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, 2014, 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 lu_dirent *ent, __u16 type)
64 struct lfsck_namespace_req *lnr;
67 size = sizeof(*lnr) + (ent->lde_namelen & ~3) + 4;
70 return ERR_PTR(-ENOMEM);
72 INIT_LIST_HEAD(&lnr->lnr_lar.lar_list);
73 lnr->lnr_lar.lar_fid = *lfsck_dto2fid(lfsck->li_obj_dir);
74 lnr->lnr_lmv = lfsck_lmv_get(lfsck->li_lmv);
75 lnr->lnr_fid = ent->lde_fid;
76 lnr->lnr_oit_cookie = lfsck->li_pos_current.lp_oit_cookie;
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 OBD_FREE(lnr, lnr->lnr_size);
99 static void lfsck_namespace_le_to_cpu(struct lfsck_namespace *dst,
100 struct lfsck_namespace *src)
102 dst->ln_magic = le32_to_cpu(src->ln_magic);
103 dst->ln_status = le32_to_cpu(src->ln_status);
104 dst->ln_flags = le32_to_cpu(src->ln_flags);
105 dst->ln_success_count = le32_to_cpu(src->ln_success_count);
106 dst->ln_run_time_phase1 = le32_to_cpu(src->ln_run_time_phase1);
107 dst->ln_run_time_phase2 = le32_to_cpu(src->ln_run_time_phase2);
108 dst->ln_time_last_complete = le64_to_cpu(src->ln_time_last_complete);
109 dst->ln_time_latest_start = le64_to_cpu(src->ln_time_latest_start);
110 dst->ln_time_last_checkpoint =
111 le64_to_cpu(src->ln_time_last_checkpoint);
112 lfsck_position_le_to_cpu(&dst->ln_pos_latest_start,
113 &src->ln_pos_latest_start);
114 lfsck_position_le_to_cpu(&dst->ln_pos_last_checkpoint,
115 &src->ln_pos_last_checkpoint);
116 lfsck_position_le_to_cpu(&dst->ln_pos_first_inconsistent,
117 &src->ln_pos_first_inconsistent);
118 dst->ln_items_checked = le64_to_cpu(src->ln_items_checked);
119 dst->ln_items_repaired = le64_to_cpu(src->ln_items_repaired);
120 dst->ln_items_failed = le64_to_cpu(src->ln_items_failed);
121 dst->ln_dirs_checked = le64_to_cpu(src->ln_dirs_checked);
122 dst->ln_objs_checked_phase2 = le64_to_cpu(src->ln_objs_checked_phase2);
123 dst->ln_objs_repaired_phase2 =
124 le64_to_cpu(src->ln_objs_repaired_phase2);
125 dst->ln_objs_failed_phase2 = le64_to_cpu(src->ln_objs_failed_phase2);
126 dst->ln_objs_nlink_repaired = le64_to_cpu(src->ln_objs_nlink_repaired);
127 fid_le_to_cpu(&dst->ln_fid_latest_scanned_phase2,
128 &src->ln_fid_latest_scanned_phase2);
129 dst->ln_dirent_repaired = le64_to_cpu(src->ln_dirent_repaired);
130 dst->ln_linkea_repaired = le64_to_cpu(src->ln_linkea_repaired);
131 dst->ln_mul_linked_checked = le64_to_cpu(src->ln_mul_linked_checked);
132 dst->ln_mul_linked_repaired = le64_to_cpu(src->ln_mul_linked_repaired);
133 dst->ln_unknown_inconsistency =
134 le64_to_cpu(src->ln_unknown_inconsistency);
135 dst->ln_unmatched_pairs_repaired =
136 le64_to_cpu(src->ln_unmatched_pairs_repaired);
137 dst->ln_dangling_repaired = le64_to_cpu(src->ln_dangling_repaired);
138 dst->ln_mul_ref_repaired = le64_to_cpu(src->ln_mul_ref_repaired);
139 dst->ln_bad_type_repaired = le64_to_cpu(src->ln_bad_type_repaired);
140 dst->ln_lost_dirent_repaired =
141 le64_to_cpu(src->ln_lost_dirent_repaired);
142 dst->ln_striped_dirs_scanned =
143 le64_to_cpu(src->ln_striped_dirs_scanned);
144 dst->ln_striped_dirs_repaired =
145 le64_to_cpu(src->ln_striped_dirs_repaired);
146 dst->ln_striped_dirs_failed =
147 le64_to_cpu(src->ln_striped_dirs_failed);
148 dst->ln_striped_dirs_disabled =
149 le64_to_cpu(src->ln_striped_dirs_disabled);
150 dst->ln_striped_dirs_skipped =
151 le64_to_cpu(src->ln_striped_dirs_skipped);
152 dst->ln_striped_shards_scanned =
153 le64_to_cpu(src->ln_striped_shards_scanned);
154 dst->ln_striped_shards_repaired =
155 le64_to_cpu(src->ln_striped_shards_repaired);
156 dst->ln_striped_shards_failed =
157 le64_to_cpu(src->ln_striped_shards_failed);
158 dst->ln_striped_shards_skipped =
159 le64_to_cpu(src->ln_striped_shards_skipped);
160 dst->ln_name_hash_repaired = le64_to_cpu(src->ln_name_hash_repaired);
161 dst->ln_local_lpf_scanned = le64_to_cpu(src->ln_local_lpf_scanned);
162 dst->ln_local_lpf_moved = le64_to_cpu(src->ln_local_lpf_moved);
163 dst->ln_local_lpf_skipped = le64_to_cpu(src->ln_local_lpf_skipped);
164 dst->ln_local_lpf_failed = le64_to_cpu(src->ln_local_lpf_failed);
165 dst->ln_bitmap_size = le32_to_cpu(src->ln_bitmap_size);
168 static void lfsck_namespace_cpu_to_le(struct lfsck_namespace *dst,
169 struct lfsck_namespace *src)
171 dst->ln_magic = cpu_to_le32(src->ln_magic);
172 dst->ln_status = cpu_to_le32(src->ln_status);
173 dst->ln_flags = cpu_to_le32(src->ln_flags);
174 dst->ln_success_count = cpu_to_le32(src->ln_success_count);
175 dst->ln_run_time_phase1 = cpu_to_le32(src->ln_run_time_phase1);
176 dst->ln_run_time_phase2 = cpu_to_le32(src->ln_run_time_phase2);
177 dst->ln_time_last_complete = cpu_to_le64(src->ln_time_last_complete);
178 dst->ln_time_latest_start = cpu_to_le64(src->ln_time_latest_start);
179 dst->ln_time_last_checkpoint =
180 cpu_to_le64(src->ln_time_last_checkpoint);
181 lfsck_position_cpu_to_le(&dst->ln_pos_latest_start,
182 &src->ln_pos_latest_start);
183 lfsck_position_cpu_to_le(&dst->ln_pos_last_checkpoint,
184 &src->ln_pos_last_checkpoint);
185 lfsck_position_cpu_to_le(&dst->ln_pos_first_inconsistent,
186 &src->ln_pos_first_inconsistent);
187 dst->ln_items_checked = cpu_to_le64(src->ln_items_checked);
188 dst->ln_items_repaired = cpu_to_le64(src->ln_items_repaired);
189 dst->ln_items_failed = cpu_to_le64(src->ln_items_failed);
190 dst->ln_dirs_checked = cpu_to_le64(src->ln_dirs_checked);
191 dst->ln_objs_checked_phase2 = cpu_to_le64(src->ln_objs_checked_phase2);
192 dst->ln_objs_repaired_phase2 =
193 cpu_to_le64(src->ln_objs_repaired_phase2);
194 dst->ln_objs_failed_phase2 = cpu_to_le64(src->ln_objs_failed_phase2);
195 dst->ln_objs_nlink_repaired = cpu_to_le64(src->ln_objs_nlink_repaired);
196 fid_cpu_to_le(&dst->ln_fid_latest_scanned_phase2,
197 &src->ln_fid_latest_scanned_phase2);
198 dst->ln_dirent_repaired = cpu_to_le64(src->ln_dirent_repaired);
199 dst->ln_linkea_repaired = cpu_to_le64(src->ln_linkea_repaired);
200 dst->ln_mul_linked_checked = cpu_to_le64(src->ln_mul_linked_checked);
201 dst->ln_mul_linked_repaired = cpu_to_le64(src->ln_mul_linked_repaired);
202 dst->ln_unknown_inconsistency =
203 cpu_to_le64(src->ln_unknown_inconsistency);
204 dst->ln_unmatched_pairs_repaired =
205 cpu_to_le64(src->ln_unmatched_pairs_repaired);
206 dst->ln_dangling_repaired = cpu_to_le64(src->ln_dangling_repaired);
207 dst->ln_mul_ref_repaired = cpu_to_le64(src->ln_mul_ref_repaired);
208 dst->ln_bad_type_repaired = cpu_to_le64(src->ln_bad_type_repaired);
209 dst->ln_lost_dirent_repaired =
210 cpu_to_le64(src->ln_lost_dirent_repaired);
211 dst->ln_striped_dirs_scanned =
212 cpu_to_le64(src->ln_striped_dirs_scanned);
213 dst->ln_striped_dirs_repaired =
214 cpu_to_le64(src->ln_striped_dirs_repaired);
215 dst->ln_striped_dirs_failed =
216 cpu_to_le64(src->ln_striped_dirs_failed);
217 dst->ln_striped_dirs_disabled =
218 cpu_to_le64(src->ln_striped_dirs_disabled);
219 dst->ln_striped_dirs_skipped =
220 cpu_to_le64(src->ln_striped_dirs_skipped);
221 dst->ln_striped_shards_scanned =
222 cpu_to_le64(src->ln_striped_shards_scanned);
223 dst->ln_striped_shards_repaired =
224 cpu_to_le64(src->ln_striped_shards_repaired);
225 dst->ln_striped_shards_failed =
226 cpu_to_le64(src->ln_striped_shards_failed);
227 dst->ln_striped_shards_skipped =
228 cpu_to_le64(src->ln_striped_shards_skipped);
229 dst->ln_name_hash_repaired = cpu_to_le64(src->ln_name_hash_repaired);
230 dst->ln_local_lpf_scanned = cpu_to_le64(src->ln_local_lpf_scanned);
231 dst->ln_local_lpf_moved = cpu_to_le64(src->ln_local_lpf_moved);
232 dst->ln_local_lpf_skipped = cpu_to_le64(src->ln_local_lpf_skipped);
233 dst->ln_local_lpf_failed = cpu_to_le64(src->ln_local_lpf_failed);
234 dst->ln_bitmap_size = cpu_to_le32(src->ln_bitmap_size);
237 static void lfsck_namespace_record_failure(const struct lu_env *env,
238 struct lfsck_instance *lfsck,
239 struct lfsck_namespace *ns)
241 struct lfsck_position pos;
243 ns->ln_items_failed++;
244 lfsck_pos_fill(env, lfsck, &pos, false);
245 if (lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent) ||
246 lfsck_pos_is_eq(&pos, &ns->ln_pos_first_inconsistent) < 0) {
247 ns->ln_pos_first_inconsistent = pos;
249 CDEBUG(D_LFSCK, "%s: namespace LFSCK hit first non-repaired "
250 "inconsistency at the pos ["LPU64", "DFID", "LPX64"]\n",
251 lfsck_lfsck2name(lfsck),
252 ns->ln_pos_first_inconsistent.lp_oit_cookie,
253 PFID(&ns->ln_pos_first_inconsistent.lp_dir_parent),
254 ns->ln_pos_first_inconsistent.lp_dir_cookie);
259 * Load the MDT bitmap from the lfsck_namespace trace file.
261 * \param[in] env pointer to the thread context
262 * \param[in] com pointer to the lfsck component
264 * \retval 0 for success
265 * \retval negative error number on failure or data corruption
267 static int lfsck_namespace_load_bitmap(const struct lu_env *env,
268 struct lfsck_component *com)
270 struct dt_object *obj = com->lc_obj;
271 struct lfsck_assistant_data *lad = com->lc_data;
272 struct lfsck_namespace *ns = com->lc_file_ram;
273 cfs_bitmap_t *bitmap = lad->lad_bitmap;
279 if (com->lc_lfsck->li_mdt_descs.ltd_tgts_bitmap->size >
281 nbits = com->lc_lfsck->li_mdt_descs.ltd_tgts_bitmap->size;
283 nbits = ns->ln_bitmap_size;
285 if (unlikely(nbits < BITS_PER_LONG))
286 nbits = BITS_PER_LONG;
288 if (nbits > bitmap->size) {
289 __u32 new_bits = bitmap->size;
290 cfs_bitmap_t *new_bitmap;
292 while (new_bits < nbits)
295 new_bitmap = CFS_ALLOCATE_BITMAP(new_bits);
296 if (new_bitmap == NULL)
299 lad->lad_bitmap = new_bitmap;
300 CFS_FREE_BITMAP(bitmap);
304 if (ns->ln_bitmap_size == 0) {
305 lad->lad_incomplete = 0;
306 CFS_RESET_BITMAP(bitmap);
311 size = (ns->ln_bitmap_size + 7) >> 3;
312 rc = dt_xattr_get(env, obj,
313 lfsck_buf_get(env, bitmap->data, size),
314 XATTR_NAME_LFSCK_BITMAP);
316 RETURN(rc >= 0 ? -EINVAL : rc);
318 if (cfs_bitmap_check_empty(bitmap))
319 lad->lad_incomplete = 0;
321 lad->lad_incomplete = 1;
327 * Load namespace LFSCK statistics information from the trace file.
329 * For old release (Lustre-2.6 or older), the statistics information was
330 * stored as XATTR_NAME_LFSCK_NAMESPACE_OLD EA. But in Lustre-2.7, we need
331 * more statistics information. To avoid confusing old MDT when downgrade,
332 * Lustre-2.7 stores the namespace LFSCK statistics information as new
333 * XATTR_NAME_LFSCK_NAMESPACE EA.
335 * \param[in] env pointer to the thread context
336 * \param[in] com pointer to the lfsck component
338 * \retval 0 for success
339 * \retval negative error number on failure
341 static int lfsck_namespace_load(const struct lu_env *env,
342 struct lfsck_component *com)
344 int len = com->lc_file_size;
347 rc = dt_xattr_get(env, com->lc_obj,
348 lfsck_buf_get(env, com->lc_file_disk, len),
349 XATTR_NAME_LFSCK_NAMESPACE);
351 struct lfsck_namespace *ns = com->lc_file_ram;
353 lfsck_namespace_le_to_cpu(ns,
354 (struct lfsck_namespace *)com->lc_file_disk);
355 if (ns->ln_magic != LFSCK_NAMESPACE_MAGIC) {
356 CDEBUG(D_LFSCK, "%s: invalid lfsck_namespace magic "
357 "%#x != %#x\n", lfsck_lfsck2name(com->lc_lfsck),
358 ns->ln_magic, LFSCK_NAMESPACE_MAGIC);
363 } else if (rc != -ENODATA) {
364 CDEBUG(D_LFSCK, "%s: fail to load lfsck_namespace, "
365 "expected = %d: rc = %d\n",
366 lfsck_lfsck2name(com->lc_lfsck), len, rc);
370 /* Check whether it is old trace file or not.
371 * If yes, it should be reset via returning -ESTALE. */
372 rc = dt_xattr_get(env, com->lc_obj,
373 lfsck_buf_get(env, com->lc_file_disk, len),
374 XATTR_NAME_LFSCK_NAMESPACE_OLD);
382 static int lfsck_namespace_store(const struct lu_env *env,
383 struct lfsck_component *com, bool init)
385 struct dt_object *obj = com->lc_obj;
386 struct lfsck_instance *lfsck = com->lc_lfsck;
387 struct lfsck_namespace *ns = com->lc_file_ram;
388 struct lfsck_assistant_data *lad = com->lc_data;
389 struct dt_device *dev = lfsck_obj2dev(obj);
390 cfs_bitmap_t *bitmap = NULL;
391 struct thandle *handle;
393 int len = com->lc_file_size;
395 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 8, 53, 0)
396 struct lu_buf tbuf = { &len, sizeof(len) };
401 bitmap = lad->lad_bitmap;
402 nbits = bitmap->size;
405 LASSERTF((nbits & 7) == 0, "Invalid nbits %u\n", nbits);
408 ns->ln_bitmap_size = nbits;
409 lfsck_namespace_cpu_to_le((struct lfsck_namespace *)com->lc_file_disk,
411 handle = dt_trans_create(env, dev);
413 GOTO(log, rc = PTR_ERR(handle));
415 rc = dt_declare_xattr_set(env, obj,
416 lfsck_buf_get(env, com->lc_file_disk, len),
417 XATTR_NAME_LFSCK_NAMESPACE, 0, handle);
421 if (bitmap != NULL) {
422 rc = dt_declare_xattr_set(env, obj,
423 lfsck_buf_get(env, bitmap->data, nbits >> 3),
424 XATTR_NAME_LFSCK_BITMAP, 0, handle);
429 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 8, 53, 0)
430 /* To be compatible with old Lustre-2.x MDT (x <= 6), generate dummy
431 * XATTR_NAME_LFSCK_NAMESPACE_OLD EA, then when downgrade to Lustre-2.x,
432 * the old LFSCK will find "invalid" XATTR_NAME_LFSCK_NAMESPACE_OLD EA,
433 * then reset the namespace LFSCK trace file. */
435 rc = dt_declare_xattr_set(env, obj, &tbuf,
436 XATTR_NAME_LFSCK_NAMESPACE_OLD,
437 LU_XATTR_CREATE, handle);
443 rc = dt_trans_start_local(env, dev, handle);
447 rc = dt_xattr_set(env, obj,
448 lfsck_buf_get(env, com->lc_file_disk, len),
449 XATTR_NAME_LFSCK_NAMESPACE, 0, handle);
450 if (rc == 0 && bitmap != NULL)
451 rc = dt_xattr_set(env, obj,
452 lfsck_buf_get(env, bitmap->data, nbits >> 3),
453 XATTR_NAME_LFSCK_BITMAP, 0, handle);
455 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 8, 53, 0)
457 rc = dt_xattr_set(env, obj, &tbuf,
458 XATTR_NAME_LFSCK_NAMESPACE_OLD,
459 LU_XATTR_CREATE, handle);
465 dt_trans_stop(env, dev, handle);
469 CDEBUG(D_LFSCK, "%s: fail to store lfsck_namespace: rc = %d\n",
470 lfsck_lfsck2name(lfsck), rc);
474 static struct dt_object *
475 lfsck_namespace_load_one_trace_file(const struct lu_env *env,
476 struct lfsck_component *com,
477 struct dt_object *parent,
478 const char *name, bool reset)
480 struct lfsck_instance *lfsck = com->lc_lfsck;
481 struct dt_object *obj;
485 rc = local_object_unlink(env, lfsck->li_bottom, parent, name);
486 if (rc != 0 && rc != -ENOENT)
490 obj = local_index_find_or_create(env, lfsck->li_los, parent, name,
491 S_IFREG | S_IRUGO | S_IWUSR,
497 static int lfsck_namespace_load_sub_trace_files(const struct lu_env *env,
498 struct lfsck_component *com,
501 char *name = lfsck_env_info(env)->lti_key;
502 struct lfsck_sub_trace_obj *lsto;
503 struct dt_object *obj;
507 for (i = 0, lsto = &com->lc_sub_trace_objs[0];
508 i < LFSCK_STF_COUNT; i++, lsto++) {
509 snprintf(name, NAME_MAX, "%s_%02d", LFSCK_NAMESPACE, i);
510 if (lsto->lsto_obj != NULL) {
514 lfsck_object_put(env, lsto->lsto_obj);
515 lsto->lsto_obj = NULL;
518 obj = lfsck_namespace_load_one_trace_file(env, com,
519 com->lc_lfsck->li_lfsck_dir, name, reset);
523 lsto->lsto_obj = obj;
524 rc = obj->do_ops->do_index_try(env, obj, &dt_lfsck_features);
532 static int lfsck_namespace_init(const struct lu_env *env,
533 struct lfsck_component *com)
535 struct lfsck_namespace *ns = com->lc_file_ram;
538 memset(ns, 0, sizeof(*ns));
539 ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
540 ns->ln_status = LS_INIT;
541 down_write(&com->lc_sem);
542 rc = lfsck_namespace_store(env, com, true);
543 up_write(&com->lc_sem);
545 rc = lfsck_namespace_load_sub_trace_files(env, com, true);
551 * Update the namespace LFSCK trace file for the given @fid
553 * \param[in] env pointer to the thread context
554 * \param[in] com pointer to the lfsck component
555 * \param[in] fid the fid which flags to be updated in the lfsck
557 * \param[in] add true if add new flags, otherwise remove flags
559 * \retval 0 for succeed or nothing to be done
560 * \retval negative error number on failure
562 int lfsck_namespace_trace_update(const struct lu_env *env,
563 struct lfsck_component *com,
564 const struct lu_fid *fid,
565 const __u8 flags, bool add)
567 struct lfsck_instance *lfsck = com->lc_lfsck;
568 struct dt_object *obj;
569 struct lu_fid *key = &lfsck_env_info(env)->lti_fid3;
570 struct dt_device *dev;
571 struct thandle *th = NULL;
580 if (unlikely(!fid_is_sane(fid)))
583 idx = lfsck_sub_trace_file_fid2idx(fid);
584 obj = com->lc_sub_trace_objs[idx].lsto_obj;
585 dev = lfsck_obj2dev(obj);
586 mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
587 fid_cpu_to_be(key, fid);
588 rc = dt_lookup(env, obj, (struct dt_rec *)&old,
589 (const struct dt_key *)key);
592 GOTO(unlock, rc = 0);
596 } else if (rc == 0) {
598 if ((old & flags) == flags)
599 GOTO(unlock, rc = 0);
603 if ((old & flags) == 0)
604 GOTO(unlock, rc = 0);
612 th = dt_trans_create(env, dev);
614 GOTO(log, rc = PTR_ERR(th));
617 rc = dt_declare_delete(env, obj,
618 (const struct dt_key *)key, th);
624 rc = dt_declare_insert(env, obj,
625 (const struct dt_rec *)&new,
626 (const struct dt_key *)key, th);
631 rc = dt_trans_start_local(env, dev, th);
636 rc = dt_delete(env, obj, (const struct dt_key *)key, th);
642 rc = dt_insert(env, obj, (const struct dt_rec *)&new,
643 (const struct dt_key *)key, th, 1);
651 if (th != NULL && !IS_ERR(th))
652 dt_trans_stop(env, dev, th);
654 CDEBUG(D_LFSCK, "%s: namespace LFSCK %s flags for "DFID" in the "
655 "trace file, flags %x, old %x, new %x: rc = %d\n",
656 lfsck_lfsck2name(lfsck), add ? "add" : "del", PFID(fid),
657 (__u32)flags, (__u32)old, (__u32)new, rc);
660 mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
665 int lfsck_namespace_check_exist(const struct lu_env *env,
666 struct dt_object *dir,
667 struct dt_object *obj, const char *name)
669 struct lu_fid *fid = &lfsck_env_info(env)->lti_fid;
673 if (unlikely(lfsck_is_dead_obj(obj)))
674 RETURN(LFSCK_NAMEENTRY_DEAD);
676 rc = dt_lookup(env, dir, (struct dt_rec *)fid,
677 (const struct dt_key *)name);
679 RETURN(LFSCK_NAMEENTRY_REMOVED);
684 if (!lu_fid_eq(fid, lfsck_dto2fid(obj)))
685 RETURN(LFSCK_NAMEENTRY_RECREATED);
690 static int lfsck_declare_namespace_exec_dir(const struct lu_env *env,
691 struct dt_object *obj,
692 struct thandle *handle)
696 /* For destroying all invalid linkEA entries. */
697 rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, handle);
701 /* For insert new linkEA entry. */
702 rc = dt_declare_xattr_set(env, obj,
703 lfsck_buf_get_const(env, NULL, DEFAULT_LINKEA_SIZE),
704 XATTR_NAME_LINK, 0, handle);
708 int __lfsck_links_read(const struct lu_env *env, struct dt_object *obj,
709 struct linkea_data *ldata)
713 if (ldata->ld_buf->lb_buf == NULL)
716 if (!dt_object_exists(obj))
719 rc = dt_xattr_get(env, obj, ldata->ld_buf, XATTR_NAME_LINK);
721 /* Buf was too small, figure out what we need. */
722 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LINK);
726 lu_buf_realloc(ldata->ld_buf, rc);
727 if (ldata->ld_buf->lb_buf == NULL)
730 rc = dt_xattr_get(env, obj, ldata->ld_buf, XATTR_NAME_LINK);
734 rc = linkea_init(ldata);
740 * Remove linkEA for the given object.
