj)) {
5332 if (handle != NULL) {
5333 LASSERT(dt_write_locked(env, obj));
5335 dt_write_unlock(env, obj);
5338 dt_trans_stop(env, dev, handle);
5341 lfsck_ibits_unlock(&lh, LCK_EX);
5344 rc = lfsck_namespace_trace_update(env, com,
5345 &lnr->lnr_fid, LNTF_SKIP_NLINK, true);
5347 /* If we cannot record this object in the
5348 * LFSCK tracing, we have to mark the LFSCK
5349 * as LF_INCOMPLETE, then the LFSCK will
5350 * skip nlink attribute verification for
5352 ns->ln_flags |= LF_INCOMPLETE;
5360 count = ldata.ld_leh->leh_reccount;
5361 if (!S_ISDIR(lfsck_object_type(obj)) ||
5362 !dt_object_remote(obj)) {
5363 ns->ln_linkea_repaired++;
5373 dt_write_unlock(env, obj);
5375 if (handle != NULL && !IS_ERR(handle))
5376 dt_trans_stop(env, dev, handle);
5379 lfsck_ibits_unlock(&lh, LCK_EX);
5381 if (!name_is_dot_or_dotdot(lnr->lnr_name, lnr->lnr_namelen) &&
5382 !lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5383 lnr->lnr_name, lnr->lnr_namelen) &&
5384 type != LNIT_BAD_DIRENT) {
5385 ns->ln_flags |= LF_INCONSISTENT;
5389 dir = lfsck_assistant_object_load(env, lfsck, lso);
5393 GOTO(trace, rc == -ENOENT ? 0 : rc);
5397 rc = lfsck_namespace_repair_bad_name_hash(env, com, dir,
5398 lnr->lnr_lmv, lnr->lnr_name);
5404 if (type != LNIT_NONE && dir == NULL) {
5405 dir = lfsck_assistant_object_load(env, lfsck, lso);
5409 GOTO(trace, rc == -ENOENT ? 0 : rc);
5416 rc = lfsck_namespace_repair_dirent(env, com, dir,
5417 obj, lnr->lnr_name, lnr->lnr_name,
5418 lnr->lnr_type, true, false);
5422 case LNIT_BAD_DIRENT:
5424 /* XXX: This is a bad dirent, we do not know whether
5425 * the original name entry reference a regular
5426 * file or a directory, then keep the parent's
5427 * nlink count unchanged here. */
5428 rc = lfsck_namespace_repair_dirent(env, com, dir,
5429 obj, lnr->lnr_name, lnr->lnr_name,
5430 lnr->lnr_type, false, false);
5438 if (count == 1 && S_ISREG(lfsck_object_type(obj)))
5439 dt_attr_get(env, obj, la);
5443 down_write(&com->lc_sem);
5445 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail to handle "
5446 "the entry: "DFID", parent "DFID", name %.*s: rc = %d\n",
5447 lfsck_lfsck2name(lfsck), PFID(&lnr->lnr_fid), PFID(pfid),
5448 lnr->lnr_namelen, lnr->lnr_name, rc);
5450 lfsck_namespace_record_failure(env, lfsck, ns);
5451 if ((rc == -ENOTCONN || rc == -ESHUTDOWN || rc == -EREMCHG ||
5452 rc == -ETIMEDOUT || rc == -EHOSTDOWN ||
5453 rc == -EHOSTUNREACH || rc == -EINPROGRESS) &&
5454 dev != NULL && dev != lfsck->li_bottom)
5455 lfsck_lad_set_bitmap(env, com, idx);
5457 if (!(bk->lb_param & LPF_FAILOUT))
5461 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant "
5462 "repaired the entry: "DFID", parent "DFID
5463 ", name %.*s\n", lfsck_lfsck2name(lfsck),
5464 PFID(&lnr->lnr_fid), PFID(pfid),
5465 lnr->lnr_namelen, lnr->lnr_name);
5468 ns->ln_items_repaired++;
5472 ns->ln_dangling_repaired++;
5475 ns->ln_bad_type_repaired++;
5477 case LNIT_BAD_DIRENT:
5478 ns->ln_dirent_repaired++;
5484 if (bk->lb_param & LPF_DRYRUN &&
5485 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent))
5486 lfsck_pos_fill(env, lfsck,
5487 &ns->ln_pos_first_inconsistent,
5492 ns->ln_name_hash_repaired++;
5494 /* Not count repeatedly. */
5496 ns->ln_items_repaired++;
5498 if (bk->lb_param & LPF_DRYRUN &&
5499 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent))
5500 lfsck_pos_fill(env, lfsck,
5501 &ns->ln_pos_first_inconsistent,
5508 if (count > 1 || la->la_nlink > 1)
5509 ns->ln_mul_linked_checked++;
5511 up_write(&com->lc_sem);
5513 if (obj != NULL && !IS_ERR(obj))
5514 lfsck_object_put(env, obj);
5516 if (dir != NULL && !IS_ERR(dir))
5517 lfsck_object_put(env, dir);
5519 lad->lad_advance_lock = bad_linkea;
5525 * Handle one orphan under the backend /lost+found directory
5527 * Insert the orphan FID into the namespace LFSCK trace file for further
5528 * processing (via the subsequent namespace LFSCK second-stage scanning).