742 * The caller should take the ldlm lock before the calling.
744 * \param[in] env pointer to the thread context
745 * \param[in] com pointer to the lfsck component
746 * \param[in] obj pointer to the dt_object to be handled
748 * \retval 0 for repaired cases
749 * \retval negative error number on failure
751 static int lfsck_namespace_links_remove(const struct lu_env *env,
752 struct lfsck_component *com,
753 struct dt_object *obj)
755 struct lfsck_instance *lfsck = com->lc_lfsck;
756 struct dt_device *dev = lfsck_obj2dev(obj);
757 struct thandle *th = NULL;
761 LASSERT(dt_object_remote(obj) == 0);
763 th = dt_trans_create(env, dev);
765 GOTO(log, rc = PTR_ERR(th));
767 rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, th);
771 rc = dt_trans_start_local(env, dev, th);
775 dt_write_lock(env, obj, 0);
776 if (unlikely(lfsck_is_dead_obj(obj)))
777 GOTO(unlock, rc = -ENOENT);
779 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
780 GOTO(unlock, rc = 0);
782 rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, th);
787 dt_write_unlock(env, obj);
790 dt_trans_stop(env, dev, th);
793 CDEBUG(D_LFSCK, "%s: namespace LFSCK remove invalid linkEA "
794 "for the object "DFID": rc = %d\n",
795 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
798 struct lfsck_namespace *ns = com->lc_file_ram;
800 ns->ln_flags |= LF_INCONSISTENT;
806 static int lfsck_links_write(const struct lu_env *env, struct dt_object *obj,
807 struct linkea_data *ldata, struct thandle *handle)
809 const struct lu_buf *buf = lfsck_buf_get_const(env,
810 ldata->ld_buf->lb_buf,
811 ldata->ld_leh->leh_len);
813 return dt_xattr_set(env, obj, buf, XATTR_NAME_LINK, 0, handle);
816 static void lfsck_namespace_unpack_linkea_entry(struct linkea_data *ldata,
817 struct lu_name *cname,
821 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, cname, pfid);
822 /* To guarantee the 'name' is terminated with '0'. */
823 memcpy(buf, cname->ln_name, cname->ln_namelen);
824 buf[cname->ln_namelen] = 0;
825 cname->ln_name = buf;
828 static int lfsck_namespace_filter_linkea_entry(struct linkea_data *ldata,
829 struct lu_name *cname,
833 struct link_ea_entry *oldlee;
837 oldlee = ldata->ld_lee;
838 oldlen = ldata->ld_reclen;
839 linkea_next_entry(ldata);
840 while (ldata->ld_lee != NULL) {
841 ldata->ld_reclen = (ldata->ld_lee->lee_reclen[0] << 8) |
842 ldata->ld_lee->lee_reclen[1];
843 if (unlikely(ldata->ld_reclen == oldlen &&
844 memcmp(ldata->ld_lee, oldlee, oldlen) == 0)) {
849 linkea_del_buf(ldata, cname);
851 linkea_next_entry(ldata);
854 ldata->ld_lee = oldlee;
855 ldata->ld_reclen = oldlen;
861 * Insert orphan into .lustre/lost+found/MDTxxxx/ locally.
863 * Add the specified orphan MDT-object to the .lustre/lost+found/MDTxxxx/
864 * with the given type to generate the name, the detailed rules for name
865 * have been described as following.
867 * The function also generates the linkEA corresponding to the name entry
868 * under the .lustre/lost+found/MDTxxxx/ for the orphan MDT-object.
870 * \param[in] env pointer to the thread context
871 * \param[in] com pointer to the lfsck component
872 * \param[in] orphan pointer to the orphan MDT-object
873 * \param[in] infix additional information for the orphan name, such as
874 * the FID for original
875 * \param[in] type the type for describing why the orphan MDT-object is
876 * created. The rules are as following:
878 * type "D": The MDT-object is a directory, it may knows its parent
879 * but because there is no valid linkEA, the LFSCK cannot
880 * know where to put it back to the namespace.
881 * type "O": The MDT-object has no linkEA, and there is no name
882 * entry that references the MDT-object.
884 * type "S": The orphan MDT-object is a shard of a striped directory
886 * \see lfsck_layout_recreate_parent() for more types.
888 * The orphan name will be like:
889 * ${FID}-${infix}-${type}-${conflict_version}
891 * \param[out] count if some others inserted some linkEA entries by race,
892 * then return the linkEA entries count.
894 * \retval positive number for repaired cases
895 * \retval 0 if needs to repair nothing
896 * \retval negative error number on failure
898 static int lfsck_namespace_insert_orphan(const struct lu_env *env,
899 struct lfsck_component *com,
900 struct dt_object *orphan,
901 const char *infix, const char *type,
904 struct lfsck_thread_info *info = lfsck_env_info(env);
905 struct lu_name *cname = &info->lti_name;
906 struct dt_insert_rec *rec = &info->lti_dt_rec;
907 struct lu_attr *la = &info->lti_la3;
908 const struct lu_fid *cfid = lfsck_dto2fid(orphan);
909 const struct lu_fid *pfid;
911 struct lfsck_instance *lfsck = com->lc_lfsck;
912 struct dt_device *dev = lfsck_obj2dev(orphan);
913 struct dt_object *parent;
914 struct thandle *th = NULL;
915 struct lfsck_lock_handle *pllh = &info->lti_llh;
916 struct lustre_handle clh = { 0 };
917 struct linkea_data ldata = { NULL };
918 struct lu_buf linkea_buf;
925 cname->ln_name = NULL;
926 if (unlikely(lfsck->li_lpf_obj == NULL))
927 GOTO(log, rc = -ENXIO);
929 parent = lfsck->li_lpf_obj;
930 pfid = lfsck_dto2fid(parent);
934 namelen = snprintf(info->lti_key, NAME_MAX, DFID"%s-%s-%d",
935 PFID(cfid), infix, type, idx++);
936 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
937 (const struct dt_key *)info->lti_key);
938 if (rc != 0 && rc != -ENOENT)
941 if (unlikely(rc == 0 && lu_fid_eq(cfid, &tfid)))
943 } while (rc == 0 && !exist);
945 rc = lfsck_lock(env, lfsck, parent, info->lti_key, pllh,
946 MDS_INODELOCK_UPDATE, LCK_PW);
950 /* Re-check whether the name conflict with othrs after taken
952 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
953 (const struct dt_key *)info->lti_key);
955 if (!lu_fid_eq(cfid, &tfid)) {
962 } else if (rc != -ENOENT) {
968 cname->ln_name = info->lti_key;
969 cname->ln_namelen = namelen;
970 rc = linkea_data_new(&ldata, &info->lti_linkea_buf2);
974 rc = linkea_add_buf(&ldata, cname, pfid);
978 rc = lfsck_ibits_lock(env, lfsck, orphan, &clh,
979 MDS_INODELOCK_UPDATE | MDS_INODELOCK_LOOKUP |
980 MDS_INODELOCK_XATTR, LCK_EX);
984 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
985 ldata.ld_leh->leh_len);
986 th = dt_trans_create(env, dev);
988 GOTO(log, rc = PTR_ERR(th));
990 if (S_ISDIR(lfsck_object_type(orphan))) {
991 rc = dt_declare_delete(env, orphan,
992 (const struct dt_key *)dotdot, th);
996 rec->rec_type = S_IFDIR;
998 rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
999 (const struct dt_key *)dotdot, th);
1004 rc = dt_declare_xattr_set(env, orphan, &linkea_buf,
1005 XATTR_NAME_LINK, 0, th);
1010 rec->rec_type = lfsck_object_type(orphan) & S_IFMT;
1011 rec->rec_fid = cfid;
1012 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1013 (const struct dt_key *)cname->ln_name,
1018 if (S_ISDIR(rec->rec_type)) {
1019 rc = dt_declare_ref_add(env, parent, th);
1025 memset(la, 0, sizeof(*la));
1026 la->la_ctime = cfs_time_current_sec();
1027 la->la_valid = LA_CTIME;
1028 rc = dt_declare_attr_set(env, orphan, la, th);
1032 rc = dt_trans_start_local(env, dev, th);
1036 dt_write_lock(env, orphan, 0);
1037 rc = lfsck_links_read(env, orphan, &ldata);
1038 if (likely((rc == -ENODATA) || (rc == -EINVAL) ||
1039 (rc == 0 && ldata.ld_leh->leh_reccount == 0))) {
1040 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1041 GOTO(unlock, rc = 1);
1043 if (S_ISDIR(lfsck_object_type(orphan))) {
1044 rc = dt_delete(env, orphan,
1045 (const struct dt_key *)dotdot, th);
1049 rec->rec_type = S_IFDIR;
1050 rec->rec_fid = pfid;
1051 rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1052 (const struct dt_key *)dotdot, th, 1);
1057 rc = dt_xattr_set(env, orphan, &linkea_buf, XATTR_NAME_LINK, 0,
1060 if (rc == 0 && count != NULL)
1061 *count = ldata.ld_leh->leh_reccount;
1065 dt_write_unlock(env, orphan);
1067 if (rc == 0 && !exist) {
1068 rec->rec_type = lfsck_object_type(orphan) & S_IFMT;
1069 rec->rec_fid = cfid;
1070 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1071 (const struct dt_key *)cname->ln_name, th, 1);
1072 if (rc == 0 && S_ISDIR(rec->rec_type)) {
1073 dt_write_lock(env, parent, 0);
1074 rc = dt_ref_add(env, parent, th);
1075 dt_write_unlock(env, parent);
1080 rc = dt_attr_set(env, orphan, la, th);
1082 GOTO(stop, rc = (rc == 0 ? 1 : rc));
1085 dt_write_unlock(env, orphan);
1088 dt_trans_stop(env, dev, th);
1091 lfsck_ibits_unlock(&clh, LCK_EX);
1093 CDEBUG(D_LFSCK, "%s: namespace LFSCK insert orphan for the "
1094 "object "DFID", name = %s: rc = %d\n",
1095 lfsck_lfsck2name(lfsck), PFID(cfid),
1096 cname->ln_name != NULL ? cname->ln_name : "<NULL>", rc);
1099 struct lfsck_namespace *ns = com->lc_file_ram;
1101 ns->ln_flags |= LF_INCONSISTENT;
1108 * Add the specified name entry back to namespace.
1110 * If there is a linkEA entry that back references a name entry under
1111 * some parent directory, but such parent directory does not have the
1112 * claimed name entry. On the other hand, the linkEA entries count is
1113 * not larger than the MDT-object's hard link count. Under such case,
1114 * it is quite possible that the name entry is lost. Then the LFSCK
1115 * should add the name entry back to the namespace.
1117 * \param[in] env pointer to the thread context
1118 * \param[in] com pointer to the lfsck component
1119 * \param[in] parent pointer to the directory under which the name entry
1120 * will be inserted into
1121 * \param[in] child pointer to the object referenced by the name entry
1122 * that to be inserted into the parent
1123 * \param[in] name the name for the child in the parent directory
1125 * \retval positive number for repaired cases
1126 * \retval 0 if nothing to be repaired
1127 * \retval negative error number on failure
1129 static int lfsck_namespace_insert_normal(const struct lu_env *env,
1130 struct lfsck_component *com,
1131 struct dt_object *parent,
1132 struct dt_object *child,
1135 struct lfsck_thread_info *info = lfsck_env_info(env);
1136 struct lu_attr *la = &info->lti_la;
1137 struct dt_insert_rec *rec = &info->lti_dt_rec;
1138 struct lfsck_instance *lfsck = com->lc_lfsck;
1139 /* The child and its name may be on different MDTs. */
1140 struct dt_device *dev = lfsck->li_next;
1141 struct dt_object *pobj = NULL;
1142 struct dt_object *cobj = NULL;
1143 struct thandle *th = NULL;
1144 struct lfsck_lock_handle *llh = &info->lti_llh;
1148 /* @parent/@child may be based on lfsck->li_bottom,
1149 * but here we need the object based on the lfsck->li_next. */
1151 pobj = lfsck_object_locate(dev, parent);
1153 GOTO(log, rc = PTR_ERR(pobj));
1155 if (unlikely(!dt_try_as_dir(env, pobj)))
1156 GOTO(log, rc = -ENOTDIR);
1158 cobj = lfsck_object_locate(dev, child);
1160 GOTO(log, rc = PTR_ERR(cobj));
1162 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1165 rc = lfsck_lock(env, lfsck, parent, name, llh,
1166 MDS_INODELOCK_UPDATE, LCK_PW);
1170 th = dt_trans_create(env, dev);
1172 GOTO(unlock, rc = PTR_ERR(th));
1174 rec->rec_type = lfsck_object_type(cobj) & S_IFMT;
1175 rec->rec_fid = lfsck_dto2fid(cobj);
1176 rc = dt_declare_insert(env, pobj, (const struct dt_rec *)rec,
1177 (const struct dt_key *)name, th);
1181 if (S_ISDIR(rec->rec_type)) {
1182 rc = dt_declare_ref_add(env, pobj, th);
1187 memset(la, 0, sizeof(*la));
1188 la->la_ctime = cfs_time_current_sec();
1189 la->la_valid = LA_CTIME;
1190 rc = dt_declare_attr_set(env, pobj, la, th);
1194 rc = dt_declare_attr_set(env, cobj, la, th);
1198 rc = dt_trans_start_local(env, dev, th);
1202 rc = dt_insert(env, pobj, (const struct dt_rec *)rec,
1203 (const struct dt_key *)name, th, 1);
1207 if (S_ISDIR(rec->rec_type)) {
1208 dt_write_lock(env, pobj, 0);
1209 rc = dt_ref_add(env, pobj, th);
1210 dt_write_unlock(env, pobj);
1215 la->la_ctime = cfs_time_current_sec();
1216 rc = dt_attr_set(env, pobj, la, th);
1220 rc = dt_attr_set(env, cobj, la, th);
1222 GOTO(stop, rc = (rc == 0 ? 1 : rc));
1225 dt_trans_stop(env, dev, th);
1231 CDEBUG(D_LFSCK, "%s: namespace LFSCK insert object "DFID" with "
1232 "the name %s and type %o to the parent "DFID": rc = %d\n",
1233 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(child)), name,
1234 lfsck_object_type(child) & S_IFMT,
1235 PFID(lfsck_dto2fid(parent)), 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;
1293 LASSERT(!dt_object_exists(orphan));
1295 cname->ln_name = NULL;
1296 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1299 if (dt_object_remote(orphan)) {
1300 LASSERT(lfsck->li_lpf_root_obj != NULL);
1302 idx = lfsck_find_mdt_idx_by_fid(env, lfsck, cfid);
1304 GOTO(log, rc = idx);
1306 snprintf(name, 8, "MDT%04x", idx);
1307 rc = dt_lookup(env, lfsck->li_lpf_root_obj,
1308 (struct dt_rec *)&tfid,
1309 (const struct dt_key *)name);
1311 GOTO(log, rc = (rc == -ENOENT ? -ENXIO : rc));
1313 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
1315 GOTO(log, rc = PTR_ERR(parent));
1317 if (unlikely(!dt_try_as_dir(env, parent)))
1318 GOTO(log, rc = -ENOTDIR);
1320 if (unlikely(lfsck->li_lpf_obj == NULL))
1321 GOTO(log, rc = -ENXIO);
1323 parent = lfsck->li_lpf_obj;
1326 dev = lfsck_find_dev_by_fid(env, lfsck, cfid);
1328 GOTO(log, rc = PTR_ERR(dev));
1334 namelen = snprintf(name, 31, DFID"-P-%d",
1336 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
1337 (const struct dt_key *)name);
1338 if (rc != 0 && rc != -ENOENT)
1342 rc = lfsck_lock(env, lfsck, parent, name, llh,
1343 MDS_INODELOCK_UPDATE, LCK_PW);
1347 /* Re-check whether the name conflict with othrs after taken
1349 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
1350 (const struct dt_key *)name);
1351 if (unlikely(rc == 0)) {
1359 cname->ln_name = name;
1360 cname->ln_namelen = namelen;
1362 memset(la, 0, sizeof(*la));
1363 la->la_mode = S_IFDIR | 0700;
1364 la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
1365 LA_ATIME | LA_MTIME | LA_CTIME;
1367 orphan->do_ops->do_ah_init(env, hint, parent, orphan,
1368 la->la_mode & S_IFMT);
1370 memset(dof, 0, sizeof(*dof));
1371 dof->dof_type = dt_mode_to_dft(S_IFDIR);
1373 rc = linkea_data_new(&ldata, &info->lti_linkea_buf2);
1377 rc = linkea_add_buf(&ldata, cname, lfsck_dto2fid(parent));
1381 th = dt_trans_create(env, dev);
1383 GOTO(unlock1, rc = PTR_ERR(th));
1385 /* Sync the remote transaction to guarantee that the subsequent
1386 * lock against the @orphan can find the @orphan in time. */
1387 if (dt_object_remote(orphan))
1390 rc = dt_declare_create(env, orphan, la, hint, dof, th);
1394 if (unlikely(!dt_try_as_dir(env, orphan)))
1395 GOTO(stop, rc = -ENOTDIR);
1397 rec->rec_type = S_IFDIR;
1398 rec->rec_fid = cfid;
1399 rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
1400 (const struct dt_key *)dot, th);
1404 rec->rec_fid = lfsck_dto2fid(parent);
1405 rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
1406 (const struct dt_key *)dotdot, th);
1408 rc = dt_declare_ref_add(env, orphan, th);
1413 rc = dt_declare_ref_add(env, orphan, th);
1418 lmv->lmv_magic = LMV_MAGIC;
1419 lmv->lmv_master_mdt_index = lfsck_dev_idx(lfsck);
1420 lfsck_lmv_header_cpu_to_le(lmv2, lmv);
1421 lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
1422 rc = dt_declare_xattr_set(env, orphan, &lmv_buf,
1423 XATTR_NAME_LMV, 0, th);
1428 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
1429 ldata.ld_leh->leh_len);
1430 rc = dt_declare_xattr_set(env, orphan, &linkea_buf,
1431 XATTR_NAME_LINK, 0, th);
1435 rec->rec_fid = cfid;
1436 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1437 (const struct dt_key *)name, th);
1439 rc = dt_declare_ref_add(env, parent, th);
1444 rc = dt_trans_start_local(env, dev, th);
1448 dt_write_lock(env, orphan, 0);
1449 rc = dt_create(env, orphan, la, hint, dof, th);
1453 rec->rec_fid = cfid;
1454 rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1455 (const struct dt_key *)dot, th, 1);
1459 rec->rec_fid = lfsck_dto2fid(parent);
1460 rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1461 (const struct dt_key *)dotdot, th, 1);
1465 rc = dt_ref_add(env, orphan, th);
1470 rc = dt_xattr_set(env, orphan, &lmv_buf, XATTR_NAME_LMV, 0, th);
1475 rc = dt_xattr_set(env, orphan, &linkea_buf,
1476 XATTR_NAME_LINK, 0, th);
1477 dt_write_unlock(env, orphan);
1481 rec->rec_fid = cfid;
1482 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1483 (const struct dt_key *)name, th, 1);
1485 dt_write_lock(env, parent, 0);
1486 rc = dt_ref_add(env, parent, th);
1487 dt_write_unlock(env, parent);
1490 GOTO(stop, rc = (rc == 0 ? 1 : rc));
1493 dt_write_unlock(env, orphan);
1496 dt_trans_stop(env, dev, th);
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 struct dt_object *pobj = NULL;
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 pobj = lfsck_object_locate(dev, parent);
1779 GOTO(log, rc = PTR_ERR(pobj));
1781 if (unlikely(!dt_try_as_dir(env, pobj)))
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, pobj, (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, lfsck_dto2fid(pobj));
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, pobj, (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, pobj, (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, pobj, (const struct dt_key *)name, th);
1907 if (rc != 0 && rc != -ENOENT)
1910 rc = dt_insert(env, pobj, (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(lfsck_dto2fid(parent)),
1936 * Overwrite the linkEA for the object with the given ldata.
1938 * The caller should take the ldlm lock before the calling.
1940 * \param[in] env pointer to the thread context
1941 * \param[in] com pointer to the lfsck component
1942 * \param[in] obj pointer to the dt_object to be handled
1943 * \param[in] ldata pointer to the new linkEA data
1945 * \retval positive number for repaired cases
1946 * \retval 0 if nothing to be repaired
1947 * \retval negative error number on failure
1949 int lfsck_namespace_rebuild_linkea(const struct lu_env *env,
1950 struct lfsck_component *com,
1951 struct dt_object *obj,
1952 struct linkea_data *ldata)
1954 struct lfsck_instance *lfsck = com->lc_lfsck;
1955 struct dt_device *dev = lfsck_obj2dev(obj);
1956 struct thandle *th = NULL;
1957 struct lu_buf linkea_buf;
1961 th = dt_trans_create(env, dev);
1963 GOTO(log, rc = PTR_ERR(th));
1965 lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
1966 ldata->ld_leh->leh_len);
1967 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
1968 XATTR_NAME_LINK, 0, th);
1972 rc = dt_trans_start_local(env, dev, th);
1976 dt_write_lock(env, obj, 0);
1977 if (unlikely(lfsck_is_dead_obj(obj)))
1978 GOTO(unlock, rc = 0);
1980 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1981 GOTO(unlock, rc = 1);
1983 rc = dt_xattr_set(env, obj, &linkea_buf,
1984 XATTR_NAME_LINK, 0, th);
1986 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
1989 dt_write_unlock(env, obj);
1992 dt_trans_stop(env, dev, th);
1995 CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild linkEA for the "
1996 "object "DFID": rc = %d\n",
1997 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
2000 struct lfsck_namespace *ns = com->lc_file_ram;
2002 ns->ln_flags |= LF_INCONSISTENT;
2009 * Repair invalid name entry.
2011 * If the name entry contains invalid information, such as bad file type
2012 * or (and) corrupted object FID, then either remove the name entry or
2013 * udpate the name entry with the given (right) information.
2015 * \param[in] env pointer to the thread context
2016 * \param[in] com pointer to the lfsck component
2017 * \param[in] parent pointer to the parent directory
2018 * \param[in] child pointer to the object referenced by the name entry
2019 * \param[in] name the old name of the child under the parent directory
2020 * \param[in] name2 the new name of the child under the parent directory
2021 * \param[in] type the type claimed by the name entry
2022 * \param[in] update update the name entry if true; otherwise, remove it
2023 * \param[in] dec decrease the parent nlink count if true
2025 * \retval positive number for repaired successfully
2026 * \retval 0 if nothing to be repaired
2027 * \retval negative error number on failure
2029 int lfsck_namespace_repair_dirent(const struct lu_env *env,
2030 struct lfsck_component *com,
2031 struct dt_object *parent,
2032 struct dt_object *child,
2033 const char *name, const char *name2,
2034 __u16 type, bool update, bool dec)
2036 struct lfsck_thread_info *info = lfsck_env_info(env);
2037 struct dt_insert_rec *rec = &info->lti_dt_rec;
2038 const struct lu_fid *cfid = lfsck_dto2fid(child);
2040 struct lfsck_instance *lfsck = com->lc_lfsck;
2041 struct dt_object *dto;
2042 struct dt_device *dev = lfsck->li_next;
2043 struct thandle *th = NULL;
2044 struct lfsck_lock_handle *llh = &info->lti_llh;
2045 struct lustre_handle lh = { 0 };
2049 if (unlikely(!dt_try_as_dir(env, parent)))
2050 GOTO(log, rc = -ENOTDIR);
2052 if (!update || strcmp(name, name2) == 0)
2053 rc = lfsck_lock(env, lfsck, parent, name, llh,
2054 MDS_INODELOCK_UPDATE, LCK_PW);
2056 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
2057 MDS_INODELOCK_UPDATE, LCK_PW);
2061 th = dt_trans_create(env, dev);
2063 GOTO(unlock1, rc = PTR_ERR(th));
2065 dto = dt_object_locate(parent, th->th_dev);
2066 rc = dt_declare_delete(env, dto, (const struct dt_key *)name, th);
2071 rec->rec_type = lfsck_object_type(child) & S_IFMT;
2072 rec->rec_fid = cfid;
2073 rc = dt_declare_insert(env, dto,
2074 (const struct dt_rec *)rec,
2075 (const struct dt_key *)name2, th);
2081 rc = dt_declare_ref_del(env, dto, th);
2086 rc = dt_trans_start_local(env, dev, th);
2091 dt_write_lock(env, dto, 0);
2092 rc = dt_lookup(env, dto, (struct dt_rec *)&tfid,
2093 (const struct dt_key *)name);
2094 /* Someone has removed the bad name entry by race. */
2096 GOTO(unlock2, rc = 0);
2101 /* Someone has removed the bad name entry and reused it for other
2102 * object by race. */
2103 if (!lu_fid_eq(&tfid, cfid))
2104 GOTO(unlock2, rc = 0);
2106 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2107 GOTO(unlock2, rc = 1);
2109 rc = dt_delete(env, dto, (const struct dt_key *)name, th);
2114 rc = dt_insert(env, dto,
2115 (const struct dt_rec *)rec,
2116 (const struct dt_key *)name2, th, 1);
2122 rc = dt_ref_del(env, dto, th);
2127 GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
2130 dt_write_unlock(env, parent);
2133 dt_trans_stop(env, dev, th);
2135 /* We are not sure whether the child will become orphan or not.
2136 * Record it in the LFSCK trace file for further checking in
2137 * the second-stage scanning. */
2138 if (!update && !dec && rc == 0)
2139 lfsck_namespace_trace_update(env, com, cfid,
2140 LNTF_CHECK_LINKEA, true);
2143 /* It is harmless even if unlock the unused lock_handle */
2144 lfsck_ibits_unlock(&lh, LCK_PW);
2148 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found bad name "
2149 "entry for: parent "DFID", child "DFID", name %s, type "
2150 "in name entry %o, type claimed by child %o. repair it "
2151 "by %s with new name2 %s: rc = %d\n", lfsck_lfsck2name(lfsck),
2152 PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)),
2153 name, type, update ? lfsck_object_type(child) : 0,
2154 update ? "updating" : "removing", name2, rc);
2157 struct lfsck_namespace *ns = com->lc_file_ram;
2159 ns->ln_flags |= LF_INCONSISTENT;
2166 * Update the ".." name entry for the given object.
2168 * The object's ".." is corrupted, this function will update the ".." name
2169 * entry with the given pfid, and the linkEA with the given ldata.
2171 * The caller should take the ldlm lock before the calling.
2173 * \param[in] env pointer to the thread context
2174 * \param[in] com pointer to the lfsck component
2175 * \param[in] obj pointer to the dt_object to be handled
2176 * \param[in] pfid the new fid for the object's ".." name entry
2177 * \param[in] cname the name for the @obj in the parent directory
2179 * \retval positive number for repaired cases
2180 * \retval 0 if nothing to be repaired
2181 * \retval negative error number on failure
2183 static int lfsck_namespace_repair_unmatched_pairs(const struct lu_env *env,
2184 struct lfsck_component *com,
2185 struct dt_object *obj,
2186 const struct lu_fid *pfid,
2187 struct lu_name *cname)
2189 struct lfsck_thread_info *info = lfsck_env_info(env);
2190 struct dt_insert_rec *rec = &info->lti_dt_rec;
2191 struct lfsck_instance *lfsck = com->lc_lfsck;
2192 struct dt_device *dev = lfsck_obj2dev(obj);
2193 struct thandle *th = NULL;
2194 struct linkea_data ldata = { NULL };
2195 struct lu_buf linkea_buf;
2199 LASSERT(!dt_object_remote(obj));
2200 LASSERT(S_ISDIR(lfsck_object_type(obj)));
2202 rc = linkea_data_new(&ldata, &info->lti_big_buf);
2206 rc = linkea_add_buf(&ldata, cname, pfid);
2210 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
2211 ldata.ld_leh->leh_len);
2213 th = dt_trans_create(env, dev);
2215 GOTO(log, rc = PTR_ERR(th));
2217 rc = dt_declare_delete(env, obj, (const struct dt_key *)dotdot, th);
2221 rec->rec_type = S_IFDIR;
2222 rec->rec_fid = pfid;
2223 rc = dt_declare_insert(env, obj, (const struct dt_rec *)rec,
2224 (const struct dt_key *)dotdot, th);
2228 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
2229 XATTR_NAME_LINK, 0, th);
2233 rc = dt_trans_start_local(env, dev, th);
2237 dt_write_lock(env, obj, 0);
2238 if (unlikely(lfsck_is_dead_obj(obj)))
2239 GOTO(unlock, rc = 0);
2241 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2242 GOTO(unlock, rc = 1);
2244 /* The old ".." name entry maybe not exist. */
2245 dt_delete(env, obj, (const struct dt_key *)dotdot, th);
2247 rc = dt_insert(env, obj, (const struct dt_rec *)rec,
2248 (const struct dt_key *)dotdot, th, 1);
2252 rc = dt_xattr_set(env, obj, &linkea_buf,
2253 XATTR_NAME_LINK, 0, th);
2255 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2258 dt_write_unlock(env, obj);
2261 dt_trans_stop(env, dev, th);
2264 CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild dotdot name entry for "
2265 "the object "DFID", new parent "DFID": rc = %d\n",
2266 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)),
2270 struct lfsck_namespace *ns = com->lc_file_ram;
2272 ns->ln_flags |= LF_INCONSISTENT;
2279 * Handle orphan @obj during Double Scan Directory.
2281 * Remove the @obj's current (invalid) linkEA entries, and insert
2282 * it in the directory .lustre/lost+found/MDTxxxx/ with the name:
2283 * ${FID}-${PFID}-D-${conflict_version}
2285 * The caller should take the ldlm lock before the calling.
2287 * \param[in] env pointer to the thread context
2288 * \param[in] com pointer to the lfsck component
2289 * \param[in] obj pointer to the orphan object to be handled
2290 * \param[in] pfid the new fid for the object's ".." name entry
2291 * \param[in,out] lh ldlm lock handler for the given @obj
2292 * \param[out] type to tell the caller what the inconsistency is
2294 * \retval positive number for repaired cases
2295 * \retval 0 if nothing to be repaired
2296 * \retval negative error number on failure
2299 lfsck_namespace_dsd_orphan(const struct lu_env *env,
2300 struct lfsck_component *com,
2301 struct dt_object *obj,
2302 const struct lu_fid *pfid,
2303 struct lustre_handle *lh,
2304 enum lfsck_namespace_inconsistency_type *type)
2306 struct lfsck_thread_info *info = lfsck_env_info(env);
2307 struct lfsck_namespace *ns = com->lc_file_ram;
2311 /* Remove the unrecognized linkEA. */
2312 rc = lfsck_namespace_links_remove(env, com, obj);
2313 lfsck_ibits_unlock(lh, LCK_EX);
2314 if (rc < 0 && rc != -ENODATA)
2317 *type = LNIT_MUL_REF;
2319 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
2320 * ever tried to verify some remote MDT-object that resides on this
2321 * MDT, but this MDT failed to respond such request. So means there
2322 * may be some remote name entry on other MDT that references this
2323 * object with another name, so we cannot know whether this linkEA
2324 * is valid or not. So keep it there and maybe resolved when next
2326 if (ns->ln_flags & LF_INCOMPLETE)
2329 /* The unique linkEA is invalid, even if the ".." name entry may be
2330 * valid, we still cannot know via which name entry this directory
2331 * will be referenced. Then handle it as pure orphan. */
2332 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2333 "-"DFID, PFID(pfid));
2334 rc = lfsck_namespace_insert_orphan(env, com, obj,
2335 info->lti_tmpbuf, "D", NULL);
2341 * Double Scan Directory object for single linkEA entry case.
2343 * The given @child has unique linkEA entry. If the linkEA entry is valid,
2344 * then check whether the name is in the namespace or not, if not, add the
2345 * missing name entry back to namespace. If the linkEA entry is invalid,
2346 * then remove it and insert the @child in the .lustre/lost+found/MDTxxxx/
2349 * \param[in] env pointer to the thread context
2350 * \param[in] com pointer to the lfsck component
2351 * \param[in] child pointer to the directory to be double scanned
2352 * \param[in] pfid the FID corresponding to the ".." entry
2353 * \param[in] ldata pointer to the linkEA data for the given @child
2354 * \param[in,out] lh ldlm lock handler for the given @child
2355 * \param[out] type to tell the caller what the inconsistency is
2356 * \param[in] retry if found inconsistency, but the caller does not hold
2357 * ldlm lock on the @child, then set @retry as true
2359 * \retval positive number for repaired cases
2360 * \retval 0 if nothing to be repaired
2361 * \retval negative error number on failure
2364 lfsck_namespace_dsd_single(const struct lu_env *env,
2365 struct lfsck_component *com,
2366 struct dt_object *child,
2367 const struct lu_fid *pfid,
2368 struct linkea_data *ldata,
2369 struct lustre_handle *lh,
2370 enum lfsck_namespace_inconsistency_type *type,
2373 struct lfsck_thread_info *info = lfsck_env_info(env);
2374 struct lu_name *cname = &info->lti_name;
2375 const struct lu_fid *cfid = lfsck_dto2fid(child);
2377 struct lfsck_namespace *ns = com->lc_file_ram;
2378 struct lfsck_instance *lfsck = com->lc_lfsck;
2379 struct dt_object *parent = NULL;
2380 struct lmv_mds_md_v1 *lmv;
2384 lfsck_namespace_unpack_linkea_entry(ldata, cname, &tfid, info->lti_key);
2385 /* The unique linkEA entry with bad parent will be handled as orphan. */
2386 if (!fid_is_sane(&tfid)) {
2387 if (!lustre_handle_is_used(lh) && retry != NULL)
2390 rc = lfsck_namespace_dsd_orphan(env, com, child,
2396 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
2398 GOTO(out, rc = PTR_ERR(parent));
2400 /* We trust the unique linkEA entry in spite of whether it matches the
2401 * ".." name entry or not. Because even if the linkEA entry is wrong
2402 * and the ".." name entry is right, we still cannot know via which
2403 * name entry the child will be referenced, since all known entries
2404 * have been verified during the first-stage scanning. */
2405 if (!dt_object_exists(parent)) {
2406 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2407 * has ever tried to verify some remote MDT-object that resides
2408 * on this MDT, but this MDT failed to respond such request. So
2409 * means there may be some remote name entry on other MDT that
2410 * references this object with another name, so we cannot know
2411 * whether this linkEA is valid or not. So keep it there and
2412 * maybe resolved when next LFSCK run. */
2413 if (ns->ln_flags & LF_INCOMPLETE)
2416 if (!lustre_handle_is_used(lh) && retry != NULL) {
2422 lfsck_ibits_unlock(lh, LCK_EX);
2425 lmv = &info->lti_lmv;
2426 rc = lfsck_read_stripe_lmv(env, child, lmv);
2427 if (rc != 0 && rc != -ENODATA)
2430 if (rc == -ENODATA || lmv->lmv_magic != LMV_MAGIC_STRIPE) {
2432 } else if (lfsck_shard_name_to_index(env,
2433 cname->ln_name, cname->ln_namelen,
2434 S_IFDIR, cfid) < 0) {
2435 /* It is an invalid name entry, we
2436 * cannot trust the parent also. */
2437 rc = lfsck_namespace_shrink_linkea(env, com, child,
2438 ldata, cname, &tfid, true);
2442 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2443 "-"DFID, PFID(pfid));
2444 rc = lfsck_namespace_insert_orphan(env, com, child,
2445 info->lti_tmpbuf, "S", NULL);
2450 /* Create the lost parent as an orphan. */
2451 rc = lfsck_namespace_create_orphan_dir(env, com, parent, lmv);
2453 /* Add the missing name entry to the parent. */
2454 rc = lfsck_namespace_insert_normal(env, com, parent,
2455 child, cname->ln_name);
2456 if (unlikely(rc == -EEXIST)) {
2457 /* Unfortunately, someone reused the name
2458 * under the parent by race. So we have
2459 * to remove the linkEA entry from
2460 * current child object. It means that the
2461 * LFSCK cannot recover the system
2462 * totally back to its original status,
2463 * but it is necessary to make the
2464 * current system to be consistent. */
2465 rc = lfsck_namespace_shrink_linkea(env,
2467 cname, &tfid, true);
2469 snprintf(info->lti_tmpbuf,
2470 sizeof(info->lti_tmpbuf),
2471 "-"DFID, PFID(pfid));
2472 rc = lfsck_namespace_insert_orphan(env,
2473 com, child, info->lti_tmpbuf,
2482 /* The unique linkEA entry with bad parent will be handled as orphan. */
2483 if (unlikely(!dt_try_as_dir(env, parent))) {
2484 if (!lustre_handle_is_used(lh) && retry != NULL)
2487 rc = lfsck_namespace_dsd_orphan(env, com, child,
2493 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
2494 (const struct dt_key *)cname->ln_name);
2495 if (rc == -ENOENT) {
2496 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2497 * has ever tried to verify some remote MDT-object that resides
2498 * on this MDT, but this MDT failed to respond such request. So
2499 * means there may be some remote name entry on other MDT that
2500 * references this object with another name, so we cannot know
2501 * whether this linkEA is valid or not. So keep it there and
2502 * maybe resolved when next LFSCK run. */
2503 if (ns->ln_flags & LF_INCOMPLETE)
2506 if (!lustre_handle_is_used(lh) && retry != NULL) {
2512 lfsck_ibits_unlock(lh, LCK_EX);
2513 rc = lfsck_namespace_check_name(env, parent, child, cname);
2520 /* It is an invalid name entry, drop it. */
2521 if (unlikely(rc > 0)) {
2522 rc = lfsck_namespace_shrink_linkea(env, com, child,
2523 ldata, cname, &tfid, true);
2525 snprintf(info->lti_tmpbuf,
2526 sizeof(info->lti_tmpbuf),
2527 "-"DFID, PFID(pfid));
2528 rc = lfsck_namespace_insert_orphan(env, com,
2529 child, info->lti_tmpbuf, "D", NULL);
2535 /* Add the missing name entry back to the namespace. */
2536 rc = lfsck_namespace_insert_normal(env, com, parent, child,
2538 if (unlikely(rc == -ESTALE))
2539 /* It may happen when the remote object has been
2540 * removed, but the local MDT is not aware of that. */
2543 if (unlikely(rc == -EEXIST)) {
2544 /* Unfortunately, someone reused the name under the
2545 * parent by race. So we have to remove the linkEA
2546 * entry from current child object. It means that the
2547 * LFSCK cannot recover the system totally back to
2548 * its original status, but it is necessary to make
2549 * the current system to be consistent.
2551 * It also may be because of the LFSCK found some
2552 * internal status of create operation. Under such
2553 * case, nothing to be done. */
2554 rc = lfsck_namespace_shrink_linkea_cond(env, com,
2555 parent, child, ldata, cname, &tfid);
2557 snprintf(info->lti_tmpbuf,
2558 sizeof(info->lti_tmpbuf),
2559 "-"DFID, PFID(pfid));
2560 rc = lfsck_namespace_insert_orphan(env, com,
2561 child, info->lti_tmpbuf, "D", NULL);
2571 if (!lu_fid_eq(&tfid, cfid)) {
2572 if (!lustre_handle_is_used(lh) && retry != NULL) {
2578 lfsck_ibits_unlock(lh, LCK_EX);
2579 /* The name entry references another MDT-object that
2580 * may be created by the LFSCK for repairing dangling
2581 * name entry. Try to replace it. */
2582 rc = lfsck_namespace_replace_cond(env, com, parent, child,
2585 rc = lfsck_namespace_dsd_orphan(env, com, child,
2591 if (fid_is_zero(pfid))
2594 /* The ".." name entry is wrong, update it. */
2595 if (!lu_fid_eq(pfid, lfsck_dto2fid(parent))) {
2596 if (!lustre_handle_is_used(lh) && retry != NULL) {
2602 *type = LNIT_UNMATCHED_PAIRS;
2603 rc = lfsck_namespace_repair_unmatched_pairs(env, com, child,
2604 lfsck_dto2fid(parent), cname);
2610 if (parent != NULL && !IS_ERR(parent))
2611 lfsck_object_put(env, parent);
2617 * Double Scan Directory object for multiple linkEA entries case.
2619 * The given @child has multiple linkEA entries. There is at most one linkEA
2620 * entry will be valid, all the others will be removed. Firstly, the function
2621 * will try to find out the linkEA entry for which the name entry exists under
2622 * the given parent (@pfid). If there is no linkEA entry that matches the given
2623 * ".." name entry, then tries to find out the first linkEA entry that both the
2624 * parent and the name entry exist to rebuild a new ".." name entry.
2626 * \param[in] env pointer to the thread context
2627 * \param[in] com pointer to the lfsck component
2628 * \param[in] child pointer to the directory to be double scanned
2629 * \param[in] pfid the FID corresponding to the ".." entry
2630 * \param[in] ldata pointer to the linkEA data for the given @child
2631 * \param[in,out] lh ldlm lock handler for the given @child
2632 * \param[out] type to tell the caller what the inconsistency is
2633 * \param[in] lpf true if the ".." entry is under lost+found/MDTxxxx/
2635 * \retval positive number for repaired cases
2636 * \retval 0 if nothing to be repaired
2637 * \retval negative error number on failure
2640 lfsck_namespace_dsd_multiple(const struct lu_env *env,
2641 struct lfsck_component *com,
2642 struct dt_object *child,
2643 const struct lu_fid *pfid,
2644 struct linkea_data *ldata,
2645 struct lustre_handle *lh,
2646 enum lfsck_namespace_inconsistency_type *type,
2649 struct lfsck_thread_info *info = lfsck_env_info(env);
2650 struct lu_name *cname = &info->lti_name;
2651 const struct lu_fid *cfid = lfsck_dto2fid(child);
2652 struct lu_fid *pfid2 = &info->lti_fid3;
2654 struct lfsck_namespace *ns = com->lc_file_ram;
2655 struct lfsck_instance *lfsck = com->lc_lfsck;
2656 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
2657 struct dt_object *parent = NULL;
2658 struct linkea_data ldata_new = { NULL };
2659 int dirent_count = 0;
2660 int linkea_count = 0;
2666 while (ldata->ld_lee != NULL) {
2667 lfsck_namespace_unpack_linkea_entry(ldata, cname, &tfid,
2669 /* Drop repeated linkEA entries. */
2670 lfsck_namespace_filter_linkea_entry(ldata, cname, &tfid, true);
2671 /* Drop invalid linkEA entry. */
2672 if (!fid_is_sane(&tfid)) {
2673 linkea_del_buf(ldata, cname);
2678 /* If current dotdot is the .lustre/lost+found/MDTxxxx/,
2679 * then it is possible that: the directry object has ever
2680 * been lost, but its name entry was there. In the former
2681 * LFSCK run, during the first-stage scanning, the LFSCK
2682 * found the dangling name entry, but it did not recreate
2683 * the lost object, and when moved to the second-stage
2684 * scanning, some children objects of the lost directory
2685 * object were found, then the LFSCK recreated such lost
2686 * directory object as an orphan.