5529 * At the same time, remove the orphan name entry from backend /lost+found
5530 * directory. There is an interval between the orphan name entry removed
5531 * from the backend /lost+found directory and the orphan FID in the LFSCK
5532 * trace file handled. In such interval, the LFSCK can be reset, then
5533 * all the FIDs recorded in the namespace LFSCK trace file will be dropped.
5534 * To guarantee that the orphans can be found when LFSCK run next time
5535 * without e2fsck again, when remove the orphan name entry, the LFSCK
5536 * will set the orphan's ctime attribute as 1. Since normal applications
5537 * cannot change the object's ctime attribute as 1. Then when LFSCK run
5538 * next time, it can record the object (that ctime is 1) in the namespace
5539 * LFSCK trace file during the first-stage scanning.
5541 * \param[in] env pointer to the thread context
5542 * \param[in] com pointer to the lfsck component
5543 * \param[in] parent pointer to the object for the backend /lost+found
5544 * \param[in] ent pointer to the name entry for the target under the
5545 * backend /lost+found
5547 * \retval positive for repaired
5548 * \retval 0 if needs to repair nothing
5549 * \retval negative error number on failure
5551 static int lfsck_namespace_scan_local_lpf_one(const struct lu_env *env,
5552 struct lfsck_component *com,
5553 struct dt_object *parent,
5554 struct lu_dirent *ent)
5556 struct lfsck_thread_info *info = lfsck_env_info(env);
5557 struct lu_fid *key = &info->lti_fid;
5558 struct lu_attr *la = &info->lti_la;
5559 struct lfsck_instance *lfsck = com->lc_lfsck;
5560 struct dt_object *obj;
5561 struct dt_device *dev = lfsck->li_bottom;
5562 struct dt_object *child = NULL;
5563 struct thandle *th = NULL;
5570 child = lfsck_object_find_by_dev(env, dev, &ent->lde_fid);
5572 RETURN(PTR_ERR(child));
5574 LASSERT(dt_object_exists(child));
5575 LASSERT(!dt_object_remote(child));
5577 idx = lfsck_sub_trace_file_fid2idx(&ent->lde_fid);
5578 obj = com->lc_sub_trace_objs[idx].lsto_obj;
5579 fid_cpu_to_be(key, &ent->lde_fid);
5580 rc = dt_lookup(env, obj, (struct dt_rec *)&flags,
5581 (const struct dt_key *)key);
5584 flags |= LNTF_CHECK_ORPHAN;
5585 } else if (rc == -ENOENT) {
5586 flags = LNTF_CHECK_ORPHAN;
5591 th = dt_trans_create(env, dev);
5593 GOTO(out, rc = PTR_ERR(th));
5595 /* a1. remove name entry from backend /lost+found */
5596 rc = dt_declare_delete(env, parent,
5597 (const struct dt_key *)ent->lde_name, th);
5601 if (S_ISDIR(lfsck_object_type(child))) {
5602 /* a2. decrease parent's nlink */
5603 rc = dt_declare_ref_del(env, parent, th);
5609 /* a3. remove child's FID from the LFSCK trace file. */
5610 rc = dt_declare_delete(env, obj,
5611 (const struct dt_key *)key, th);
5615 /* a4. set child's ctime as 1 */
5616 memset(la, 0, sizeof(*la));
5618 la->la_valid = LA_CTIME;
5619 rc = dt_declare_attr_set(env, child, la, th);
5624 /* a5. insert child's FID into the LFSCK trace file. */
5625 rc = dt_declare_insert(env, obj, (const struct dt_rec *)&flags,
5626 (const struct dt_key *)key, th);
5630 rc = dt_trans_start_local(env, dev, th);
5634 /* b1. remove name entry from backend /lost+found */
5635 rc = dt_delete(env, parent, (const struct dt_key *)ent->lde_name, th);
5639 if (S_ISDIR(lfsck_object_type(child))) {
5640 /* b2. decrease parent's nlink */
5641 dt_write_lock(env, parent, 0);
5642 rc = dt_ref_del(env, parent, th);
5643 dt_write_unlock(env, parent);
5649 /* a3. remove child's FID from the LFSCK trace file. */
5650 rc = dt_delete(env, obj, (const struct dt_key *)key, th);
5654 /* b4. set child's ctime as 1 */
5655 rc = dt_attr_set(env, child, la, th);
5660 /* b5. insert child's FID into the LFSCK trace file. */
5661 rc = dt_insert(env, obj, (const struct dt_rec *)&flags,
5662 (const struct dt_key *)key, th, 1);
5664 GOTO(stop, rc = (rc == 0 ? 1 : rc));
5667 dt_trans_stop(env, dev, th);
5670 lfsck_object_put(env, child);
5676 * Handle orphans under the backend /lost+found directory
5678 * Some backend checker, such as e2fsck for ldiskfs may find some orphans
5679 * and put them under the backend /lost+found directory that is invisible
5680 * to client. The LFSCK will scan such directory, for the original client
5681 * visible orphans, add their fids into the namespace LFSCK trace file,
5682 * then the subsenquent namespace LFSCK second-stage scanning can handle
5683 * them as other objects to be double scanned: either move back to normal
5684 * namespace, or to the global visible orphan directory:
5685 * /ROOT/.lustre/lost+found/MDTxxxx/
5687 * \param[in] env pointer to the thread context
5688 * \param[in] com pointer to the lfsck component
5690 static void lfsck_namespace_scan_local_lpf(const struct lu_env *env,
5691 struct lfsck_component *com)
5693 struct lfsck_thread_info *info = lfsck_env_info(env);
5694 struct lu_dirent *ent =
5695 (struct lu_dirent *)info->lti_key;
5696 struct lu_seq_range *range = &info->lti_range;
5697 struct lfsck_instance *lfsck = com->lc_lfsck;
5698 struct ptlrpc_thread *thread = &lfsck->li_thread;
5699 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5700 struct lfsck_namespace *ns = com->lc_file_ram;
5701 struct dt_object *parent;
5702 const struct dt_it_ops *iops;
5704 struct seq_server_site *ss = lfsck_dev_site(lfsck);
5706 __u32 idx = lfsck_dev_idx(lfsck);
5711 parent = lfsck_object_find_by_dev(env, lfsck->li_bottom,
5712 &LU_BACKEND_LPF_FID);
5713 if (IS_ERR(parent)) {
5714 CERROR("%s: fail to find backend /lost+found: rc = %ld\n",
5715 lfsck_lfsck2name(lfsck), PTR_ERR(parent));
5719 /* It is normal that the /lost+found does not exist for ZFS backend. */
5720 if (!dt_object_exists(parent))
5723 if (unlikely(!dt_try_as_dir(env, parent)))
5724 GOTO(out, rc = -ENOTDIR);
5726 CDEBUG(D_LFSCK, "%s: start to scan backend /lost+found\n",
5727 lfsck_lfsck2name(lfsck));
5729 com->lc_new_scanned = 0;
5730 iops = &parent->do_index_ops->dio_it;
5731 di = iops->init(env, parent, LUDA_64BITHASH | LUDA_TYPE);
5733 GOTO(out, rc = PTR_ERR(di));
5735 rc = iops->load(env, di, 0);
5737 rc = iops->next(env, di);
5742 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
5743 unlikely(!thread_is_running(thread)))
5746 rc = iops->rec(env, di, (struct dt_rec *)ent,
5747 LUDA_64BITHASH | LUDA_TYPE);
5749 rc = lfsck_unpack_ent(ent, &cookie, &type);
5751 if (unlikely(rc != 0)) {
5752 CDEBUG(D_LFSCK, "%s: fail to iterate backend "
5753 "/lost+found: rc = %d\n",
5754 lfsck_lfsck2name(lfsck), rc);
5759 /* skip dot and dotdot entries */
5760 if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen))
5763 if (!fid_seq_in_fldb(fid_seq(&ent->lde_fid)))
5766 if (fid_is_norm(&ent->lde_fid)) {
5767 fld_range_set_mdt(range);
5768 rc = fld_local_lookup(env, ss->ss_server_fld,
5769 fid_seq(&ent->lde_fid), range);
5772 } else if (idx != 0) {
5773 /* If the returned FID is IGIF, then there are three
5776 * 1) The object is upgraded from old Lustre-1.8 with
5777 * IGIF assigned to such object.
5778 * 2) The object is a backend local object and is
5779 * invisible to client.
5780 * 3) The object lost its LMV EA, and since there is
5781 * no FID-in-dirent for the orphan in the backend
5782 * /lost+found directory, then the low layer will
5783 * return IGIF for such object.
5785 * For MDTx (x != 0), it is either case 2) or case 3),
5786 * but from the LFSCK view, they are indistinguishable.