2688 * When the LFSCK runs again, if the dangling name is still
2689 * there, the LFSCK should move the orphan directory object
2690 * back to the normal namespace. */
2691 if (!lpf && !lu_fid_eq(pfid, &tfid) && once) {
2692 linkea_next_entry(ldata);
2696 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
2698 RETURN(PTR_ERR(parent));
2700 if (!dt_object_exists(parent)) {
2701 lfsck_object_put(env, parent);
2702 if (ldata->ld_leh->leh_reccount > 1) {
2703 /* If it is NOT the last linkEA entry, then
2704 * there is still other chance to make the
2705 * child to be visible via other parent, then
2706 * remove this linkEA entry. */
2707 linkea_del_buf(ldata, cname);
2715 /* The linkEA entry with bad parent will be removed. */
2716 if (unlikely(!dt_try_as_dir(env, parent))) {
2717 lfsck_object_put(env, parent);
2718 linkea_del_buf(ldata, cname);
2723 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
2724 (const struct dt_key *)cname->ln_name);
2725 *pfid2 = *lfsck_dto2fid(parent);
2726 if (rc == -ENOENT) {
2727 lfsck_object_put(env, parent);
2728 linkea_next_entry(ldata);
2733 lfsck_object_put(env, parent);
2738 if (lu_fid_eq(&tfid, cfid)) {
2739 lfsck_object_put(env, parent);
2740 if (!lu_fid_eq(pfid, pfid2)) {
2741 *type = LNIT_UNMATCHED_PAIRS;
2742 rc = lfsck_namespace_repair_unmatched_pairs(env,
2743 com, child, pfid2, cname);
2749 /* It is the most common case that we find the
2750 * name entry corresponding to the linkEA entry
2751 * that matches the ".." name entry. */
2752 rc = linkea_data_new(&ldata_new, &info->lti_big_buf);
2756 rc = linkea_add_buf(&ldata_new, cname, pfid2);
2760 rc = lfsck_namespace_rebuild_linkea(env, com, child,
2765 linkea_del_buf(ldata, cname);
2767 linkea_first_entry(ldata);
2768 /* There may be some invalid dangling name entries under
2769 * other parent directories, remove all of them. */
2770 while (ldata->ld_lee != NULL) {
2771 lfsck_namespace_unpack_linkea_entry(ldata,
2772 cname, &tfid, info->lti_key);
2773 if (!fid_is_sane(&tfid))
2776 parent = lfsck_object_find_bottom(env, lfsck,
2778 if (IS_ERR(parent)) {
2779 rc = PTR_ERR(parent);
2780 if (rc != -ENOENT &&
2781 bk->lb_param & LPF_FAILOUT)
2787 if (!dt_object_exists(parent)) {
2788 lfsck_object_put(env, parent);
2792 rc = lfsck_namespace_repair_dirent(env, com,
2793 parent, child, cname->ln_name,
2794 cname->ln_name, S_IFDIR, false, true);
2795 lfsck_object_put(env, parent);
2797 if (bk->lb_param & LPF_FAILOUT)
2806 linkea_del_buf(ldata, cname);
2809 ns->ln_dirent_repaired += dirent_count;
2814 lfsck_ibits_unlock(lh, LCK_EX);
2815 /* The name entry references another MDT-object that may be
2816 * created by the LFSCK for repairing dangling name entry.
2817 * Try to replace it. */
2818 rc = lfsck_namespace_replace_cond(env, com, parent, child,
2820 lfsck_object_put(env, parent);
2827 linkea_del_buf(ldata, cname);
2830 linkea_first_entry(ldata);
2831 if (ldata->ld_leh->leh_reccount == 1) {
2832 rc = lfsck_namespace_dsd_single(env, com, child, pfid, ldata,
2835 if (rc == 0 && fid_is_zero(pfid) && linkea_count > 0)
2836 rc = lfsck_namespace_rebuild_linkea(env, com, child,
2842 /* All linkEA entries are invalid and removed, then handle the @child
2844 if (ldata->ld_leh->leh_reccount == 0) {
2845 rc = lfsck_namespace_dsd_orphan(env, com, child, pfid, lh,
2851 /* If the dangling name entry for the orphan directory object has
2852 * been remvoed, then just check whether the directory object is
2853 * still under the .lustre/lost+found/MDTxxxx/ or not. */
2859 /* There is no linkEA entry that matches the ".." name entry. Find
2860 * the first linkEA entry that both parent and name entry exist to
2861 * rebuild a new ".." name entry. */
2871 * Repair the object's nlink attribute.
2873 * If all the known name entries have been verified, then the object's hard
2874 * link attribute should match the object's linkEA entries count unless the
2875 * object's has too much hard link to be recorded in the linkEA. Such cases
2876 * should have been marked in the LFSCK trace file. Otherwise, trust the
2877 * linkEA to update the object's nlink attribute.
2879 * \param[in] env pointer to the thread context
2880 * \param[in] com pointer to the lfsck component
2881 * \param[in] obj pointer to the dt_object to be handled
2882 * \param[in,out] la pointer to buffer to object's attribute before
2883 * and after the repairing
2885 * \retval positive number for repaired cases
2886 * \retval 0 if nothing to be repaired
2887 * \retval negative error number on failure
2889 static int lfsck_namespace_repair_nlink(const struct lu_env *env,
2890 struct lfsck_component *com,
2891 struct dt_object *obj,
2894 struct lfsck_thread_info *info = lfsck_env_info(env);
2895 struct lu_fid *tfid = &info->lti_fid3;
2896 struct lfsck_namespace *ns = com->lc_file_ram;
2897 struct lfsck_instance *lfsck = com->lc_lfsck;
2898 struct dt_device *dev = lfsck_obj2dev(obj);
2899 const struct lu_fid *cfid = lfsck_dto2fid(obj);
2900 struct thandle *th = NULL;
2901 struct linkea_data ldata = { NULL };
2902 struct lustre_handle lh = { 0 };
2903 __u32 old = la->la_nlink;
2909 LASSERT(!dt_object_remote(obj));
2910 LASSERT(S_ISREG(lfsck_object_type(obj)));
2912 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
2913 MDS_INODELOCK_UPDATE, LCK_PW);
2917 th = dt_trans_create(env, dev);
2919 GOTO(log, rc = PTR_ERR(th));
2921 la->la_valid = LA_NLINK;
2922 rc = dt_declare_attr_set(env, obj, la, th);
2926 rc = dt_trans_start_local(env, dev, th);
2930 dt_write_lock(env, obj, 0);
2931 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
2932 * ever tried to verify some remote MDT-object that resides on this
2933 * MDT, but this MDT failed to respond such request. So means there
2934 * may be some remote name entry on other MDT that references this
2935 * object with another name, so we cannot know whether this linkEA
2936 * is valid or not. So keep it there and maybe resolved when next
2938 if (ns->ln_flags & LF_INCOMPLETE)
2939 GOTO(unlock, rc = 0);
2941 fid_cpu_to_be(tfid, cfid);
2942 idx = lfsck_sub_trace_file_fid2idx(cfid);
2943 rc = dt_lookup(env, com->lc_sub_trace_objs[idx].lsto_obj,
2944 (struct dt_rec *)&flags, (const struct dt_key *)tfid);
2948 if (flags & LNTF_SKIP_NLINK)
2949 GOTO(unlock, rc = 0);
2951 rc = dt_attr_get(env, obj, la);
2953 GOTO(unlock, rc = (rc == -ENOENT ? 0 : rc));
2955 rc = lfsck_links_read2(env, obj, &ldata);
2957 GOTO(unlock, rc = (rc == -ENODATA ? 0 : rc));
2959 if (la->la_nlink == ldata.ld_leh->leh_reccount ||
2960 unlikely(la->la_nlink == 0))
2961 GOTO(unlock, rc = 0);
2963 la->la_nlink = ldata.ld_leh->leh_reccount;
2964 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2965 GOTO(unlock, rc = 1);
2967 rc = dt_attr_set(env, obj, la, th);
2969 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2972 dt_write_unlock(env, obj);
2975 dt_trans_stop(env, dev, th);
2978 lfsck_ibits_unlock(&lh, LCK_PW);
2979 CDEBUG(D_LFSCK, "%s: namespace LFSCK repaired the object "DFID"'s "
2980 "nlink count from %u to %u: rc = %d\n",
2981 lfsck_lfsck2name(lfsck), PFID(cfid), old, la->la_nlink, rc);
2984 ns->ln_flags |= LF_INCONSISTENT;
2990 * Double scan the directory object for namespace LFSCK.
2992 * This function will verify the <parent, child> pairs in the namespace tree:
2993 * the parent references the child via some name entry that should be in the
2994 * child's linkEA entry, the child should back references the parent via its
2997 * The LFSCK will scan every linkEA entry in turn until find out the first
2998 * matched pairs. If found, then all other linkEA entries will be dropped.
2999 * If all the linkEA entries cannot match the ".." name entry, then there
3000 * are serveral possible cases:
3002 * 1) If there is only one linkEA entry, then trust it as long as the PFID
3003 * in the linkEA entry is valid.
3005 * 2) If there are multiple linkEA entries, then try to find the linkEA
3006 * that matches the ".." name entry. If found, then all other entries
3007 * are invalid; otherwise, it is quite possible that the ".." name entry
3008 * is corrupted. Under such case, the LFSCK will rebuild the ".." name
3009 * entry according to the first valid linkEA entry (both the parent and
3010 * the name entry should exist).
3012 * 3) If the directory object has no (valid) linkEA entry, then the
3013 * directory object will be handled as pure orphan and inserted
3014 * in the .lustre/lost+found/MDTxxxx/ with the name:
3015 * ${self_FID}-${PFID}-D-${conflict_version}
3017 * \param[in] env pointer to the thread context
3018 * \param[in] com pointer to the lfsck component
3019 * \param[in] child pointer to the directory object to be handled
3020 * \param[in] flags to indicate the specical checking on the @child
3022 * \retval positive number for repaired cases
3023 * \retval 0 if nothing to be repaired
3024 * \retval negative error number on failure
3026 static int lfsck_namespace_double_scan_dir(const struct lu_env *env,
3027 struct lfsck_component *com,
3028 struct dt_object *child, __u8 flags)
3030 struct lfsck_thread_info *info = lfsck_env_info(env);
3031 const struct lu_fid *cfid = lfsck_dto2fid(child);
3032 struct lu_fid *pfid = &info->lti_fid2;
3033 struct lfsck_namespace *ns = com->lc_file_ram;
3034 struct lfsck_instance *lfsck = com->lc_lfsck;
3035 struct lustre_handle lh = { 0 };
3036 struct linkea_data ldata = { NULL };
3037 bool unknown = false;
3040 enum lfsck_namespace_inconsistency_type type = LNIT_BAD_LINKEA;
3044 LASSERT(!dt_object_remote(child));
3046 if (flags & LNTF_UNCERTAIN_LMV) {
3047 if (flags & LNTF_RECHECK_NAME_HASH) {
3048 rc = lfsck_namespace_scan_shard(env, com, child);
3052 ns->ln_striped_shards_scanned++;
3054 ns->ln_striped_shards_skipped++;
3058 flags &= ~(LNTF_RECHECK_NAME_HASH | LNTF_UNCERTAIN_LMV);
3062 if (flags & (LNTF_CHECK_LINKEA | LNTF_CHECK_PARENT) &&
3063 !(lfsck->li_bookmark_ram.lb_param & LPF_ALL_TGT)) {
3064 CDEBUG(D_LFSCK, "%s: some MDT(s) maybe NOT take part in the"
3065 "the namespace LFSCK, then the LFSCK cannot guarantee"
3066 "all the name entries have been verified in first-stage"
3067 "scanning. So have to skip orphan related handling for"
3068 "the directory object "DFID" with remote name entry\n",
3069 lfsck_lfsck2name(lfsck), PFID(cfid));
3074 if (unlikely(!dt_try_as_dir(env, child)))
3075 GOTO(out, rc = -ENOTDIR);
3077 /* We only take ldlm lock on the @child when required. When the
3078 * logic comes here for the first time, it is always false. */
3082 rc = lfsck_ibits_lock(env, lfsck, child, &lh,
3083 MDS_INODELOCK_UPDATE |
3084 MDS_INODELOCK_XATTR, LCK_EX);
3089 dt_read_lock(env, child, 0);
3090 if (unlikely(lfsck_is_dead_obj(child))) {
3091 dt_read_unlock(env, child);
3096 rc = dt_lookup(env, child, (struct dt_rec *)pfid,
3097 (const struct dt_key *)dotdot);
3099 if (rc != -ENOENT && rc != -ENODATA && rc != -EINVAL) {
3100 dt_read_unlock(env, child);
3105 if (!lustre_handle_is_used(&lh)) {
3106 dt_read_unlock(env, child);
3111 } else if (lfsck->li_lpf_obj != NULL &&
3112 lu_fid_eq(pfid, lfsck_dto2fid(lfsck->li_lpf_obj))) {
3114 } else if (unlikely(!fid_is_sane(pfid))) {
3118 rc = lfsck_links_read(env, child, &ldata);
3119 dt_read_unlock(env, child);
3121 if (rc != -ENODATA && rc != -EINVAL)
3124 if (!lustre_handle_is_used(&lh))
3127 if (rc == -EINVAL && !fid_is_zero(pfid)) {
3128 /* Remove the corrupted linkEA. */
3129 rc = lfsck_namespace_links_remove(env, com, child);
3131 /* Here, because of the crashed linkEA, we
3132 * cannot know whether there is some parent
3133 * that references the child directory via
3134 * some name entry or not. So keep it there,
3135 * when the LFSCK run next time, if there is
3136 * some parent that references this object,
3137 * then the LFSCK can rebuild the linkEA;
3138 * otherwise, this object will be handled
3139 * as orphan as above. */
3142 /* 1. If we have neither ".." nor linkEA,
3143 * then it is an orphan.
3145 * 2. If we only have the ".." name entry,
3146 * but no parent references this child
3147 * directory, then handle it as orphan. */
3148 lfsck_ibits_unlock(&lh, LCK_EX);
3149 type = LNIT_MUL_REF;
3151 /* If the LFSCK is marked as LF_INCOMPLETE,
3152 * then means some MDT has ever tried to
3153 * verify some remote MDT-object that resides
3154 * on this MDT, but this MDT failed to respond
3155 * such request. So means there may be some
3156 * remote name entry on other MDT that
3157 * references this object with another name,
3158 * so we cannot know whether this linkEA is
3159 * valid or not. So keep it there and maybe
3160 * resolved when next LFSCK run. */
3161 if (ns->ln_flags & LF_INCOMPLETE)
3164 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
3165 "-"DFID, PFID(pfid));
3166 rc = lfsck_namespace_insert_orphan(env, com, child,
3167 info->lti_tmpbuf, "D", NULL);
3173 linkea_first_entry(&ldata);
3174 /* This is the most common case: the object has unique linkEA entry. */
3175 if (ldata.ld_leh->leh_reccount == 1) {
3176 rc = lfsck_namespace_dsd_single(env, com, child, pfid, &ldata,
3177 &lh, &type, &retry);
3179 LASSERT(!lustre_handle_is_used(&lh));
3188 if (!lustre_handle_is_used(&lh))
3191 if (unlikely(ldata.ld_leh->leh_reccount == 0)) {
3192 rc = lfsck_namespace_dsd_orphan(env, com, child, pfid, &lh,
3198 /* When we come here, the cases usually like that:
3199 * 1) The directory object has a corrupted linkEA entry. During the
3200 * first-stage scanning, the LFSCK cannot know such corruption,
3201 * then it appends the right linkEA entry according to the found
3202 * name entry after the bad one.
3204 * 2) The directory object has a right linkEA entry. During the
3205 * first-stage scanning, the LFSCK finds some bad name entry,
3206 * but the LFSCK cannot aware that at that time, then it adds
3207 * the bad linkEA entry for further processing. */
3208 rc = lfsck_namespace_dsd_multiple(env, com, child, pfid, &ldata,
3214 lfsck_ibits_unlock(&lh, LCK_EX);
3217 case LNIT_BAD_LINKEA:
3218 ns->ln_linkea_repaired++;
3220 case LNIT_UNMATCHED_PAIRS:
3221 ns->ln_unmatched_pairs_repaired++;
3224 ns->ln_mul_ref_repaired++;
3232 ns->ln_unknown_inconsistency++;
3238 * Double scan the MDT-object for namespace LFSCK.
3240 * If the MDT-object contains invalid or repeated linkEA entries, then drop
3241 * those entries from the linkEA; if the linkEA becomes empty or the object
3242 * has no linkEA, then it is an orphan and will be added into the directory
3243 * .lustre/lost+found/MDTxxxx/; if the remote parent is lost, then recreate
3244 * the remote parent; if the name entry corresponding to some linkEA entry
3245 * is lost, then add the name entry back to the namespace.
3247 * \param[in] env pointer to the thread context
3248 * \param[in] com pointer to the lfsck component
3249 * \param[in] child pointer to the dt_object to be handled
3250 * \param[in] flags some hints to indicate how the @child should be handled
3252 * \retval positive number for repaired cases
3253 * \retval 0 if nothing to be repaired
3254 * \retval negative error number on failure
3256 static int lfsck_namespace_double_scan_one(const struct lu_env *env,
3257 struct lfsck_component *com,
3258 struct dt_object *child, __u8 flags)
3260 struct lfsck_thread_info *info = lfsck_env_info(env);
3261 struct lu_attr *la = &info->lti_la;
3262 struct lu_name *cname = &info->lti_name;
3263 struct lu_fid *pfid = &info->lti_fid;
3264 struct lu_fid *cfid = &info->lti_fid2;
3265 struct lfsck_instance *lfsck = com->lc_lfsck;
3266 struct lfsck_namespace *ns = com->lc_file_ram;
3267 struct dt_object *parent = NULL;
3268 struct linkea_data ldata = { NULL };
3269 bool repaired = false;
3274 dt_read_lock(env, child, 0);
3275 if (unlikely(lfsck_is_dead_obj(child))) {
3276 dt_read_unlock(env, child);
3281 if (S_ISDIR(lfsck_object_type(child))) {
3282 dt_read_unlock(env, child);
3283 rc = lfsck_namespace_double_scan_dir(env, com, child, flags);
3288 rc = lfsck_links_read(env, child, &ldata);
3289 dt_read_unlock(env, child);
3293 linkea_first_entry(&ldata);
3294 while (ldata.ld_lee != NULL) {
3295 lfsck_namespace_unpack_linkea_entry(&ldata, cname, pfid,
3297 rc = lfsck_namespace_filter_linkea_entry(&ldata, cname, pfid,
3299 /* Found repeated linkEA entries */
3301 rc = lfsck_namespace_shrink_linkea(env, com, child,
3302 &ldata, cname, pfid, false);
3314 /* Invalid PFID in the linkEA entry. */
3315 if (!fid_is_sane(pfid)) {
3316 rc = lfsck_namespace_shrink_linkea(env, com, child,
3317 &ldata, cname, pfid, true);
3327 parent = lfsck_object_find_bottom(env, lfsck, pfid);
3329 GOTO(out, rc = PTR_ERR(parent));
3331 if (!dt_object_exists(parent)) {
3334 if (ldata.ld_leh->leh_reccount > 1) {
3335 /* If it is NOT the last linkEA entry, then
3336 * there is still other chance to make the
3337 * child to be visible via other parent, then
3338 * remove this linkEA entry. */
3339 rc = lfsck_namespace_shrink_linkea(env, com,
3340 child, &ldata, cname, pfid, true);
3342 /* If the LFSCK is marked as LF_INCOMPLETE,
3343 * then means some MDT has ever tried to
3344 * verify some remote MDT-object that resides
3345 * on this MDT, but this MDT failed to respond
3346 * such request. So means there may be some
3347 * remote name entry on other MDT that
3348 * references this object with another name,
3349 * so we cannot know whether this linkEA is
3350 * valid or not. So keep it there and maybe
3351 * resolved when next LFSCK run. */
3352 if (ns->ln_flags & LF_INCOMPLETE) {
3353 lfsck_object_put(env, parent);
3358 /* Create the lost parent as an orphan. */
3359 rc = lfsck_namespace_create_orphan_dir(env, com,
3362 lfsck_object_put(env, parent);
3370 /* Add the missing name entry to the parent. */
3371 rc = lfsck_namespace_insert_normal(env, com,
3372 parent, child, cname->ln_name);
3373 if (unlikely(rc == -EEXIST))
3374 /* Unfortunately, someone reused the
3375 * name under the parent by race. So we
3376 * have to remove the linkEA entry from
3377 * current child object. It means that
3378 * the LFSCK cannot recover the system
3379 * totally back to its original status,
3380 * but it is necessary to make the
3381 * current system to be consistent. */
3382 rc = lfsck_namespace_shrink_linkea(env,
3386 linkea_next_entry(&ldata);
3389 lfsck_object_put(env, parent);
3399 /* The linkEA entry with bad parent will be removed. */
3400 if (unlikely(!dt_try_as_dir(env, parent))) {
3401 lfsck_object_put(env, parent);
3402 rc = lfsck_namespace_shrink_linkea(env, com, child,
3403 &ldata, cname, pfid, true);
3413 rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
3414 (const struct dt_key *)cname->ln_name);
3415 if (rc != 0 && rc != -ENOENT) {
3416 lfsck_object_put(env, parent);
3422 if (lu_fid_eq(cfid, lfsck_dto2fid(child))) {
3423 /* It is the most common case that we
3424 * find the name entry corresponding
3425 * to the linkEA entry. */
3426 lfsck_object_put(env, parent);
3427 linkea_next_entry(&ldata);
3429 /* The name entry references another
3430 * MDT-object that may be created by
3431 * the LFSCK for repairing dangling
3432 * name entry. Try to replace it. */
3433 rc = lfsck_namespace_replace_cond(env, com,
3434 parent, child, cfid, cname);
3435 lfsck_object_put(env, parent);
3441 linkea_next_entry(&ldata);
3443 rc = lfsck_namespace_shrink_linkea(env,
3457 rc = dt_attr_get(env, child, la);
3461 /* If there is no name entry in the parent dir and the object
3462 * link count is less than the linkea entries count, then the
3463 * linkea entry should be removed. */
3464 if (ldata.ld_leh->leh_reccount > la->la_nlink) {
3465 rc = lfsck_namespace_shrink_linkea_cond(env, com,
3466 parent, child, &ldata, cname, pfid);
3467 lfsck_object_put(env, parent);
3477 /* If the LFSCK is marked as LF_INCOMPLETE, then means some
3478 * MDT has ever tried to verify some remote MDT-object that
3479 * resides on this MDT, but this MDT failed to respond such
3480 * request. So means there may be some remote name entry on
3481 * other MDT that references this object with another name,
3482 * so we cannot know whether this linkEA is valid or not.