5787 * To be safe, the LFSCK will keep it there and report
5788 * some message, then the adminstrator can handle that
5791 * For MDT0, it is more possible the case 1). The LFSCK
5792 * will handle the orphan as an upgraded object. */
5793 CDEBUG(D_LFSCK, "%s: the orphan %.*s with IGIF "DFID
5794 "in the backend /lost+found on the MDT %04x, "
5795 "to be safe, skip it.\n",
5796 lfsck_lfsck2name(lfsck), ent->lde_namelen,
5797 ent->lde_name, PFID(&ent->lde_fid), idx);
5801 rc = lfsck_namespace_scan_local_lpf_one(env, com, parent, ent);
5804 down_write(&com->lc_sem);
5805 com->lc_new_scanned++;
5806 ns->ln_local_lpf_scanned++;
5808 ns->ln_local_lpf_moved++;
5810 ns->ln_local_lpf_skipped++;
5812 ns->ln_local_lpf_failed++;
5813 up_write(&com->lc_sem);
5815 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
5819 lfsck_control_speed_by_self(com);
5820 if (unlikely(!thread_is_running(thread))) {
5825 rc = iops->next(env, di);
5829 iops->fini(env, di);
5834 CDEBUG(D_LFSCK, "%s: stop to scan backend /lost+found: rc = %d\n",
5835 lfsck_lfsck2name(lfsck), rc);
5837 lfsck_object_put(env, parent);
5841 * Rescan the striped directory after the master LMV EA reset.
5843 * Sometimes, the master LMV EA of the striped directory maybe lost, so when
5844 * the namespace LFSCK engine scan the striped directory for the first time,
5845 * it will be regarded as a normal directory. As the LFSCK processing, some
5846 * other LFSCK instance on other MDT will find the shard of this striped dir,
5847 * and find that the master MDT-object of the striped directory lost its LMV
5848 * EA, then such remote LFSCK instance will regenerate the master LMV EA and
5849 * notify the LFSCK instance on this MDT to rescan the striped directory.
5851 * \param[in] env pointer to the thread context
5852 * \param[in] com pointer to the lfsck component
5853 * \param[in] llu the lfsck_lmv_unit that contains the striped directory
5856 * \retval positive number for success
5857 * \retval 0 for LFSCK stopped/paused
5858 * \retval negative error number on failure
5860 static int lfsck_namespace_rescan_striped_dir(const struct lu_env *env,
5861 struct lfsck_component *com,
5862 struct lfsck_lmv_unit *llu)
5864 struct lfsck_thread_info *info = lfsck_env_info(env);
5865 struct lfsck_instance *lfsck = com->lc_lfsck;
5866 struct lfsck_assistant_data *lad = com->lc_data;
5867 struct dt_object *dir;
5868 const struct dt_it_ops *iops;
5870 struct lu_dirent *ent =
5871 (struct lu_dirent *)info->lti_key;
5872 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5873 struct ptlrpc_thread *thread = &lfsck->li_thread;
5874 struct lfsck_assistant_object *lso = NULL;
5875 struct lfsck_namespace_req *lnr;
5876 struct lfsck_assistant_req *lar;
5881 LASSERT(list_empty(&lad->lad_req_list));
5883 lfsck->li_lmv = &llu->llu_lmv;
5884 lfsck->li_obj_dir = lfsck_object_get(llu->llu_obj);
5885 rc = lfsck_open_dir(env, lfsck, 0);
5889 dir = lfsck->li_obj_dir;
5890 di = lfsck->li_di_dir;
5891 iops = &dir->do_index_ops->dio_it;
5893 rc = iops->rec(env, di, (struct dt_rec *)ent,
5894 lfsck->li_args_dir);
5896 rc = lfsck_unpack_ent(ent, &lfsck->li_cookie_dir,
5900 if (bk->lb_param & LPF_FAILOUT)
5906 if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen))
5910 lso = lfsck_assistant_object_init(env,
5911 lfsck_dto2fid(dir), NULL,
5912 lfsck->li_pos_current.lp_oit_cookie, true);
5914 if (bk->lb_param & LPF_FAILOUT)
5915 GOTO(out, rc = PTR_ERR(lso));
5922 lnr = lfsck_namespace_assistant_req_init(lfsck, lso, ent, type);
5924 if (bk->lb_param & LPF_FAILOUT)
5925 GOTO(out, rc = PTR_ERR(lnr));
5930 lar = &lnr->lnr_lar;
5931 rc = lfsck_namespace_assistant_handler_p1(env, com, lar);
5932 lfsck_namespace_assistant_req_fini(env, lar);
5933 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
5936 if (unlikely(!thread_is_running(thread)))
5940 rc = iops->next(env, di);
5944 if (lso != NULL && !IS_ERR(lso))
5945 lfsck_assistant_object_put(env, lso);
5947 lfsck_close_dir(env, lfsck, rc);
5951 /* The close_dir() may insert a dummy lnr in the lad->lad_req_list. */
5952 if (list_empty(&lad->lad_req_list))
5955 spin_lock(&lad->lad_lock);
5956 lar = list_entry(lad->lad_req_list.next, struct lfsck_assistant_req,
5958 list_del_init(&lar->lar_list);
5959 spin_unlock(&lad->lad_lock);
5961 rc = lfsck_namespace_assistant_handler_p1(env, com, lar);
5962 lfsck_namespace_assistant_req_fini(env, lar);
5964 RETURN(rc == 0 ? 1 : rc);