3483 * So keep it there and maybe resolved when next LFSCK run. */
3484 if (ns->ln_flags & LF_INCOMPLETE) {
3485 lfsck_object_put(env, parent);
3490 rc = lfsck_namespace_check_name(env, parent, child, cname);
3495 lfsck_object_put(env, parent);
3500 /* It is an invalid name entry, drop it. */
3501 if (unlikely(rc > 0)) {
3502 lfsck_object_put(env, parent);
3503 rc = lfsck_namespace_shrink_linkea(env, com, child,
3504 &ldata, cname, pfid, true);
3514 /* Add the missing name entry back to the namespace. */
3515 rc = lfsck_namespace_insert_normal(env, com, parent, child,
3517 if (unlikely(rc == -ESTALE))
3518 /* It may happen when the remote object has been
3519 * removed, but the local MDT is not aware of that. */
3522 if (unlikely(rc == -EEXIST))
3523 /* Unfortunately, someone reused the name under the
3524 * parent by race. So we have to remove the linkEA
3525 * entry from current child object. It means that the
3526 * LFSCK cannot recover the system totally back to
3527 * its original status, but it is necessary to make
3528 * the current system to be consistent.
3530 * It also may be because of the LFSCK found some
3531 * internal status of create operation. Under such
3532 * case, nothing to be done. */
3533 rc = lfsck_namespace_shrink_linkea_cond(env, com,
3534 parent, child, &ldata, cname, pfid);
3536 linkea_next_entry(&ldata);
3538 lfsck_object_put(env, parent);
3549 if (rc < 0 && rc != -ENODATA)
3553 LASSERT(ldata.ld_leh != NULL);
3555 count = ldata.ld_leh->leh_reccount;
3559 /* If the LFSCK is marked as LF_INCOMPLETE, then means some
3560 * MDT has ever tried to verify some remote MDT-object that
3561 * resides on this MDT, but this MDT failed to respond such
3562 * request. So means there may be some remote name entry on
3563 * other MDT that references this object with another name,
3564 * so we cannot know whether this linkEA is valid or not.
3565 * So keep it there and maybe resolved when next LFSCK run. */
3566 if (!(ns->ln_flags & LF_INCOMPLETE)) {
3567 /* If the child becomes orphan, then insert it into
3568 * the global .lustre/lost+found/MDTxxxx directory. */
3569 rc = lfsck_namespace_insert_orphan(env, com, child,
3575 ns->ln_mul_ref_repaired++;
3580 rc = dt_attr_get(env, child, la);
3584 if (la->la_nlink != 0 && la->la_nlink != count) {
3585 rc = lfsck_namespace_repair_nlink(env, com, child, la);
3587 ns->ln_objs_nlink_repaired++;
3594 if (la->la_nlink > 1)
3595 ns->ln_mul_linked_repaired++;
3604 static void lfsck_namespace_dump_statistics(struct seq_file *m,
3605 struct lfsck_namespace *ns,
3606 __u64 checked_phase1,
3607 __u64 checked_phase2,
3611 seq_printf(m, "checked_phase1: "LPU64"\n"
3612 "checked_phase2: "LPU64"\n"
3613 "updated_phase1: "LPU64"\n"
3614 "updated_phase2: "LPU64"\n"
3615 "failed_phase1: "LPU64"\n"
3616 "failed_phase2: "LPU64"\n"
3617 "directories: "LPU64"\n"
3618 "dirent_repaired: "LPU64"\n"
3619 "linkea_repaired: "LPU64"\n"
3620 "nlinks_repaired: "LPU64"\n"
3621 "multiple_linked_checked: "LPU64"\n"
3622 "multiple_linked_repaired: "LPU64"\n"
3623 "unknown_inconsistency: "LPU64"\n"
3624 "unmatched_pairs_repaired: "LPU64"\n"
3625 "dangling_repaired: "LPU64"\n"
3626 "multiple_referenced_repaired: "LPU64"\n"
3627 "bad_file_type_repaired: "LPU64"\n"
3628 "lost_dirent_repaired: "LPU64"\n"
3629 "local_lost_found_scanned: "LPU64"\n"
3630 "local_lost_found_moved: "LPU64"\n"
3631 "local_lost_found_skipped: "LPU64"\n"
3632 "local_lost_found_failed: "LPU64"\n"
3633 "striped_dirs_scanned: "LPU64"\n"
3634 "striped_dirs_repaired: "LPU64"\n"
3635 "striped_dirs_failed: "LPU64"\n"
3636 "striped_dirs_disabled: "LPU64"\n"
3637 "striped_dirs_skipped: "LPU64"\n"
3638 "striped_shards_scanned: "LPU64"\n"
3639 "striped_shards_repaired: "LPU64"\n"
3640 "striped_shards_failed: "LPU64"\n"
3641 "striped_shards_skipped: "LPU64"\n"
3642 "name_hash_repaired: "LPU64"\n"
3643 "success_count: %u\n"
3644 "run_time_phase1: %u seconds\n"
3645 "run_time_phase2: %u seconds\n",
3648 ns->ln_items_repaired,
3649 ns->ln_objs_repaired_phase2,
3650 ns->ln_items_failed,
3651 ns->ln_objs_failed_phase2,
3652 ns->ln_dirs_checked,
3653 ns->ln_dirent_repaired,
3654 ns->ln_linkea_repaired,
3655 ns->ln_objs_nlink_repaired,
3656 ns->ln_mul_linked_checked,
3657 ns->ln_mul_linked_repaired,
3658 ns->ln_unknown_inconsistency,
3659 ns->ln_unmatched_pairs_repaired,
3660 ns->ln_dangling_repaired,
3661 ns->ln_mul_ref_repaired,
3662 ns->ln_bad_type_repaired,
3663 ns->ln_lost_dirent_repaired,
3664 ns->ln_local_lpf_scanned,
3665 ns->ln_local_lpf_moved,
3666 ns->ln_local_lpf_skipped,
3667 ns->ln_local_lpf_failed,
3668 ns->ln_striped_dirs_scanned,
3669 ns->ln_striped_dirs_repaired,
3670 ns->ln_striped_dirs_failed,
3671 ns->ln_striped_dirs_disabled,
3672 ns->ln_striped_dirs_skipped,
3673 ns->ln_striped_shards_scanned,
3674 ns->ln_striped_shards_repaired,
3675 ns->ln_striped_shards_failed,
3676 ns->ln_striped_shards_skipped,
3677 ns->ln_name_hash_repaired,
3678 ns->ln_success_count,
3683 static void lfsck_namespace_release_lmv(const struct lu_env *env,
3684 struct lfsck_component *com)
3686 struct lfsck_instance *lfsck = com->lc_lfsck;
3687 struct lfsck_namespace *ns = com->lc_file_ram;
3689 while (!list_empty(&lfsck->li_list_lmv)) {
3690 struct lfsck_lmv_unit *llu;
3691 struct lfsck_lmv *llmv;
3693 llu = list_entry(lfsck->li_list_lmv.next,
3694 struct lfsck_lmv_unit, llu_link);
3695 llmv = &llu->llu_lmv;
3697 LASSERTF(atomic_read(&llmv->ll_ref) == 1,
3698 "still in using: %u\n",
3699 atomic_read(&llmv->ll_ref));
3701 ns->ln_striped_dirs_skipped++;
3702 lfsck_lmv_put(env, llmv);
3706 static int lfsck_namespace_check_for_double_scan(const struct lu_env *env,
3707 struct lfsck_component *com,
3708 struct dt_object *obj)
3710 struct lu_attr *la = &lfsck_env_info(env)->lti_la;
3713 rc = dt_attr_get(env, obj, la);
3717 /* zero-linkEA object may be orphan, but it also maybe because
3718 * of upgrading. Currently, we cannot record it for double scan.
3719 * Because it may cause the LFSCK trace file to be too large. */
3721 /* "la_ctime" == 1 means that it has ever been removed from
3722 * backend /lost+found directory but not been added back to
3723 * the normal namespace yet. */
3725 if ((S_ISREG(lfsck_object_type(obj)) && la->la_nlink > 1) ||
3726 unlikely(la->la_ctime == 1))
3727 rc = lfsck_namespace_trace_update(env, com, lfsck_dto2fid(obj),
3728 LNTF_CHECK_LINKEA, true);
3733 /* namespace APIs */
3735 static int lfsck_namespace_reset(const struct lu_env *env,
3736 struct lfsck_component *com, bool init)
3738 struct lfsck_instance *lfsck = com->lc_lfsck;
3739 struct lfsck_namespace *ns = com->lc_file_ram;
3740 struct lfsck_assistant_data *lad = com->lc_data;
3741 struct dt_object *root;
3742 struct dt_object *dto;
3746 root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
3748 GOTO(log, rc = PTR_ERR(root));
3750 if (unlikely(!dt_try_as_dir(env, root)))
3751 GOTO(put, rc = -ENOTDIR);
3753 down_write(&com->lc_sem);
3755 memset(ns, 0, sizeof(*ns));
3757 __u32 count = ns->ln_success_count;
3758 __u64 last_time = ns->ln_time_last_complete;
3760 memset(ns, 0, sizeof(*ns));
3761 ns->ln_success_count = count;
3762 ns->ln_time_last_complete = last_time;
3764 ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
3765 ns->ln_status = LS_INIT;
3767 lfsck_object_put(env, com->lc_obj);
3769 dto = lfsck_namespace_load_one_trace_file(env, com, root,
3770 LFSCK_NAMESPACE, true);
3772 GOTO(out, rc = PTR_ERR(dto));
3775 rc = lfsck_namespace_load_sub_trace_files(env, com, true);
3779 lad->lad_incomplete = 0;
3780 CFS_RESET_BITMAP(lad->lad_bitmap);
3782 rc = lfsck_namespace_store(env, com, true);
3787 up_write(&com->lc_sem);
3790 lfsck_object_put(env, root);
3792 CDEBUG(D_LFSCK, "%s: namespace LFSCK reset: rc = %d\n",
3793 lfsck_lfsck2name(lfsck), rc);
3798 lfsck_namespace_fail(const struct lu_env *env, struct lfsck_component *com,
3801 struct lfsck_namespace *ns = com->lc_file_ram;
3803 down_write(&com->lc_sem);
3805 com->lc_new_checked++;
3806 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
3807 up_write(&com->lc_sem);
3810 static void lfsck_namespace_close_dir(const struct lu_env *env,
3811 struct lfsck_component *com)
3813 struct lfsck_namespace *ns = com->lc_file_ram;
3814 struct lfsck_assistant_data *lad = com->lc_data;
3815 struct lfsck_instance *lfsck = com->lc_lfsck;
3816 struct lfsck_lmv *llmv = lfsck->li_lmv;
3817 struct lfsck_namespace_req *lnr;
3819 sizeof(*lnr) + LFSCK_TMPBUF_LEN;
3820 bool wakeup = false;
3826 OBD_ALLOC(lnr, size);
3828 ns->ln_striped_dirs_skipped++;
3833 /* Generate a dummy request to indicate that all shards' name entry
3834 * in this striped directory has been scanned for the first time. */
3835 INIT_LIST_HEAD(&lnr->lnr_lar.lar_list);
3836 lnr->lnr_lar.lar_fid = *lfsck_dto2fid(lfsck->li_obj_dir);
3837 lnr->lnr_lmv = lfsck_lmv_get(llmv);
3838 lnr->lnr_fid = *lfsck_dto2fid(lfsck->li_obj_dir);
3839 lnr->lnr_oit_cookie = lfsck->li_pos_current.lp_oit_cookie;
3840 lnr->lnr_dir_cookie = MDS_DIR_END_OFF;
3841 lnr->lnr_size = size;
3843 spin_lock(&lad->lad_lock);
3844 if (lad->lad_assistant_status < 0) {
3845 spin_unlock(&lad->lad_lock);
3846 lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar);
3847 ns->ln_striped_dirs_skipped++;
3852 list_add_tail(&lnr->lnr_lar.lar_list, &lad->lad_req_list);
3853 if (lad->lad_prefetched == 0)
3856 lad->lad_prefetched++;
3857 spin_unlock(&lad->lad_lock);
3859 wake_up_all(&lad->lad_thread.t_ctl_waitq);
3864 static int lfsck_namespace_open_dir(const struct lu_env *env,
3865 struct lfsck_component *com)
3867 struct lfsck_instance *lfsck = com->lc_lfsck;
3868 struct lfsck_namespace *ns = com->lc_file_ram;
3869 struct lfsck_lmv *llmv = lfsck->li_lmv;
3876 if (llmv->ll_lmv_master) {
3877 struct lmv_mds_md_v1 *lmv = &llmv->ll_lmv;
3879 if (lmv->lmv_master_mdt_index != lfsck_dev_idx(lfsck)) {
3880 lmv->lmv_master_mdt_index =
3881 lfsck_dev_idx(lfsck);
3882 ns->ln_flags |= LF_INCONSISTENT;
3883 llmv->ll_lmv_updated = 1;
3886 rc = lfsck_namespace_verify_stripe_slave(env, com,
3887 lfsck->li_obj_dir, llmv);
3890 RETURN(rc > 0 ? 0 : rc);
3893 static int lfsck_namespace_checkpoint(const struct lu_env *env,
3894 struct lfsck_component *com, bool init)
3896 struct lfsck_instance *lfsck = com->lc_lfsck;
3897 struct lfsck_namespace *ns = com->lc_file_ram;
3901 rc = lfsck_checkpoint_generic(env, com);
3906 down_write(&com->lc_sem);
3908 ns->ln_pos_latest_start = lfsck->li_pos_checkpoint;
3910 ns->ln_pos_last_checkpoint = lfsck->li_pos_checkpoint;
3911 ns->ln_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
3912 HALF_SEC - lfsck->li_time_last_checkpoint);
3913 ns->ln_time_last_checkpoint = cfs_time_current_sec();
3914 ns->ln_items_checked += com->lc_new_checked;
3915 com->lc_new_checked = 0;
3918 rc = lfsck_namespace_store(env, com, false);
3919 up_write(&com->lc_sem);
3922 CDEBUG(D_LFSCK, "%s: namespace LFSCK checkpoint at the pos ["LPU64
3923 ", "DFID", "LPX64"]: rc = %d\n", lfsck_lfsck2name(lfsck),
3924 lfsck->li_pos_current.lp_oit_cookie,
3925 PFID(&lfsck->li_pos_current.lp_dir_parent),
3926 lfsck->li_pos_current.lp_dir_cookie, rc);
3928 return rc > 0 ? 0 : rc;
3931 static int lfsck_namespace_prep(const struct lu_env *env,
3932 struct lfsck_component *com,
3933 struct lfsck_start_param *lsp)
3935 struct lfsck_instance *lfsck = com->lc_lfsck;
3936 struct lfsck_namespace *ns = com->lc_file_ram;
3937 struct lfsck_position *pos = &com->lc_pos_start;
3940 rc = lfsck_namespace_load_bitmap(env, com);
3941 if (rc != 0 || ns->ln_status == LS_COMPLETED) {
3942 rc = lfsck_namespace_reset(env, com, false);
3944 rc = lfsck_set_param(env, lfsck, lsp->lsp_start, true);
3947 CDEBUG(D_LFSCK, "%s: namespace LFSCK prep failed: "
3948 "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
3954 down_write(&com->lc_sem);
3955 ns->ln_time_latest_start = cfs_time_current_sec();
3956 spin_lock(&lfsck->li_lock);
3958 if (ns->ln_flags & LF_SCANNED_ONCE) {
3959 if (!lfsck->li_drop_dryrun ||
3960 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent)) {
3961 ns->ln_status = LS_SCANNING_PHASE2;
3962 list_move_tail(&com->lc_link,
3963 &lfsck->li_list_double_scan);
3964 if (!list_empty(&com->lc_link_dir))
3965 list_del_init(&com->lc_link_dir);
3966 lfsck_pos_set_zero(pos);
3968 ns->ln_status = LS_SCANNING_PHASE1;
3969 ns->ln_run_time_phase1 = 0;
3970 ns->ln_run_time_phase2 = 0;
3971 ns->ln_items_checked = 0;
3972 ns->ln_items_repaired = 0;
3973 ns->ln_items_failed = 0;
3974 ns->ln_dirs_checked = 0;
3975 ns->ln_objs_checked_phase2 = 0;
3976 ns->ln_objs_repaired_phase2 = 0;
3977 ns->ln_objs_failed_phase2 = 0;
3978 ns->ln_objs_nlink_repaired = 0;
3979 ns->ln_dirent_repaired = 0;
3980 ns->ln_linkea_repaired = 0;
3981 ns->ln_mul_linked_checked = 0;
3982 ns->ln_mul_linked_repaired = 0;
3983 ns->ln_unknown_inconsistency = 0;
3984 ns->ln_unmatched_pairs_repaired = 0;
3985 ns->ln_dangling_repaired = 0;
3986 ns->ln_mul_ref_repaired = 0;
3987 ns->ln_bad_type_repaired = 0;
3988 ns->ln_lost_dirent_repaired = 0;
3989 ns->ln_striped_dirs_scanned = 0;
3990 ns->ln_striped_dirs_repaired = 0;
3991 ns->ln_striped_dirs_failed = 0;
3992 ns->ln_striped_dirs_disabled = 0;
3993 ns->ln_striped_dirs_skipped = 0;
3994 ns->ln_striped_shards_scanned = 0;
3995 ns->ln_striped_shards_repaired = 0;
3996 ns->ln_striped_shards_failed = 0;
3997 ns->ln_striped_shards_skipped = 0;
3998 ns->ln_name_hash_repaired = 0;
3999 fid_zero(&ns->ln_fid_latest_scanned_phase2);
4000 if (list_empty(&com->lc_link_dir))
4001 list_add_tail(&com->lc_link_dir,
4002 &lfsck->li_list_dir);
4003 *pos = ns->ln_pos_first_inconsistent;
4006 ns->ln_status = LS_SCANNING_PHASE1;
4007 if (list_empty(&com->lc_link_dir))
4008 list_add_tail(&com->lc_link_dir,
4009 &lfsck->li_list_dir);
4010 if (!lfsck->li_drop_dryrun ||
4011 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent)) {
4012 *pos = ns->ln_pos_last_checkpoint;
4013 pos->lp_oit_cookie++;
4015 *pos = ns->ln_pos_first_inconsistent;
4019 spin_unlock(&lfsck->li_lock);
4020 up_write(&com->lc_sem);
4022 rc = lfsck_start_assistant(env, com, lsp);
4024 CDEBUG(D_LFSCK, "%s: namespace LFSCK prep done, start pos ["LPU64", "
4025 DFID", "LPX64"]: rc = %d\n",
4026 lfsck_lfsck2name(lfsck), pos->lp_oit_cookie,
4027 PFID(&pos->lp_dir_parent), pos->lp_dir_cookie, rc);
4032 static int lfsck_namespace_exec_oit(const struct lu_env *env,
4033 struct lfsck_component *com,
4034 struct dt_object *obj)
4036 struct lfsck_thread_info *info = lfsck_env_info(env);
4037 struct lfsck_namespace *ns = com->lc_file_ram;
4038 struct lfsck_instance *lfsck = com->lc_lfsck;
4039 const struct lu_fid *fid = lfsck_dto2fid(obj);
4040 struct lu_fid *pfid = &info->lti_fid2;
4041 struct lu_name *cname = &info->lti_name;
4042 struct lu_seq_range *range = &info->lti_range;
4043 struct seq_server_site *ss = lfsck_dev_site(lfsck);
4044 struct linkea_data ldata = { NULL };
4045 __u32 idx = lfsck_dev_idx(lfsck);
4049 rc = lfsck_links_read(env, obj, &ldata);
4053 /* -EINVAL means crashed linkEA, should be verified. */
4054 if (rc == -EINVAL) {
4055 rc = lfsck_namespace_trace_update(env, com, fid,
4056 LNTF_CHECK_LINKEA, true);
4058 struct lustre_handle lh = { 0 };
4060 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
4061 MDS_INODELOCK_UPDATE |
4062 MDS_INODELOCK_XATTR, LCK_EX);
4064 rc = lfsck_namespace_links_remove(env, com,
4066 lfsck_ibits_unlock(&lh, LCK_EX);
4070 GOTO(out, rc = (rc == -ENOENT ? 0 : rc));
4073 if (rc == -ENODATA) {
4074 rc = lfsck_namespace_check_for_double_scan(env, com, obj);
4082 /* Record multiple-linked object. */
4083 if (ldata.ld_leh->leh_reccount > 1) {
4084 rc = lfsck_namespace_trace_update(env, com, fid,
4085 LNTF_CHECK_LINKEA, true);
4090 linkea_first_entry(&ldata);
4091 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, cname, pfid);
4092 if (!fid_is_sane(pfid)) {
4093 rc = lfsck_namespace_trace_update(env, com, fid,
4094 LNTF_CHECK_PARENT, true);
4096 fld_range_set_mdt(range);
4097 rc = fld_local_lookup(env, ss->ss_server_fld,
4098 fid_seq(pfid), range);
4099 if ((rc == -ENOENT) ||
4100 (rc == 0 && range->lsr_index != idx))
4101 rc = lfsck_namespace_trace_update(env, com, fid,
4102 LNTF_CHECK_LINKEA, true);
4104 rc = lfsck_namespace_check_for_double_scan(env, com,
4111 down_write(&com->lc_sem);
4112 if (S_ISDIR(lfsck_object_type(obj)))
4113 ns->ln_dirs_checked++;
4115 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
4116 up_write(&com->lc_sem);
4121 static int lfsck_namespace_exec_dir(const struct lu_env *env,
4122 struct lfsck_component *com,
4123 struct lu_dirent *ent, __u16 type)
4125 struct lfsck_assistant_data *lad = com->lc_data;
4126 struct lfsck_instance *lfsck = com->lc_lfsck;
4127 struct lfsck_namespace_req *lnr;