5968 lfsck_namespace_double_scan_one_trace_file(const struct lu_env *env,
5969 struct lfsck_component *com,
5970 struct dt_object *obj, bool first)
5972 struct lfsck_instance *lfsck = com->lc_lfsck;
5973 struct ptlrpc_thread *thread = &lfsck->li_thread;
5974 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5975 struct lfsck_namespace *ns = com->lc_file_ram;
5976 const struct dt_it_ops *iops = &obj->do_index_ops->dio_it;
5977 struct dt_object *target;
5985 di = iops->init(env, obj, 0);
5987 RETURN(PTR_ERR(di));
5990 fid_cpu_to_be(&fid, &ns->ln_fid_latest_scanned_phase2);
5993 rc = iops->get(env, di, (const struct dt_key *)&fid);
5998 /* The start one either has been processed or does not exist,
6000 rc = iops->next(env, di);
6006 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
6007 unlikely(!thread_is_running(thread)))
6010 key = iops->key(env, di);
6019 fid_be_to_cpu(&fid, (const struct lu_fid *)key);
6020 if (!fid_is_sane(&fid)) {
6025 target = lfsck_object_find_bottom(env, lfsck, &fid);
6026 if (IS_ERR(target)) {
6027 rc = PTR_ERR(target);
6031 if (dt_object_exists(target)) {
6032 rc = iops->rec(env, di, (struct dt_rec *)&flags, 0);
6034 rc = lfsck_namespace_double_scan_one(env, com,
6041 lfsck_object_put(env, target);
6044 down_write(&com->lc_sem);
6045 com->lc_new_checked++;
6046 com->lc_new_scanned++;
6047 if (rc >= 0 && fid_is_sane(&fid))
6048 ns->ln_fid_latest_scanned_phase2 = fid;
6050 ns->ln_objs_repaired_phase2++;
6052 ns->ln_objs_failed_phase2++;
6053 up_write(&com->lc_sem);
6055 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
6058 if (unlikely(cfs_time_beforeq(com->lc_time_next_checkpoint,
6059 cfs_time_current())) &&
6060 com->lc_new_checked != 0) {
6061 down_write(&com->lc_sem);
6062 ns->ln_run_time_phase2 +=
6063 cfs_duration_sec(cfs_time_current() +
6064 HALF_SEC - com->lc_time_last_checkpoint);
6065 ns->ln_time_last_checkpoint = cfs_time_current_sec();
6066 ns->ln_objs_checked_phase2 += com->lc_new_checked;
6067 com->lc_new_checked = 0;
6068 rc = lfsck_namespace_store(env, com, false);
6069 up_write(&com->lc_sem);
6073 com->lc_time_last_checkpoint = cfs_time_current();
6074 com->lc_time_next_checkpoint =
6075 com->lc_time_last_checkpoint +
6076 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
6079 lfsck_control_speed_by_self(com);
6080 if (unlikely(!thread_is_running(thread)))
6083 rc = iops->next(env, di);
6092 iops->fini(env, di);
6097 static int lfsck_namespace_assistant_handler_p2(const struct lu_env *env,
6098 struct lfsck_component *com)
6100 struct lfsck_instance *lfsck = com->lc_lfsck;
6101 struct lfsck_namespace *ns = com->lc_file_ram;
6106 while (!list_empty(&lfsck->li_list_lmv)) {
6107 struct lfsck_lmv_unit *llu;
6109 spin_lock(&lfsck->li_lock);
6110 llu = list_entry(lfsck->li_list_lmv.next,
6111 struct lfsck_lmv_unit, llu_link);
6112 list_del_init(&llu->llu_link);
6113 spin_unlock(&lfsck->li_lock);
6115 rc = lfsck_namespace_rescan_striped_dir(env, com, llu);
6120 CDEBUG(D_LFSCK, "%s: namespace LFSCK phase2 scan start\n",
6121 lfsck_lfsck2name(lfsck));
6123 lfsck_namespace_scan_local_lpf(env, com);
6125 com->lc_new_checked = 0;
6126 com->lc_new_scanned = 0;
6127 com->lc_time_last_checkpoint = cfs_time_current();
6128 com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
6129 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
6131 i = lfsck_sub_trace_file_fid2idx(&ns->ln_fid_latest_scanned_phase2);
6132 rc = lfsck_namespace_double_scan_one_trace_file(env, com,
6133 com->lc_sub_trace_objs[i].lsto_obj, true);
6134 while (rc > 0 && ++i < LFSCK_STF_COUNT)
6135 rc = lfsck_namespace_double_scan_one_trace_file(env, com,
6136 com->lc_sub_trace_objs[i].lsto_obj, false);
6138 CDEBUG(D_LFSCK, "%s: namespace LFSCK phase2 scan stop at the No. %d "
6139 "trace file: rc = %d\n", lfsck_lfsck2name(lfsck), i, rc);
6144 static void lfsck_namespace_assistant_fill_pos(const struct lu_env *env,
6145 struct lfsck_component *com,
6146 struct lfsck_position *pos)
6148 struct lfsck_assistant_data *lad = com->lc_data;
6149 struct lfsck_namespace_req *lnr;
6151 if (list_empty(&lad->lad_req_list))
6154 lnr = list_entry(lad->lad_req_list.next,
6155 struct lfsck_namespace_req,
6157 pos->lp_oit_cookie = lnr->lnr_lar.lar_parent->lso_oit_cookie;
6158 pos->lp_dir_cookie = lnr->lnr_dir_cookie - 1;
6159 pos->lp_dir_parent = lnr->lnr_lar.lar_parent->lso_fid;
6162 static int lfsck_namespace_double_scan_result(const struct lu_env *env,
6163 struct lfsck_component *com,
6166 struct lfsck_instance *lfsck = com->lc_lfsck;
6167 struct lfsck_namespace *ns = com->lc_file_ram;
6169 down_write(&com->lc_sem);
6170 ns->ln_run_time_phase2 += cfs_duration_sec(cfs_time_current() +
6171 HALF_SEC - com->lc_time_last_checkpoint);
6172 ns->ln_time_last_checkpoint = cfs_time_current_sec();
6173 ns->ln_objs_checked_phase2 += com->lc_new_checked;
6174 com->lc_new_checked = 0;
6177 if (ns->ln_flags & LF_INCOMPLETE)
6178 ns->ln_status = LS_PARTIAL;
6180 ns->ln_status = LS_COMPLETED;
6181 if (!(lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN))
6182 ns->ln_flags &= ~(LF_SCANNED_ONCE | LF_INCONSISTENT);
6183 ns->ln_time_last_complete = ns->ln_time_last_checkpoint;
6184 ns->ln_success_count++;
6185 } else if (rc == 0) {
6186 if (lfsck->li_status != 0)
6187 ns->ln_status = lfsck->li_status;
6189 ns->ln_status = LS_STOPPED;
6191 ns->ln_status = LS_FAILED;
6194 rc = lfsck_namespace_store(env, com, false);
6195 up_write(&com->lc_sem);
6201 lfsck_namespace_assistant_sync_failures_interpret(const struct lu_env *env,
6202 struct ptlrpc_request *req,
6206 struct lfsck_async_interpret_args *laia = args;
6207 struct lfsck_tgt_desc *ltd = laia->laia_ltd;
6209 ltd->ltd_synced_failures = 1;
6216 * Notify remote LFSCK instances about former failures.
6218 * The local LFSCK instance has recorded which MDTs have ever failed to respond
6219 * some LFSCK verification requests (maybe because of network issues or the MDT
6220 * itself trouble). During the respond gap the MDT may missed some name entries
6221 * verification, then the MDT cannot know whether related MDT-objects have been
6222 * referenced by related name entries or not, then in the second-stage scanning,
6223 * these MDT-objects will be regarded as orphan, if the MDT-object contains bad
6224 * linkEA for back reference, then it will misguide the LFSCK to generate wrong
6225 * name entry for repairing the orphan.
6227 * To avoid above trouble, when layout LFSCK finishes the first-stage scanning,
6228 * it will scan the bitmap for the ever failed MDTs, and notify them that they
6229 * have ever missed some name entries verification and should skip the handling
6230 * for orphan MDT-objects.
6232 * \param[in] env pointer to the thread context
6233 * \param[in] com pointer to the lfsck component
6234 * \param[in] lr pointer to the lfsck request
6236 static void lfsck_namespace_assistant_sync_failures(const struct lu_env *env,
6237 struct lfsck_component *com,
6238 struct lfsck_request *lr)
6240 struct lfsck_async_interpret_args *laia =
6241 &lfsck_env_info(env)->lti_laia2;
6242 struct lfsck_assistant_data *lad = com->lc_data;
6243 struct lfsck_namespace *ns = com->lc_file_ram;
6244 struct lfsck_instance *lfsck = com->lc_lfsck;
6245 struct lfsck_tgt_descs *ltds = &lfsck->li_mdt_descs;
6246 struct lfsck_tgt_desc *ltd;
6247 struct ptlrpc_request_set *set;
6252 if (!lad->lad_incomplete)
6255 set = ptlrpc_prep_set();
6257 GOTO(out, rc = -ENOMEM);
6259 lr->lr_flags2 = ns->ln_flags | LF_INCOMPLETE;
6260 memset(laia, 0, sizeof(*laia));
6261 lad->lad_touch_gen++;
6263 down_read(<ds->ltd_rw_sem);
6264 cfs_foreach_bit(lad->lad_bitmap, idx) {
6265 ltd = lfsck_ltd2tgt(ltds, idx);
6266 LASSERT(ltd != NULL);
6268 laia->laia_ltd = ltd;
6269 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
6270 lfsck_namespace_assistant_sync_failures_interpret,
6271 laia, LFSCK_NOTIFY);
6273 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail "
6274 "to sync failure with MDT %x: rc = %d\n",
6275 lfsck_lfsck2name(lfsck), ltd->ltd_index, rc);
6277 up_read(<ds->ltd_rw_sem);
6279 rc = ptlrpc_set_wait(set);
6280 ptlrpc_set_destroy(set);
6286 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail "
6287 "to sync failure with MDTs, and related MDTs "
6288 "may handle orphan improperly: rc = %d\n",
6289 lfsck_lfsck2name(lfsck), rc);
6294 struct lfsck_assistant_operations lfsck_namespace_assistant_ops = {
6295 .la_handler_p1 = lfsck_namespace_assistant_handler_p1,
6296 .la_handler_p2 = lfsck_namespace_assistant_handler_p2,
6297 .la_fill_pos = lfsck_namespace_assistant_fill_pos,
6298 .la_double_scan_result = lfsck_namespace_double_scan_result,
6299 .la_req_fini = lfsck_namespace_assistant_req_fini,
6300 .la_sync_failures = lfsck_namespace_assistant_sync_failures,
6304 * Verify the specified linkEA entry for the given directory object.