4128 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4129 struct ptlrpc_thread *mthread = &lfsck->li_thread;
4130 struct ptlrpc_thread *athread = &lad->lad_thread;
4131 struct l_wait_info lwi = { 0 };
4132 bool wakeup = false;
4134 l_wait_event(mthread->t_ctl_waitq,
4135 lad->lad_prefetched < bk->lb_async_windows ||
4136 !thread_is_running(mthread) ||
4137 thread_is_stopped(athread),
4140 if (unlikely(!thread_is_running(mthread)) ||
4141 thread_is_stopped(athread))
4144 if (unlikely(lfsck_is_dead_obj(lfsck->li_obj_dir)))
4147 lnr = lfsck_namespace_assistant_req_init(com->lc_lfsck, ent, type);
4149 struct lfsck_namespace *ns = com->lc_file_ram;
4151 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
4152 return PTR_ERR(lnr);
4155 spin_lock(&lad->lad_lock);
4156 if (lad->lad_assistant_status < 0) {
4157 spin_unlock(&lad->lad_lock);
4158 lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar);
4159 return lad->lad_assistant_status;
4162 list_add_tail(&lnr->lnr_lar.lar_list, &lad->lad_req_list);
4163 if (lad->lad_prefetched == 0)
4166 lad->lad_prefetched++;
4167 spin_unlock(&lad->lad_lock);
4169 wake_up_all(&lad->lad_thread.t_ctl_waitq);
4171 down_write(&com->lc_sem);
4172 com->lc_new_checked++;
4173 up_write(&com->lc_sem);
4178 static int lfsck_namespace_post(const struct lu_env *env,
4179 struct lfsck_component *com,
4180 int result, bool init)
4182 struct lfsck_instance *lfsck = com->lc_lfsck;
4183 struct lfsck_namespace *ns = com->lc_file_ram;
4187 lfsck_post_generic(env, com, &result);
4189 down_write(&com->lc_sem);
4190 lfsck_namespace_release_lmv(env, com);
4192 spin_lock(&lfsck->li_lock);
4194 ns->ln_pos_last_checkpoint = lfsck->li_pos_checkpoint;
4196 ns->ln_status = LS_SCANNING_PHASE2;
4197 ns->ln_flags |= LF_SCANNED_ONCE;
4198 ns->ln_flags &= ~LF_UPGRADE;
4199 list_del_init(&com->lc_link_dir);
4200 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
4201 } else if (result == 0) {
4202 if (lfsck->li_status != 0)
4203 ns->ln_status = lfsck->li_status;
4205 ns->ln_status = LS_STOPPED;
4206 if (ns->ln_status != LS_PAUSED) {
4207 list_del_init(&com->lc_link_dir);
4208 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4211 ns->ln_status = LS_FAILED;
4212 list_del_init(&com->lc_link_dir);
4213 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4215 spin_unlock(&lfsck->li_lock);
4218 ns->ln_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
4219 HALF_SEC - lfsck->li_time_last_checkpoint);
4220 ns->ln_time_last_checkpoint = cfs_time_current_sec();
4221 ns->ln_items_checked += com->lc_new_checked;
4222 com->lc_new_checked = 0;
4225 rc = lfsck_namespace_store(env, com, false);
4226 up_write(&com->lc_sem);
4228 CDEBUG(D_LFSCK, "%s: namespace LFSCK post done: rc = %d\n",
4229 lfsck_lfsck2name(lfsck), rc);
4235 lfsck_namespace_dump(const struct lu_env *env, struct lfsck_component *com,
4238 struct lfsck_instance *lfsck = com->lc_lfsck;
4239 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4240 struct lfsck_namespace *ns = com->lc_file_ram;
4243 down_read(&com->lc_sem);
4244 seq_printf(m, "name: lfsck_namespace\n"
4250 lfsck_status2names(ns->ln_status));
4252 rc = lfsck_bits_dump(m, ns->ln_flags, lfsck_flags_names, "flags");
4256 rc = lfsck_bits_dump(m, bk->lb_param, lfsck_param_names, "param");
4260 rc = lfsck_time_dump(m, ns->ln_time_last_complete,
4265 rc = lfsck_time_dump(m, ns->ln_time_latest_start,
4270 rc = lfsck_time_dump(m, ns->ln_time_last_checkpoint,
4275 rc = lfsck_pos_dump(m, &ns->ln_pos_latest_start,
4276 "latest_start_position");
4280 rc = lfsck_pos_dump(m, &ns->ln_pos_last_checkpoint,
4281 "last_checkpoint_position");
4285 rc = lfsck_pos_dump(m, &ns->ln_pos_first_inconsistent,
4286 "first_failure_position");
4290 if (ns->ln_status == LS_SCANNING_PHASE1) {
4291 struct lfsck_position pos;
4292 const struct dt_it_ops *iops;
4293 cfs_duration_t duration = cfs_time_current() -
4294 lfsck->li_time_last_checkpoint;
4295 __u64 checked = ns->ln_items_checked + com->lc_new_checked;
4296 __u64 speed = checked;
4297 __u64 new_checked = com->lc_new_checked *
4298 msecs_to_jiffies(MSEC_PER_SEC);
4299 __u32 rtime = ns->ln_run_time_phase1 +
4300 cfs_duration_sec(duration + HALF_SEC);
4303 do_div(new_checked, duration);
4306 do_div(speed, rtime);
4308 lfsck_namespace_dump_statistics(m, ns, checked, 0, rtime, 0);
4309 seq_printf(m, "average_speed_phase1: "LPU64" items/sec\n"
4310 "average_speed_phase2: N/A\n"
4311 "average_speed_total: "LPU64" items/sec\n"
4312 "real_time_speed_phase1: "LPU64" items/sec\n"
4313 "real_time_speed_phase2: N/A\n",
4318 LASSERT(lfsck->li_di_oit != NULL);
4320 iops = &lfsck->li_obj_oit->do_index_ops->dio_it;
4322 /* The low layer otable-based iteration position may NOT
4323 * exactly match the namespace-based directory traversal
4324 * cookie. Generally, it is not a serious issue. But the
4325 * caller should NOT make assumption on that. */
4326 pos.lp_oit_cookie = iops->store(env, lfsck->li_di_oit);
4327 if (!lfsck->li_current_oit_processed)
4328 pos.lp_oit_cookie--;
4330 spin_lock(&lfsck->li_lock);
4331 if (lfsck->li_di_dir != NULL) {
4332 pos.lp_dir_cookie = lfsck->li_cookie_dir;
4333 if (pos.lp_dir_cookie >= MDS_DIR_END_OFF) {
4334 fid_zero(&pos.lp_dir_parent);
4335 pos.lp_dir_cookie = 0;
4338 *lfsck_dto2fid(lfsck->li_obj_dir);
4341 fid_zero(&pos.lp_dir_parent);
4342 pos.lp_dir_cookie = 0;
4344 spin_unlock(&lfsck->li_lock);
4345 lfsck_pos_dump(m, &pos, "current_position");
4346 } else if (ns->ln_status == LS_SCANNING_PHASE2) {
4347 cfs_duration_t duration = cfs_time_current() -
4348 lfsck->li_time_last_checkpoint;
4349 __u64 checked = ns->ln_objs_checked_phase2 +
4350 com->lc_new_checked;
4351 __u64 speed1 = ns->ln_items_checked;
4352 __u64 speed2 = checked;
4353 __u64 speed0 = speed1 + speed2;
4354 __u64 new_checked = com->lc_new_checked *
4355 msecs_to_jiffies(MSEC_PER_SEC);
4356 __u32 rtime = ns->ln_run_time_phase2 +
4357 cfs_duration_sec(duration + HALF_SEC);
4358 __u32 time0 = ns->ln_run_time_phase1 + rtime;
4361 do_div(new_checked, duration);
4363 if (ns->ln_run_time_phase1 != 0)
4364 do_div(speed1, ns->ln_run_time_phase1);
4365 else if (ns->ln_items_checked != 0)
4369 do_div(speed2, rtime);
4370 else if (checked != 0)
4374 do_div(speed0, time0);
4376 lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked,
4378 ns->ln_run_time_phase1, rtime);
4379 seq_printf(m, "average_speed_phase1: "LPU64" items/sec\n"
4380 "average_speed_phase2: "LPU64" objs/sec\n"
4381 "average_speed_total: "LPU64" items/sec\n"
4382 "real_time_speed_phase1: N/A\n"
4383 "real_time_speed_phase2: "LPU64" objs/sec\n"
4384 "current_position: "DFID"\n",
4389 PFID(&ns->ln_fid_latest_scanned_phase2));
4391 __u64 speed1 = ns->ln_items_checked;
4392 __u64 speed2 = ns->ln_objs_checked_phase2;
4393 __u64 speed0 = speed1 + speed2;
4394 __u32 time0 = ns->ln_run_time_phase1 + ns->ln_run_time_phase2;
4396 if (ns->ln_run_time_phase1 != 0)
4397 do_div(speed1, ns->ln_run_time_phase1);
4398 else if (ns->ln_items_checked != 0)
4401 if (ns->ln_run_time_phase2 != 0)
4402 do_div(speed2, ns->ln_run_time_phase2);
4403 else if (ns->ln_objs_checked_phase2 != 0)
4407 do_div(speed0, time0);
4409 lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked,
4410 ns->ln_objs_checked_phase2,
4411 ns->ln_run_time_phase1,
4412 ns->ln_run_time_phase2);
4413 seq_printf(m, "average_speed_phase1: "LPU64" items/sec\n"
4414 "average_speed_phase2: "LPU64" objs/sec\n"
4415 "average_speed_total: "LPU64" items/sec\n"
4416 "real_time_speed_phase1: N/A\n"
4417 "real_time_speed_phase2: N/A\n"
4418 "current_position: N/A\n",
4424 up_read(&com->lc_sem);
4428 static int lfsck_namespace_double_scan(const struct lu_env *env,
4429 struct lfsck_component *com)
4431 struct lfsck_namespace *ns = com->lc_file_ram;
4432 struct lfsck_assistant_data *lad = com->lc_data;
4433 struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
4434 struct lfsck_tgt_desc *ltd;
4435 struct lfsck_tgt_desc *next;
4438 rc = lfsck_double_scan_generic(env, com, ns->ln_status);
4439 if (thread_is_stopped(&lad->lad_thread)) {
4440 LASSERT(list_empty(&lad->lad_req_list));
4441 LASSERT(list_empty(&lad->lad_mdt_phase1_list));
4443 spin_lock(<ds->ltd_lock);
4444 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4445 ltd_namespace_phase_list) {
4446 list_del_init(<d->ltd_namespace_phase_list);
4448 spin_unlock(<ds->ltd_lock);
4454 static void lfsck_namespace_data_release(const struct lu_env *env,
4455 struct lfsck_component *com)
4457 struct lfsck_assistant_data *lad = com->lc_data;
4458 struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
4459 struct lfsck_tgt_desc *ltd;
4460 struct lfsck_tgt_desc *next;
4462 LASSERT(lad != NULL);
4463 LASSERT(thread_is_init(&lad->lad_thread) ||
4464 thread_is_stopped(&lad->lad_thread));
4465 LASSERT(list_empty(&lad->lad_req_list));
4467 com->lc_data = NULL;
4468 lfsck_namespace_release_lmv(env, com);
4470 spin_lock(<ds->ltd_lock);
4471 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
4472 ltd_namespace_phase_list) {
4473 list_del_init(<d->ltd_namespace_phase_list);
4475 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4476 ltd_namespace_phase_list) {
4477 list_del_init(<d->ltd_namespace_phase_list);
4479 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_list,
4480 ltd_namespace_list) {
4481 list_del_init(<d->ltd_namespace_list);
4483 spin_unlock(<ds->ltd_lock);
4485 if (likely(lad->lad_bitmap != NULL))
4486 CFS_FREE_BITMAP(lad->lad_bitmap);
4491 static void lfsck_namespace_quit(const struct lu_env *env,
4492 struct lfsck_component *com)
4494 struct lfsck_assistant_data *lad = com->lc_data;
4495 struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
4496 struct lfsck_tgt_desc *ltd;
4497 struct lfsck_tgt_desc *next;
4499 LASSERT(lad != NULL);
4501 lfsck_quit_generic(env, com);
4503 LASSERT(thread_is_init(&lad->lad_thread) ||
4504 thread_is_stopped(&lad->lad_thread));
4505 LASSERT(list_empty(&lad->lad_req_list));
4507 lfsck_namespace_release_lmv(env, com);
4509 spin_lock(<ds->ltd_lock);
4510 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
4511 ltd_namespace_phase_list) {
4512 list_del_init(<d->ltd_namespace_phase_list);
4514 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4515 ltd_namespace_phase_list) {
4516 list_del_init(<d->ltd_namespace_phase_list);
4518 spin_unlock(<ds->ltd_lock);
4521 static int lfsck_namespace_in_notify(const struct lu_env *env,
4522 struct lfsck_component *com,
4523 struct lfsck_request *lr,
4526 struct lfsck_instance *lfsck = com->lc_lfsck;
4527 struct lfsck_namespace *ns = com->lc_file_ram;
4528 struct lfsck_assistant_data *lad = com->lc_data;
4529 struct lfsck_tgt_descs *ltds = &lfsck->li_mdt_descs;
4530 struct lfsck_tgt_desc *ltd;
4535 switch (lr->lr_event) {
4536 case LE_SKIP_NLINK_DECLARE: {
4537 struct dt_object *obj;
4538 struct lu_fid *key = &lfsck_env_info(env)->lti_fid3;
4542 LASSERT(th != NULL);
4544 idx = lfsck_sub_trace_file_fid2idx(&lr->lr_fid);
4545 obj = com->lc_sub_trace_objs[idx].lsto_obj;
4546 fid_cpu_to_be(key, &lr->lr_fid);
4547 mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
4548 rc = dt_declare_delete(env, obj,
4549 (const struct dt_key *)key, th);
4551 rc = dt_declare_insert(env, obj,
4552 (const struct dt_rec *)&flags,
4553 (const struct dt_key *)key, th);
4554 mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
4558 case LE_SKIP_NLINK: {
4559 struct dt_object *obj;
4560 struct lu_fid *key = &lfsck_env_info(env)->lti_fid3;
4566 LASSERT(th != NULL);
4568 idx = lfsck_sub_trace_file_fid2idx(&lr->lr_fid);
4569 obj = com->lc_sub_trace_objs[idx].lsto_obj;
4570 fid_cpu_to_be(key, &lr->lr_fid);
4571 mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
4572 rc = dt_lookup(env, obj, (struct dt_rec *)&flags,
4573 (const struct dt_key *)key);
4575 if (flags & LNTF_SKIP_NLINK) {
4577 &com->lc_sub_trace_objs[idx].lsto_mutex);
4583 } else if (rc != -ENOENT) {
4587 flags |= LNTF_SKIP_NLINK;
4589 rc = dt_delete(env, obj, (const struct dt_key *)key,
4595 rc = dt_insert(env, obj, (const struct dt_rec *)&flags,
4596 (const struct dt_key *)key, th, 1);
4601 mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
4602 CDEBUG(D_LFSCK, "%s: RPC service thread mark the "DFID
4603 " to be skipped for namespace double scan: rc = %d\n",
4604 lfsck_lfsck2name(com->lc_lfsck), PFID(&lr->lr_fid), rc);
4607 /* If we cannot record this object in the LFSCK tracing,
4608 * we have to mark the LFSC as LF_INCOMPLETE, then the
4609 * LFSCK will skip nlink attribute verification for
4611 ns->ln_flags |= LF_INCOMPLETE;
4615 case LE_SET_LMV_MASTER: {
4616 struct dt_object *obj;
4618 obj = lfsck_object_find_bottom(env, lfsck, &lr->lr_fid);
4620 RETURN(PTR_ERR(obj));
4622 rc = lfsck_namespace_notify_lmv_master_local(env, com, obj);
4623 lfsck_object_put(env, obj);
4625 RETURN(rc > 0 ? 0 : rc);
4627 case LE_SET_LMV_SLAVE: {
4628 if (!(lr->lr_flags & LEF_RECHECK_NAME_HASH))
4629 ns->ln_striped_shards_repaired++;
4631 rc = lfsck_namespace_trace_update(env, com, &lr->lr_fid,
4632 LNTF_RECHECK_NAME_HASH, true);
4634 RETURN(rc > 0 ? 0 : rc);
4636 case LE_PHASE1_DONE:
4637 case LE_PHASE2_DONE:
4644 CDEBUG(D_LFSCK, "%s: namespace LFSCK handles notify %u from MDT %x, "
4645 "status %d, flags %x\n", lfsck_lfsck2name(lfsck), lr->lr_event,
4646 lr->lr_index, lr->lr_status, lr->lr_flags2);
4648 spin_lock(<ds->ltd_lock);
4649 ltd = LTD_TGT(ltds, lr->lr_index);
4651 spin_unlock(<ds->ltd_lock);
4656 list_del_init(<d->ltd_namespace_phase_list);
4657 switch (lr->lr_event) {
4658 case LE_PHASE1_DONE:
4659 if (lr->lr_status <= 0) {
4660 ltd->ltd_namespace_done = 1;
4661 list_del_init(<d->ltd_namespace_list);
4662 CDEBUG(D_LFSCK, "%s: MDT %x failed/stopped at "
4663 "phase1 for namespace LFSCK: rc = %d.\n",
4664 lfsck_lfsck2name(lfsck),
4665 ltd->ltd_index, lr->lr_status);
4666 ns->ln_flags |= LF_INCOMPLETE;
4671 if (lr->lr_flags2 & LF_INCOMPLETE)
4672 ns->ln_flags |= LF_INCOMPLETE;
4674 if (list_empty(<d->ltd_namespace_list))
4675 list_add_tail(<d->ltd_namespace_list,
4676 &lad->lad_mdt_list);
4677 list_add_tail(<d->ltd_namespace_phase_list,
4678 &lad->lad_mdt_phase2_list);
4680 case LE_PHASE2_DONE:
4681 ltd->ltd_namespace_done = 1;
4682 list_del_init(<d->ltd_namespace_list);
4686 ltd->ltd_namespace_done = 1;
4687 list_del_init(<d->ltd_namespace_list);
4688 if (!(lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT)) {
4690 "%s: the peer MDT %x exit namespace LFSCK\n",
4691 lfsck_lfsck2name(lfsck), ltd->ltd_index);
4692 ns->ln_flags |= LF_INCOMPLETE;
4698 spin_unlock(<ds->ltd_lock);
4700 if (fail && lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) {
4701 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
4703 memset(stop, 0, sizeof(*stop));
4704 stop->ls_status = lr->lr_status;
4705 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
4706 lfsck_stop(env, lfsck->li_bottom, stop);
4707 } else if (lfsck_phase2_next_ready(lad)) {
4708 wake_up_all(&lad->lad_thread.t_ctl_waitq);
4714 static int lfsck_namespace_query(const struct lu_env *env,
4715 struct lfsck_component *com)
4717 struct lfsck_namespace *ns = com->lc_file_ram;
4719 return ns->ln_status;
4722 static struct lfsck_operations lfsck_namespace_ops = {
4723 .lfsck_reset = lfsck_namespace_reset,
4724 .lfsck_fail = lfsck_namespace_fail,
4725 .lfsck_close_dir = lfsck_namespace_close_dir,
4726 .lfsck_open_dir = lfsck_namespace_open_dir,
4727 .lfsck_checkpoint = lfsck_namespace_checkpoint,
4728 .lfsck_prep = lfsck_namespace_prep,
4729 .lfsck_exec_oit = lfsck_namespace_exec_oit,
4730 .lfsck_exec_dir = lfsck_namespace_exec_dir,
4731 .lfsck_post = lfsck_namespace_post,
4732 .lfsck_dump = lfsck_namespace_dump,
4733 .lfsck_double_scan = lfsck_namespace_double_scan,
4734 .lfsck_data_release = lfsck_namespace_data_release,
4735 .lfsck_quit = lfsck_namespace_quit,
4736 .lfsck_in_notify = lfsck_namespace_in_notify,
4737 .lfsck_query = lfsck_namespace_query,
4741 * Repair dangling name entry.
4743 * For the name entry with dangling reference, we need to repare the
4744 * inconsistency according to the LFSCK sponsor's requirement:
4746 * 1) Keep the inconsistency there and report the inconsistency case,
4747 * then give the chance to the application to find related issues,
4748 * and the users can make the decision about how to handle it with
4749 * more human knownledge. (by default)
4751 * 2) Re-create the missing MDT-object with the FID information.
4753 * \param[in] env pointer to the thread context
4754 * \param[in] com pointer to the lfsck component
4755 * \param[in] parent pointer to the dir object that contains the dangling
4757 * \param[in] child pointer to the object corresponding to the dangling
4759 * \param[in] lnr pointer to the namespace request that contains the
4760 * name's name, parent object, parent's LMV, and ect.
4762 * \retval positive number if no need to repair
4763 * \retval zero for repaired successfully
4764 * \retval negative error number on failure
4766 int lfsck_namespace_repair_dangling(const struct lu_env *env,
4767 struct lfsck_component *com,
4768 struct dt_object *parent,
4769 struct dt_object *child,
4770 struct lfsck_namespace_req *lnr)
4772 struct lfsck_thread_info *info = lfsck_env_info(env);
4773 struct lu_attr *la = &info->lti_la;
4774 struct dt_allocation_hint *hint = &info->lti_hint;
4775 struct dt_object_format *dof = &info->lti_dof;
4776 struct dt_insert_rec *rec = &info->lti_dt_rec;
4777 struct lmv_mds_md_v1 *lmv2 = &info->lti_lmv2;
4778 struct dt_object *pobj = NULL;
4779 struct dt_object *cobj = NULL;
4780 const struct lu_name *cname;
4781 struct linkea_data ldata = { NULL };
4782 struct lfsck_lock_handle *llh = &info->lti_llh;
4783 struct lu_buf linkea_buf;
4784 struct lu_buf lmv_buf;
4785 struct lfsck_instance *lfsck = com->lc_lfsck;
4786 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4787 struct dt_device *dev = lfsck->li_next;
4788 struct thandle *th = NULL;
4790 __u16 type = lnr->lnr_type;
4794 cname = lfsck_name_get_const(env, lnr->lnr_name, lnr->lnr_namelen);
4795 if (bk->lb_param & LPF_CREATE_MDTOBJ)
4800 if (!create || bk->lb_param & LPF_DRYRUN)
4803 /* We may need to create the sub-objects of the @child via LOD,
4804 * so make the modification based on lfsck->li_next. */
4806 pobj = lfsck_object_locate(dev, parent);
4808 GOTO(log, rc = PTR_ERR(pobj));
4810 if (unlikely(!dt_try_as_dir(env, pobj)))
4811 GOTO(log, rc = -ENOTDIR);
4813 cobj = lfsck_object_locate(dev, child);
4815 GOTO(log, rc = PTR_ERR(cobj));
4817 rc = linkea_data_new(&ldata, &info->lti_linkea_buf2);
4821 rc = linkea_add_buf(&ldata, cname, lfsck_dto2fid(pobj));
4825 rc = lfsck_lock(env, lfsck, parent, lnr->lnr_name, llh,
4826 MDS_INODELOCK_UPDATE, LCK_PR);
4830 rc = lfsck_namespace_check_exist(env, pobj, cobj, lnr->lnr_name);
4834 /* Set the ctime as zero, then others can know it is created for
4835 * repairing dangling name entry by LFSCK. And if the LFSCK made
4836 * wrong decision and the real MDT-object has been found later,
4837 * then the LFSCK has chance to fix the incosistency properly. */
4838 memset(la, 0, sizeof(*la));
4839 la->la_mode = (type & S_IFMT) | 0600;
4840 la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
4841 LA_ATIME | LA_MTIME | LA_CTIME;
4843 cobj->do_ops->do_ah_init(env, hint, pobj, cobj,
4844 la->la_mode & S_IFMT);
4846 memset(dof, 0, sizeof(*dof));
4847 dof->dof_type = dt_mode_to_dft(type);
4848 /* If the target is a regular file, then the LFSCK will only create
4849 * the MDT-object without stripes (dof->dof_reg.striped = 0). related
4850 * OST-objects will be created when write open. */
4852 th = dt_trans_create(env, dev);
4854 GOTO(log, rc = PTR_ERR(th));
4856 /* 1a. create child. */
4857 rc = dt_declare_create(env, cobj, la, hint, dof, th);
4861 if (S_ISDIR(type)) {
4862 if (unlikely(!dt_try_as_dir(env, cobj)))
4863 GOTO(stop, rc = -ENOTDIR);
4865 /* 2a. insert dot into child dir */
4866 rec->rec_type = S_IFDIR;
4867 rec->rec_fid = lfsck_dto2fid(cobj);
4868 rc = dt_declare_insert(env, cobj,
4869 (const struct dt_rec *)rec,
4870 (const struct dt_key *)dot, th);
4874 /* 3a. insert dotdot into child dir */
4875 rec->rec_fid = lfsck_dto2fid(pobj);
4876 rc = dt_declare_insert(env, cobj,
4877 (const struct dt_rec *)rec,
4878 (const struct dt_key *)dotdot, th);
4882 /* 4a. increase child nlink */
4883 rc = dt_declare_ref_add(env, cobj, th);
4887 /* 5a. generate slave LMV EA. */
4888 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
4891 idx = lfsck_shard_name_to_index(env,
4892 lnr->lnr_name, lnr->lnr_namelen,
4893 type, lfsck_dto2fid(cobj));
4894 if (unlikely(idx < 0))
4895 GOTO(stop, rc = idx);
4897 *lmv2 = lnr->lnr_lmv->ll_lmv;
4898 lmv2->lmv_magic = LMV_MAGIC_STRIPE;
4899 lmv2->lmv_master_mdt_index = idx;
4901 lfsck_lmv_header_cpu_to_le(lmv2, lmv2);
4902 lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
4903 rc = dt_declare_xattr_set(env, cobj, &lmv_buf,
4904 XATTR_NAME_LMV, 0, th);
4910 /* 6a. insert linkEA for child */
4911 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
4912 ldata.ld_leh->leh_len);
4913 rc = dt_declare_xattr_set(env, cobj, &linkea_buf,
4914 XATTR_NAME_LINK, 0, th);
4918 rc = dt_trans_start_local(env, dev, th);
4920 GOTO(stop, rc = (rc == -EEXIST ? 1 : rc));
4922 dt_write_lock(env, cobj, 0);
4923 /* 1b. create child */
4924 rc = dt_create(env, cobj, la, hint, dof, th);
4926 GOTO(unlock, rc = (rc == -EEXIST ? 1 : rc));
4928 if (S_ISDIR(type)) {
4929 /* 2b. insert dot into child dir */
4930 rec->rec_type = S_IFDIR;
4931 rec->rec_fid = lfsck_dto2fid(cobj);
4932 rc = dt_insert(env, cobj, (const struct dt_rec *)rec,
4933 (const struct dt_key *)dot, th, 1);
4937 /* 3b. insert dotdot into child dir */
4938 rec->rec_fid = lfsck_dto2fid(pobj);
4939 rc = dt_insert(env, cobj, (const struct dt_rec *)rec,
4940 (const struct dt_key *)dotdot, th, 1);
4944 /* 4b. increase child nlink */
4945 rc = dt_ref_add(env, cobj, th);
4949 /* 5b. generate slave LMV EA. */
4950 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
4951 rc = dt_xattr_set(env, cobj, &lmv_buf, XATTR_NAME_LMV,
4958 /* 6b. insert linkEA for child. */
4959 rc = dt_xattr_set(env, cobj, &linkea_buf,
4960 XATTR_NAME_LINK, 0, th);
4965 dt_write_unlock(env, cobj);
4968 dt_trans_stop(env, dev, th);
4972 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found dangling "
4973 "reference for: parent "DFID", child "DFID", type %u, "
4974 "name %s. %s: rc = %d\n", lfsck_lfsck2name(lfsck),
4975 PFID(&lnr->lnr_lar.lar_fid), PFID(lfsck_dto2fid(child)),
4976 type, cname->ln_name,
4977 create ? "Create the lost OST-object as required" :
4978 "Keep the MDT-object there by default", rc);
4981 struct lfsck_namespace *ns = com->lc_file_ram;
4983 ns->ln_flags |= LF_INCONSISTENT;
4989 static int lfsck_namespace_assistant_handler_p1(const struct lu_env *env,
4990 struct lfsck_component *com,
4991 struct lfsck_assistant_req *lar)
4993 struct lfsck_thread_info *info = lfsck_env_info(env);
4994 struct lu_attr *la = &info->lti_la;
4995 struct lfsck_instance *lfsck = com->lc_lfsck;
4996 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4997 struct lfsck_namespace *ns = com->lc_file_ram;
4998 struct linkea_data ldata = { NULL };
4999 const struct lu_name *cname;
5000 struct thandle *handle = NULL;
5001 struct lfsck_namespace_req *lnr =
5002 container_of0(lar, struct lfsck_namespace_req, lnr_lar);
5003 struct dt_object *dir = NULL;
5004 struct dt_object *obj = NULL;
5005 const struct lu_fid *pfid;
5006 struct dt_device *dev = NULL;
5007 struct lustre_handle lh = { 0 };
5008 bool repaired = false;
5009 bool dtlocked = false;
5013 bool bad_hash = false;
5017 enum lfsck_namespace_inconsistency_type type = LNIT_NONE;
5020 dir = lfsck_object_find_bottom(env, lfsck, &lar->lar_fid);
5022 RETURN(PTR_ERR(dir));
5024 if (unlikely(lfsck_is_dead_obj(dir)))
5025 GOTO(put_dir, rc = 0);
5027 if (unlikely(!dt_try_as_dir(env, dir)))
5028 GOTO(put_dir, rc = -ENOTDIR);
5030 pfid = lfsck_dto2fid(dir);
5032 if (lnr->lnr_attr & LUDA_UPGRADE) {
5033 ns->ln_flags |= LF_UPGRADE;
5034 ns->ln_dirent_repaired++;
5036 } else if (lnr->lnr_attr & LUDA_REPAIR) {
5037 ns->ln_flags |= LF_INCONSISTENT;
5038 ns->ln_dirent_repaired++;
5042 if (unlikely(fid_is_zero(&lnr->lnr_fid))) {
5043 if (strcmp(lnr->lnr_name, dotdot) != 0)
5046 rc = lfsck_namespace_trace_update(env, com, pfid,
5047 LNTF_CHECK_PARENT, true);
5052 if (unlikely(!fid_is_sane(&lnr->lnr_fid))) {
5053 CDEBUG(D_LFSCK, "%s: dir scan find invalid FID "DFID
5054 " for the name entry %.*s under "DFID"\n",
5055 lfsck_lfsck2name(lfsck), PFID(&lnr->lnr_fid),
5056 lnr->lnr_namelen, lnr->lnr_name, PFID(pfid));
5058 if (strcmp(lnr->lnr_name, dotdot) != 0)
5059 /* invalid FID means bad name entry, remove it. */
5060 type = LNIT_BAD_DIRENT;
5062 /* If the parent FID is invalid, we cannot remove
5063 * the ".." entry directly. */
5064 rc = lfsck_namespace_trace_update(env, com, pfid,
5065 LNTF_CHECK_PARENT, true);
5070 if (unlikely(lnr->lnr_dir_cookie == MDS_DIR_END_OFF)) {
5071 rc = lfsck_namespace_striped_dir_rescan(env, com, dir, lnr);
5076 if (fid_seq_is_dot(fid_seq(&lnr->lnr_fid)))
5079 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
5080 rc = lfsck_namespace_handle_striped_master(env, com, dir, lnr);
5085 idx = lfsck_find_mdt_idx_by_fid(env, lfsck, &lnr->lnr_fid);
5087 GOTO(out, rc = idx);
5089 if (idx == lfsck_dev_idx(lfsck)) {
5090 if (unlikely(strcmp(lnr->lnr_name, dotdot) == 0))
5093 dev = lfsck->li_bottom;
5095 struct lfsck_tgt_desc *ltd;
5097 /* Usually, some local filesystem consistency verification
5098 * tools can guarantee the local namespace tree consistenct.
5099 * So the LFSCK will only verify the remote directory. */
5100 if (unlikely(strcmp(lnr->lnr_name, dotdot) == 0)) {
5101 rc = lfsck_namespace_trace_update(env, com, pfid,
5102 LNTF_CHECK_PARENT, true);
5107 ltd = LTD_TGT(&lfsck->li_mdt_descs, idx);
5108 if (unlikely(ltd == NULL)) {
5109 CDEBUG(D_LFSCK, "%s: cannot talk with MDT %x which "
5110 "did not join the namespace LFSCK\n",
5111 lfsck_lfsck2name(lfsck), idx);
5112 lfsck_lad_set_bitmap(env, com, idx);
5114 GOTO(out, rc = -ENODEV);
5120 obj = lfsck_object_find_by_dev(env, dev, &lnr->lnr_fid);
5122 GOTO(out, rc = PTR_ERR(obj));
5124 cname = lfsck_name_get_const(env, lnr->lnr_name, lnr->lnr_namelen);
5125 if (dt_object_exists(obj) == 0) {
5128 rc = lfsck_namespace_check_exist(env, dir, obj, lnr->lnr_name);
5130 if (!lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5131 lnr->lnr_name, lnr->lnr_namelen)) {
5132 type = LNIT_BAD_DIRENT;
5137 type = LNIT_DANGLING;
5138 rc = lfsck_namespace_repair_dangling(env, com, dir,
5147 if (!(bk->lb_param & LPF_DRYRUN) && repaired) {
5150 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
5151 MDS_INODELOCK_UPDATE |
5152 MDS_INODELOCK_XATTR, LCK_EX);
5156 handle = dt_trans_create(env, dev);
5158 GOTO(out, rc = PTR_ERR(handle));
5160 rc = lfsck_declare_namespace_exec_dir(env, obj, handle);
5164 rc = dt_trans_start_local(env, dev, handle);
5168 dt_write_lock(env, obj, 0);
5172 rc = lfsck_namespace_check_exist(env, dir, obj, lnr->lnr_name);
5176 rc = lfsck_links_read(env, obj, &ldata);
5177 if (unlikely(rc == -ENOENT)) {
5178 if (handle != NULL) {
5179 dt_write_unlock(env, obj);
5182 dt_trans_stop(env, dev, handle);
5185 lfsck_ibits_unlock(&lh, LCK_EX);
5188 /* It may happen when the remote object has been removed,
5189 * but the local MDT is not aware of that. */
5191 } else if (rc == 0) {
5192 count = ldata.ld_leh->leh_reccount;
5193 rc = linkea_links_find(&ldata, cname, pfid);
5195 (count == 1 || !S_ISDIR(lfsck_object_type(obj)))) {
5196 if ((lfsck_object_type(obj) & S_IFMT) !=
5198 ns->ln_flags |= LF_INCONSISTENT;
5199 type = LNIT_BAD_TYPE;
5205 ns->ln_flags |= LF_INCONSISTENT;
5207 /* If the name entry hash does not match the slave striped
5208 * directory, and the name entry does not match also, then
5209 * it is quite possible that name entry is corrupted. */
5210 if (!lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5211 lnr->lnr_name, lnr->lnr_namelen)) {
5212 type = LNIT_BAD_DIRENT;
5217 /* If the file type stored in the name entry does not match
5218 * the file type claimed by the object, and the object does
5219 * not recognize the name entry, then it is quite possible
5220 * that the name entry is corrupted. */
5221 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type) {
5222 type = LNIT_BAD_DIRENT;
5227 /* For sub-dir object, we cannot make sure whether the sub-dir
5228 * back references the parent via ".." name entry correctly or
5229 * not in the LFSCK first-stage scanning. It may be that the
5230 * (remote) sub-dir ".." name entry has no parent FID after
5231 * file-level backup/restore and its linkEA may be wrong.
5232 * So under such case, we should replace the linkEA according
5233 * to current name entry. But this needs to be done during the
5234 * LFSCK second-stage scanning. The LFSCK will record the name
5235 * entry for further possible using. */
5239 } else if (unlikely(rc == -EINVAL)) {
5240 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type)
5241 type = LNIT_BAD_TYPE;
5244 ns->ln_flags |= LF_INCONSISTENT;
5245 /* The magic crashed, we are not sure whether there are more
5246 * corrupt data in the linkea, so remove all linkea entries. */
5250 } else if (rc == -ENODATA) {
5251 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type)
5252 type = LNIT_BAD_TYPE;
5255 ns->ln_flags |= LF_UPGRADE;
5260 if (bk->lb_param & LPF_DRYRUN) {
5261 ns->ln_linkea_repaired++;
5267 if (!lustre_handle_is_used(&lh))
5273 rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, handle);
5279 rc = linkea_data_new(&ldata,
5280 &lfsck_env_info(env)->lti_linkea_buf);
5285 rc = linkea_add_buf(&ldata, cname, pfid);
5289 rc = lfsck_links_write(env, obj, &ldata, handle);
5290 if (unlikely(rc == -ENOSPC) &&
5291 S_ISREG(lfsck_object_type(obj)) && !dt_object_remote(obj)) {
5292 if (handle != NULL) {
5293 LASSERT(dt_write_locked(env, obj));
5295 dt_write_unlock(env, obj);
5298 dt_trans_stop(env, dev, handle);
5301 lfsck_ibits_unlock(&lh, LCK_EX);
5304 rc = lfsck_namespace_trace_update(env, com,
5305 &lnr->lnr_fid, LNTF_SKIP_NLINK, true);
5307 /* If we cannot record this object in the
5308 * LFSCK tracing, we have to mark the LFSCK
5309 * as LF_INCOMPLETE, then the LFSCK will
5310 * skip nlink attribute verification for
5312 ns->ln_flags |= LF_INCOMPLETE;
5320 count = ldata.ld_leh->leh_reccount;
5321 if (!S_ISDIR(lfsck_object_type(obj)) ||
5322 !dt_object_remote(obj)) {
5323 ns->ln_linkea_repaired++;
5333 dt_write_unlock(env, obj);
5335 if (handle != NULL && !IS_ERR(handle))
5336 dt_trans_stop(env, dev, handle);
5339 lfsck_ibits_unlock(&lh, LCK_EX);
5341 if (!name_is_dot_or_dotdot(lnr->lnr_name, lnr->lnr_namelen) &&
5342 !lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5343 lnr->lnr_name, lnr->lnr_namelen) &&
5344 type != LNIT_BAD_DIRENT) {
5345 ns->ln_flags |= LF_INCONSISTENT;
5348 rc = lfsck_namespace_repair_bad_name_hash(env, com, dir,
5349 lnr->lnr_lmv, lnr->lnr_name);
5358 rc = lfsck_namespace_repair_dirent(env, com, dir,
5359 obj, lnr->lnr_name, lnr->lnr_name,
5360 lnr->lnr_type, true, false);
5364 case LNIT_BAD_DIRENT:
5366 /* XXX: This is a bad dirent, we do not know whether
5367 * the original name entry reference a regular
5368 * file or a directory, then keep the parent's
5369 * nlink count unchanged here. */
5370 rc = lfsck_namespace_repair_dirent(env, com, dir,
5371 obj, lnr->lnr_name, lnr->lnr_name,
5372 lnr->lnr_type, false, false);
5380 if (count == 1 && S_ISREG(lfsck_object_type(obj)))
5381 dt_attr_get(env, obj, la);
5384 down_write(&com->lc_sem);
5386 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail to handle "
5387 "the entry: "DFID", parent "DFID", name %.*s: rc = %d\n",
5388 lfsck_lfsck2name(lfsck), PFID(&lnr->lnr_fid),
5389 PFID(lfsck_dto2fid(dir)),
5390 lnr->lnr_namelen, lnr->lnr_name, rc);
5392 lfsck_namespace_record_failure(env, lfsck, ns);
5393 if ((rc == -ENOTCONN || rc == -ESHUTDOWN || rc == -EREMCHG ||
5394 rc == -ETIMEDOUT || rc == -EHOSTDOWN ||
5395 rc == -EHOSTUNREACH || rc == -EINPROGRESS) &&
5396 dev != NULL && dev != lfsck->li_bottom)
5397 lfsck_lad_set_bitmap(env, com, idx);
5399 if (!(bk->lb_param & LPF_FAILOUT))
5403 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant "
5404 "repaired the entry: "DFID", parent "DFID
5405 ", name %.*s\n", lfsck_lfsck2name(lfsck),
5406 PFID(&lnr->lnr_fid),
5407 PFID(lfsck_dto2fid(dir)),
5408 lnr->lnr_namelen, lnr->lnr_name);
5411 ns->ln_items_repaired++;
5415 ns->ln_dangling_repaired++;
5418 ns->ln_bad_type_repaired++;
5420 case LNIT_BAD_DIRENT:
5421 ns->ln_dirent_repaired++;
5427 if (bk->lb_param & LPF_DRYRUN &&
5428 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent))
5429 lfsck_pos_fill(env, lfsck,
5430 &ns->ln_pos_first_inconsistent,
5435 ns->ln_name_hash_repaired++;
5437 /* Not count repeatedly. */
5439 ns->ln_items_repaired++;
5441 if (bk->lb_param & LPF_DRYRUN &&
5442 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent))
5443 lfsck_pos_fill(env, lfsck,
5444 &ns->ln_pos_first_inconsistent,
5451 if (count > 1 || la->la_nlink > 1)
5452 ns->ln_mul_linked_checked++;
5454 up_write(&com->lc_sem);
5456 if (obj != NULL && !IS_ERR(obj))
5457 lfsck_object_put(env, obj);
5460 lfsck_object_put(env, dir);
5466 * Handle one orphan under the backend /lost+found directory
5468 * Insert the orphan FID into the namespace LFSCK trace file for further
5469 * processing (via the subsequent namespace LFSCK second-stage scanning).
5470 * At the same time, remove the orphan name entry from backend /lost+found
5471 * directory. There is an interval between the orphan name entry removed
5472 * from the backend /lost+found directory and the orphan FID in the LFSCK
5473 * trace file handled. In such interval, the LFSCK can be reset, then
5474 * all the FIDs recorded in the namespace LFSCK trace file will be dropped.
5475 * To guarantee that the orphans can be found when LFSCK run next time
5476 * without e2fsck again, when remove the orphan name entry, the LFSCK
5477 * will set the orphan's ctime attribute as 1. Since normal applications
5478 * cannot change the object's ctime attribute as 1. Then when LFSCK run
5479 * next time, it can record the object (that ctime is 1) in the namespace
5480 * LFSCK trace file during the first-stage scanning.