6305 * If the object has no such linkEA entry or it has more other linkEA
6306 * entries, then re-generate the linkEA with the given information.
6308 * \param[in] env pointer to the thread context
6309 * \param[in] obj pointer to the dt_object to be handled
6310 * \param[in] cname the name for the child in the parent directory
6311 * \param[in] pfid the parent directory's FID for the linkEA
6313 * \retval 0 for success
6314 * \retval negative error number on failure
6316 int lfsck_verify_linkea(const struct lu_env *env, struct dt_object *obj,
6317 const struct lu_name *cname, const struct lu_fid *pfid)
6319 struct dt_device *dev = lfsck_obj2dev(obj);
6320 struct linkea_data ldata = { NULL };
6321 struct lu_buf linkea_buf;
6324 int fl = LU_XATTR_CREATE;
6328 LASSERT(S_ISDIR(lfsck_object_type(obj)));
6330 rc = lfsck_links_read(env, obj, &ldata);
6331 if (rc == -ENODATA) {
6333 } else if (rc == 0) {
6334 fl = LU_XATTR_REPLACE;
6335 if (ldata.ld_leh->leh_reccount != 1) {
6338 rc = linkea_links_find(&ldata, cname, pfid);
6347 rc = linkea_data_new(&ldata, &lfsck_env_info(env)->lti_linkea_buf);
6351 rc = linkea_add_buf(&ldata, cname, pfid);
6355 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
6356 ldata.ld_leh->leh_len);
6357 th = dt_trans_create(env, dev);
6359 RETURN(PTR_ERR(th));
6361 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
6362 XATTR_NAME_LINK, fl, th);
6366 rc = dt_trans_start_local(env, dev, th);
6370 dt_write_lock(env, obj, 0);
6371 rc = dt_xattr_set(env, obj, &linkea_buf,
6372 XATTR_NAME_LINK, fl, th);
6373 dt_write_unlock(env, obj);
6378 dt_trans_stop(env, dev, th);
6383 * Get the name and parent directory's FID from the first linkEA entry.
6385 * \param[in] env pointer to the thread context
6386 * \param[in] obj pointer to the object which get linkEA from
6387 * \param[out] name pointer to the buffer to hold the name
6388 * in the first linkEA entry
6389 * \param[out] pfid pointer to the buffer to hold the parent
6390 * directory's FID in the first linkEA entry
6392 * \retval 0 for success
6393 * \retval negative error number on failure
6395 int lfsck_links_get_first(const struct lu_env *env, struct dt_object *obj,
6396 char *name, struct lu_fid *pfid)
6398 struct lu_name *cname = &lfsck_env_info(env)->lti_name;
6399 struct linkea_data ldata = { NULL };
6402 rc = lfsck_links_read(env, obj, &ldata);
6406 linkea_first_entry(&ldata);
6407 if (ldata.ld_lee == NULL)
6410 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, cname, pfid);
6411 /* To guarantee the 'name' is terminated with '0'. */
6412 memcpy(name, cname->ln_name, cname->ln_namelen);
6413 name[cname->ln_namelen] = 0;
6419 * Update the object's name entry with the given FID.
6421 * \param[in] env pointer to the thread context
6422 * \param[in] lfsck pointer to the lfsck instance
6423 * \param[in] dir pointer to the directory that holds
6425 * \param[in] name the name for the entry to be updated
6426 * \param[in] fid the new FID for the name entry referenced
6427 * \param[in] type the type for the name entry to be updated
6429 * \retval 0 for success
6430 * \retval negative error number on failure
6432 int lfsck_update_name_entry(const struct lu_env *env,
6433 struct lfsck_instance *lfsck,
6434 struct dt_object *dir, const char *name,
6435 const struct lu_fid *fid, __u32 type)
6437 struct lfsck_thread_info *info = lfsck_env_info(env);
6438 struct dt_insert_rec *rec = &info->lti_dt_rec;
6439 struct lfsck_lock_handle *llh = &info->lti_llh;
6440 struct dt_device *dev = lfsck_obj2dev(dir);
6446 rc = lfsck_lock(env, lfsck, dir, name, llh,
6447 MDS_INODELOCK_UPDATE, LCK_PW);
6451 th = dt_trans_create(env, dev);
6453 GOTO(unlock, rc = PTR_ERR(th));
6455 rc = dt_declare_delete(env, dir, (const struct dt_key *)name, th);
6459 rec->rec_type = type;
6461 rc = dt_declare_insert(env, dir, (const struct dt_rec *)rec,
6462 (const struct dt_key *)name, th);
6466 rc = dt_declare_ref_add(env, dir, th);
6470 rc = dt_trans_start_local(env, dev, th);
6474 rc = dt_delete(env, dir, (const struct dt_key *)name, th);
6475 if (rc == -ENOENT) {
6483 rc = dt_insert(env, dir, (const struct dt_rec *)rec,
6484 (const struct dt_key *)name, th, 1);
6485 if (rc == 0 && S_ISDIR(type) && !exists) {
6486 dt_write_lock(env, dir, 0);
6487 rc = dt_ref_add(env, dir, th);
6488 dt_write_unlock(env, dir);
6494 dt_trans_stop(env, dev, th);
6498 CDEBUG(D_LFSCK, "%s: update name entry "DFID"/%s with the FID "DFID
6499 " and the type %o: rc = %d\n", lfsck_lfsck2name(lfsck),
6500 PFID(lfsck_dto2fid(dir)), name, PFID(fid), type, rc);
6505 int lfsck_namespace_setup(const struct lu_env *env,
6506 struct lfsck_instance *lfsck)
6508 struct lfsck_component *com;
6509 struct lfsck_namespace *ns;
6510 struct dt_object *root = NULL;
6511 struct dt_object *obj;
6516 LASSERT(lfsck->li_master);
6522 INIT_LIST_HEAD(&com->lc_link);
6523 INIT_LIST_HEAD(&com->lc_link_dir);
6524 init_rwsem(&com->lc_sem);
6525 atomic_set(&com->lc_ref, 1);
6526 com->lc_lfsck = lfsck;
6527 com->lc_type = LFSCK_TYPE_NAMESPACE;
6528 com->lc_ops = &lfsck_namespace_ops;
6529 com->lc_data = lfsck_assistant_data_init(
6530 &lfsck_namespace_assistant_ops,
6532 if (com->lc_data == NULL)
6533 GOTO(out, rc = -ENOMEM);
6535 com->lc_file_size = sizeof(struct lfsck_namespace);
6536 OBD_ALLOC(com->lc_file_ram, com->lc_file_size);
6537 if (com->lc_file_ram == NULL)
6538 GOTO(out, rc = -ENOMEM);
6540 OBD_ALLOC(com->lc_file_disk, com->lc_file_size);
6541 if (com->lc_file_disk == NULL)
6542 GOTO(out, rc = -ENOMEM);
6544 for (i = 0; i < LFSCK_STF_COUNT; i++)
6545 mutex_init(&com->lc_sub_trace_objs[i].lsto_mutex);
6547 root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
6549 GOTO(out, rc = PTR_ERR(root));
6551 if (unlikely(!dt_try_as_dir(env, root)))
6552 GOTO(out, rc = -ENOTDIR);
6554 obj = local_index_find_or_create(env, lfsck->li_los, root,
6556 S_IFREG | S_IRUGO | S_IWUSR,
6557 &dt_lfsck_features);
6559 GOTO(out, rc = PTR_ERR(obj));
6562 rc = lfsck_namespace_load(env, com);
6564 rc = lfsck_namespace_init(env, com);
6566 rc = lfsck_namespace_reset(env, com, true);
6568 rc = lfsck_namespace_load_sub_trace_files(env, com, false);
6572 ns = com->lc_file_ram;
6573 switch (ns->ln_status) {
6578 spin_lock(&lfsck->li_lock);
6579 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
6580 spin_unlock(&lfsck->li_lock);
6583 CERROR("%s: unknown lfsck_namespace status %d\n",
6584 lfsck_lfsck2name(lfsck), ns->ln_status);
6586 case LS_SCANNING_PHASE1:
6587 case LS_SCANNING_PHASE2:
6588 /* No need to store the status to disk right now.
6589 * If the system crashed before the status stored,
6590 * it will be loaded back when next time. */
6591 ns->ln_status = LS_CRASHED;
6595 spin_lock(&lfsck->li_lock);
6596 list_add_tail(&com->lc_link, &lfsck->li_list_scan);
6597 list_add_tail(&com->lc_link_dir, &lfsck->li_list_dir);
6598 spin_unlock(&lfsck->li_lock);
6605 if (root != NULL && !IS_ERR(root))
6606 lfsck_object_put(env, root);
6608 lfsck_component_cleanup(env, com);
6609 CERROR("%s: fail to init namespace LFSCK component: rc = %d\n",
6610 lfsck_lfsck2name(lfsck), rc);