5482 * \param[in] env pointer to the thread context
5483 * \param[in] com pointer to the lfsck component
5484 * \param[in] parent pointer to the object for the backend /lost+found
5485 * \param[in] ent pointer to the name entry for the target under the
5486 * backend /lost+found
5488 * \retval positive for repaired
5489 * \retval 0 if needs to repair nothing
5490 * \retval negative error number on failure
5492 static int lfsck_namespace_scan_local_lpf_one(const struct lu_env *env,
5493 struct lfsck_component *com,
5494 struct dt_object *parent,
5495 struct lu_dirent *ent)
5497 struct lfsck_thread_info *info = lfsck_env_info(env);
5498 struct lu_fid *key = &info->lti_fid;
5499 struct lu_attr *la = &info->lti_la;
5500 struct lfsck_instance *lfsck = com->lc_lfsck;
5501 struct dt_object *obj;
5502 struct dt_device *dev = lfsck->li_bottom;
5503 struct dt_object *child = NULL;
5504 struct thandle *th = NULL;
5511 child = lfsck_object_find_by_dev(env, dev, &ent->lde_fid);
5513 RETURN(PTR_ERR(child));
5515 LASSERT(dt_object_exists(child));
5516 LASSERT(!dt_object_remote(child));
5518 idx = lfsck_sub_trace_file_fid2idx(&ent->lde_fid);
5519 obj = com->lc_sub_trace_objs[idx].lsto_obj;
5520 fid_cpu_to_be(key, &ent->lde_fid);
5521 rc = dt_lookup(env, obj, (struct dt_rec *)&flags,
5522 (const struct dt_key *)key);
5525 flags |= LNTF_CHECK_ORPHAN;
5526 } else if (rc == -ENOENT) {
5527 flags = LNTF_CHECK_ORPHAN;
5532 th = dt_trans_create(env, dev);
5534 GOTO(out, rc = PTR_ERR(th));
5536 /* a1. remove name entry from backend /lost+found */
5537 rc = dt_declare_delete(env, parent,
5538 (const struct dt_key *)ent->lde_name, th);
5542 if (S_ISDIR(lfsck_object_type(child))) {
5543 /* a2. decrease parent's nlink */
5544 rc = dt_declare_ref_del(env, parent, th);
5550 /* a3. remove child's FID from the LFSCK trace file. */
5551 rc = dt_declare_delete(env, obj,
5552 (const struct dt_key *)key, th);
5556 /* a4. set child's ctime as 1 */
5557 memset(la, 0, sizeof(*la));
5559 la->la_valid = LA_CTIME;
5560 rc = dt_declare_attr_set(env, child, la, th);
5565 /* a5. insert child's FID into the LFSCK trace file. */
5566 rc = dt_declare_insert(env, obj, (const struct dt_rec *)&flags,
5567 (const struct dt_key *)key, th);
5571 rc = dt_trans_start_local(env, dev, th);
5575 /* b1. remove name entry from backend /lost+found */
5576 rc = dt_delete(env, parent, (const struct dt_key *)ent->lde_name, th);
5580 if (S_ISDIR(lfsck_object_type(child))) {
5581 /* b2. decrease parent's nlink */
5582 dt_write_lock(env, parent, 0);
5583 rc = dt_ref_del(env, parent, th);
5584 dt_write_unlock(env, parent);
5590 /* a3. remove child's FID from the LFSCK trace file. */
5591 rc = dt_delete(env, obj, (const struct dt_key *)key, th);
5595 /* b4. set child's ctime as 1 */
5596 rc = dt_attr_set(env, child, la, th);
5601 /* b5. insert child's FID into the LFSCK trace file. */
5602 rc = dt_insert(env, obj, (const struct dt_rec *)&flags,
5603 (const struct dt_key *)key, th, 1);
5605 GOTO(stop, rc = (rc == 0 ? 1 : rc));
5608 dt_trans_stop(env, dev, th);
5611 lfsck_object_put(env, child);
5617 * Handle orphans under the backend /lost+found directory
5619 * Some backend checker, such as e2fsck for ldiskfs may find some orphans
5620 * and put them under the backend /lost+found directory that is invisible
5621 * to client. The LFSCK will scan such directory, for the original client
5622 * visible orphans, add their fids into the namespace LFSCK trace file,
5623 * then the subsenquent namespace LFSCK second-stage scanning can handle
5624 * them as other objects to be double scanned: either move back to normal
5625 * namespace, or to the global visible orphan directory:
5626 * /ROOT/.lustre/lost+found/MDTxxxx/
5628 * \param[in] env pointer to the thread context
5629 * \param[in] com pointer to the lfsck component
5631 static void lfsck_namespace_scan_local_lpf(const struct lu_env *env,
5632 struct lfsck_component *com)
5634 struct lfsck_thread_info *info = lfsck_env_info(env);
5635 struct lu_dirent *ent =
5636 (struct lu_dirent *)info->lti_key;
5637 struct lu_seq_range *range = &info->lti_range;
5638 struct lfsck_instance *lfsck = com->lc_lfsck;
5639 struct ptlrpc_thread *thread = &lfsck->li_thread;
5640 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5641 struct lfsck_namespace *ns = com->lc_file_ram;
5642 struct dt_object *parent;
5643 const struct dt_it_ops *iops;
5645 struct seq_server_site *ss = lfsck_dev_site(lfsck);
5647 __u32 idx = lfsck_dev_idx(lfsck);
5652 parent = lfsck_object_find_by_dev(env, lfsck->li_bottom,
5653 &LU_BACKEND_LPF_FID);
5654 if (IS_ERR(parent)) {
5655 CERROR("%s: fail to find backend /lost+found: rc = %ld\n",
5656 lfsck_lfsck2name(lfsck), PTR_ERR(parent));
5660 /* It is normal that the /lost+found does not exist for ZFS backend. */
5661 if (!dt_object_exists(parent))
5664 if (unlikely(!dt_try_as_dir(env, parent)))
5665 GOTO(out, rc = -ENOTDIR);
5667 CDEBUG(D_LFSCK, "%s: start to scan backend /lost+found\n",
5668 lfsck_lfsck2name(lfsck));
5670 com->lc_new_scanned = 0;
5671 iops = &parent->do_index_ops->dio_it;
5672 di = iops->init(env, parent, LUDA_64BITHASH | LUDA_TYPE);
5674 GOTO(out, rc = PTR_ERR(di));
5676 rc = iops->load(env, di, 0);
5678 rc = iops->next(env, di);
5683 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
5684 unlikely(!thread_is_running(thread)))
5687 rc = iops->rec(env, di, (struct dt_rec *)ent,
5688 LUDA_64BITHASH | LUDA_TYPE);
5690 rc = lfsck_unpack_ent(ent, &cookie, &type);
5692 if (unlikely(rc != 0)) {
5693 CDEBUG(D_LFSCK, "%s: fail to iterate backend "
5694 "/lost+found: rc = %d\n",
5695 lfsck_lfsck2name(lfsck), rc);
5700 /* skip dot and dotdot entries */
5701 if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen))
5704 if (!fid_seq_in_fldb(fid_seq(&ent->lde_fid)))
5707 if (fid_is_norm(&ent->lde_fid)) {
5708 fld_range_set_mdt(range);
5709 rc = fld_local_lookup(env, ss->ss_server_fld,
5710 fid_seq(&ent->lde_fid), range);
5713 } else if (idx != 0) {
5714 /* If the returned FID is IGIF, then there are three
5717 * 1) The object is upgraded from old Lustre-1.8 with
5718 * IGIF assigned to such object.
5719 * 2) The object is a backend local object and is
5720 * invisible to client.
5721 * 3) The object lost its LMV EA, and since there is
5722 * no FID-in-dirent for the orphan in the backend
5723 * /lost+found directory, then the low layer will
5724 * return IGIF for such object.
5726 * For MDTx (x != 0), it is either case 2) or case 3),
5727 * but from the LFSCK view, they are indistinguishable.
5728 * To be safe, the LFSCK will keep it there and report
5729 * some message, then the adminstrator can handle that
5732 * For MDT0, it is more possible the case 1). The LFSCK
5733 * will handle the orphan as an upgraded object. */
5734 CDEBUG(D_LFSCK, "%s: the orphan %.*s with IGIF "DFID
5735 "in the backend /lost+found on the MDT %04x, "
5736 "to be safe, skip it.\n",
5737 lfsck_lfsck2name(lfsck), ent->lde_namelen,
5738 ent->lde_name, PFID(&ent->lde_fid), idx);
5742 rc = lfsck_namespace_scan_local_lpf_one(env, com, parent, ent);
5745 down_write(&com->lc_sem);
5746 com->lc_new_scanned++;
5747 ns->ln_local_lpf_scanned++;
5749 ns->ln_local_lpf_moved++;
5751 ns->ln_local_lpf_skipped++;
5753 ns->ln_local_lpf_failed++;
5754 up_write(&com->lc_sem);
5756 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
5760 lfsck_control_speed_by_self(com);
5761 if (unlikely(!thread_is_running(thread))) {
5766 rc = iops->next(env, di);
5770 iops->fini(env, di);
5775 CDEBUG(D_LFSCK, "%s: stop to scan backend /lost+found: rc = %d\n",
5776 lfsck_lfsck2name(lfsck), rc);
5778 lfsck_object_put(env, parent);
5782 * Rescan the striped directory after the master LMV EA reset.
5784 * Sometimes, the master LMV EA of the striped directory maybe lost, so when
5785 * the namespace LFSCK engine scan the striped directory for the first time,
5786 * it will be regarded as a normal directory. As the LFSCK processing, some
5787 * other LFSCK instance on other MDT will find the shard of this striped dir,
5788 * and find that the master MDT-object of the striped directory lost its LMV
5789 * EA, then such remote LFSCK instance will regenerate the master LMV EA and
5790 * notify the LFSCK instance on this MDT to rescan the striped directory.
5792 * \param[in] env pointer to the thread context
5793 * \param[in] com pointer to the lfsck component
5794 * \param[in] llu the lfsck_lmv_unit that contains the striped directory
5797 * \retval positive number for success
5798 * \retval 0 for LFSCK stopped/paused
5799 * \retval negative error number on failure
5801 static int lfsck_namespace_rescan_striped_dir(const struct lu_env *env,
5802 struct lfsck_component *com,
5803 struct lfsck_lmv_unit *llu)
5805 struct lfsck_thread_info *info = lfsck_env_info(env);
5806 struct lfsck_instance *lfsck = com->lc_lfsck;
5807 struct lfsck_assistant_data *lad = com->lc_data;
5808 struct dt_object *dir;
5809 const struct dt_it_ops *iops;
5811 struct lu_dirent *ent =
5812 (struct lu_dirent *)info->lti_key;
5813 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5814 struct ptlrpc_thread *thread = &lfsck->li_thread;
5815 struct lfsck_namespace_req *lnr;
5816 struct lfsck_assistant_req *lar;
5821 LASSERT(list_empty(&lad->lad_req_list));
5823 lfsck->li_lmv = &llu->llu_lmv;
5824 lfsck->li_obj_dir = lfsck_object_get(llu->llu_obj);
5825 rc = lfsck_open_dir(env, lfsck, 0);
5829 dir = lfsck->li_obj_dir;
5830 di = lfsck->li_di_dir;
5831 iops = &dir->do_index_ops->dio_it;
5833 rc = iops->rec(env, di, (struct dt_rec *)ent,
5834 lfsck->li_args_dir);
5836 rc = lfsck_unpack_ent(ent, &lfsck->li_cookie_dir,
5840 if (bk->lb_param & LPF_FAILOUT)
5846 if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen))
5849 lnr = lfsck_namespace_assistant_req_init(lfsck, ent, type);
5851 if (bk->lb_param & LPF_FAILOUT)
5852 GOTO(out, rc = PTR_ERR(lnr));
5857 lar = &lnr->lnr_lar;
5858 rc = lfsck_namespace_assistant_handler_p1(env, com, lar);
5859 lfsck_namespace_assistant_req_fini(env, lar);
5860 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
5863 if (unlikely(!thread_is_running(thread)))
5867 rc = iops->next(env, di);
5871 lfsck_close_dir(env, lfsck, rc);
5875 /* The close_dir() may insert a dummy lnr in the lad->lad_req_list. */
5876 if (list_empty(&lad->lad_req_list))
5879 spin_lock(&lad->lad_lock);
5880 lar = list_entry(lad->lad_req_list.next, struct lfsck_assistant_req,
5882 list_del_init(&lar->lar_list);
5883 spin_unlock(&lad->lad_lock);
5885 rc = lfsck_namespace_assistant_handler_p1(env, com, lar);
5886 lfsck_namespace_assistant_req_fini(env, lar);
5888 RETURN(rc == 0 ? 1 : rc);
5892 lfsck_namespace_double_scan_one_trace_file(const struct lu_env *env,
5893 struct lfsck_component *com,
5894 struct dt_object *obj, bool first)
5896 struct lfsck_instance *lfsck = com->lc_lfsck;
5897 struct ptlrpc_thread *thread = &lfsck->li_thread;
5898 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5899 struct lfsck_namespace *ns = com->lc_file_ram;
5900 const struct dt_it_ops *iops = &obj->do_index_ops->dio_it;
5901 struct dt_object *target;
5909 di = iops->init(env, obj, 0);
5911 RETURN(PTR_ERR(di));
5914 fid_cpu_to_be(&fid, &ns->ln_fid_latest_scanned_phase2);
5917 rc = iops->get(env, di, (const struct dt_key *)&fid);
5922 /* The start one either has been processed or does not exist,
5924 rc = iops->next(env, di);
5930 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
5931 unlikely(!thread_is_running(thread)))
5934 key = iops->key(env, di);
5943 fid_be_to_cpu(&fid, (const struct lu_fid *)key);
5944 if (!fid_is_sane(&fid)) {
5949 target = lfsck_object_find_bottom(env, lfsck, &fid);
5950 if (IS_ERR(target)) {
5951 rc = PTR_ERR(target);
5955 if (dt_object_exists(target)) {
5956 rc = iops->rec(env, di, (struct dt_rec *)&flags, 0);
5958 rc = lfsck_namespace_double_scan_one(env, com,
5965 lfsck_object_put(env, target);
5968 down_write(&com->lc_sem);
5969 com->lc_new_checked++;
5970 com->lc_new_scanned++;
5971 if (rc >= 0 && fid_is_sane(&fid))
5972 ns->ln_fid_latest_scanned_phase2 = fid;
5974 ns->ln_objs_repaired_phase2++;
5976 ns->ln_objs_failed_phase2++;
5977 up_write(&com->lc_sem);
5979 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
5982 if (unlikely(cfs_time_beforeq(com->lc_time_next_checkpoint,
5983 cfs_time_current())) &&
5984 com->lc_new_checked != 0) {
5985 down_write(&com->lc_sem);
5986 ns->ln_run_time_phase2 +=
5987 cfs_duration_sec(cfs_time_current() +
5988 HALF_SEC - com->lc_time_last_checkpoint);
5989 ns->ln_time_last_checkpoint = cfs_time_current_sec();
5990 ns->ln_objs_checked_phase2 += com->lc_new_checked;
5991 com->lc_new_checked = 0;
5992 rc = lfsck_namespace_store(env, com, false);
5993 up_write(&com->lc_sem);
5997 com->lc_time_last_checkpoint = cfs_time_current();
5998 com->lc_time_next_checkpoint =
5999 com->lc_time_last_checkpoint +
6000 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
6003 lfsck_control_speed_by_self(com);
6004 if (unlikely(!thread_is_running(thread)))
6007 rc = iops->next(env, di);
6016 iops->fini(env, di);
6021 static int lfsck_namespace_assistant_handler_p2(const struct lu_env *env,
6022 struct lfsck_component *com)
6024 struct lfsck_instance *lfsck = com->lc_lfsck;
6025 struct lfsck_namespace *ns = com->lc_file_ram;
6030 while (!list_empty(&lfsck->li_list_lmv)) {
6031 struct lfsck_lmv_unit *llu;
6033 spin_lock(&lfsck->li_lock);
6034 llu = list_entry(lfsck->li_list_lmv.next,
6035 struct lfsck_lmv_unit, llu_link);
6036 list_del_init(&llu->llu_link);
6037 spin_unlock(&lfsck->li_lock);
6039 rc = lfsck_namespace_rescan_striped_dir(env, com, llu);
6044 CDEBUG(D_LFSCK, "%s: namespace LFSCK phase2 scan start\n",
6045 lfsck_lfsck2name(lfsck));
6047 lfsck_namespace_scan_local_lpf(env, com);
6049 com->lc_new_checked = 0;
6050 com->lc_new_scanned = 0;
6051 com->lc_time_last_checkpoint = cfs_time_current();
6052 com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
6053 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
6055 i = lfsck_sub_trace_file_fid2idx(&ns->ln_fid_latest_scanned_phase2);
6056 rc = lfsck_namespace_double_scan_one_trace_file(env, com,
6057 com->lc_sub_trace_objs[i].lsto_obj, true);
6058 while (rc > 0 && ++i < LFSCK_STF_COUNT)
6059 rc = lfsck_namespace_double_scan_one_trace_file(env, com,
6060 com->lc_sub_trace_objs[i].lsto_obj, false);
6062 CDEBUG(D_LFSCK, "%s: namespace LFSCK phase2 scan stop at the No. %d "
6063 "trace file: rc = %d\n", lfsck_lfsck2name(lfsck), i, rc);
6068 static void lfsck_namespace_assistant_fill_pos(const struct lu_env *env,
6069 struct lfsck_component *com,
6070 struct lfsck_position *pos)
6072 struct lfsck_assistant_data *lad = com->lc_data;
6073 struct lfsck_namespace_req *lnr;
6075 if (list_empty(&lad->lad_req_list))
6078 lnr = list_entry(lad->lad_req_list.next,
6079 struct lfsck_namespace_req,
6081 pos->lp_oit_cookie = lnr->lnr_oit_cookie;
6082 pos->lp_dir_cookie = lnr->lnr_dir_cookie - 1;
6083 pos->lp_dir_parent = lnr->lnr_lar.lar_fid;
6086 static int lfsck_namespace_double_scan_result(const struct lu_env *env,
6087 struct lfsck_component *com,
6090 struct lfsck_instance *lfsck = com->lc_lfsck;
6091 struct lfsck_namespace *ns = com->lc_file_ram;
6093 down_write(&com->lc_sem);
6094 ns->ln_run_time_phase2 += cfs_duration_sec(cfs_time_current() +
6095 HALF_SEC - lfsck->li_time_last_checkpoint);
6096 ns->ln_time_last_checkpoint = cfs_time_current_sec();
6097 ns->ln_objs_checked_phase2 += com->lc_new_checked;
6098 com->lc_new_checked = 0;
6101 if (ns->ln_flags & LF_INCOMPLETE)
6102 ns->ln_status = LS_PARTIAL;
6104 ns->ln_status = LS_COMPLETED;
6105 if (!(lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN))
6106 ns->ln_flags &= ~(LF_SCANNED_ONCE | LF_INCONSISTENT);
6107 ns->ln_time_last_complete = ns->ln_time_last_checkpoint;
6108 ns->ln_success_count++;
6109 } else if (rc == 0) {
6110 if (lfsck->li_status != 0)
6111 ns->ln_status = lfsck->li_status;
6113 ns->ln_status = LS_STOPPED;
6115 ns->ln_status = LS_FAILED;
6118 rc = lfsck_namespace_store(env, com, false);
6119 up_write(&com->lc_sem);
6125 lfsck_namespace_assistant_sync_failures_interpret(const struct lu_env *env,
6126 struct ptlrpc_request *req,
6130 struct lfsck_async_interpret_args *laia = args;
6131 struct lfsck_tgt_desc *ltd = laia->laia_ltd;
6133 ltd->ltd_synced_failures = 1;
6140 * Notify remote LFSCK instances about former failures.
6142 * The local LFSCK instance has recorded which MDTs have ever failed to respond
6143 * some LFSCK verification requests (maybe because of network issues or the MDT
6144 * itself trouble). During the respond gap the MDT may missed some name entries
6145 * verification, then the MDT cannot know whether related MDT-objects have been
6146 * referenced by related name entries or not, then in the second-stage scanning,
6147 * these MDT-objects will be regarded as orphan, if the MDT-object contains bad
6148 * linkEA for back reference, then it will misguide the LFSCK to generate wrong
6149 * name entry for repairing the orphan.
6151 * To avoid above trouble, when layout LFSCK finishes the first-stage scanning,
6152 * it will scan the bitmap for the ever failed MDTs, and notify them that they
6153 * have ever missed some name entries verification and should skip the handling
6154 * for orphan MDT-objects.
6156 * \param[in] env pointer to the thread context
6157 * \param[in] com pointer to the lfsck component
6158 * \param[in] lr pointer to the lfsck request
6160 static void lfsck_namespace_assistant_sync_failures(const struct lu_env *env,
6161 struct lfsck_component *com,
6162 struct lfsck_request *lr)
6164 struct lfsck_async_interpret_args *laia =
6165 &lfsck_env_info(env)->lti_laia2;
6166 struct lfsck_assistant_data *lad = com->lc_data;
6167 struct lfsck_namespace *ns = com->lc_file_ram;
6168 struct lfsck_instance *lfsck = com->lc_lfsck;
6169 struct lfsck_tgt_descs *ltds = &lfsck->li_mdt_descs;
6170 struct lfsck_tgt_desc *ltd;
6171 struct ptlrpc_request_set *set;
6176 if (!lad->lad_incomplete)
6179 set = ptlrpc_prep_set();
6181 GOTO(out, rc = -ENOMEM);
6183 lr->lr_flags2 = ns->ln_flags | LF_INCOMPLETE;
6184 memset(laia, 0, sizeof(*laia));
6185 lad->lad_touch_gen++;
6187 down_read(<ds->ltd_rw_sem);
6188 cfs_foreach_bit(lad->lad_bitmap, idx) {
6189 ltd = LTD_TGT(ltds, idx);
6190 LASSERT(ltd != NULL);
6192 laia->laia_ltd = ltd;
6193 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
6194 lfsck_namespace_assistant_sync_failures_interpret,
6195 laia, LFSCK_NOTIFY);
6197 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail "
6198 "to sync failure with MDT %x: rc = %d\n",
6199 lfsck_lfsck2name(lfsck), ltd->ltd_index, rc);
6201 up_read(<ds->ltd_rw_sem);
6203 rc = ptlrpc_set_wait(set);
6204 ptlrpc_set_destroy(set);
6210 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail "
6211 "to sync failure with MDTs, and related MDTs "
6212 "may handle orphan improperly: rc = %d\n",
6213 lfsck_lfsck2name(lfsck), rc);
6218 struct lfsck_assistant_operations lfsck_namespace_assistant_ops = {
6219 .la_handler_p1 = lfsck_namespace_assistant_handler_p1,
6220 .la_handler_p2 = lfsck_namespace_assistant_handler_p2,
6221 .la_fill_pos = lfsck_namespace_assistant_fill_pos,
6222 .la_double_scan_result = lfsck_namespace_double_scan_result,
6223 .la_req_fini = lfsck_namespace_assistant_req_fini,
6224 .la_sync_failures = lfsck_namespace_assistant_sync_failures,
6228 * Verify the specified linkEA entry for the given directory object.
6229 * If the object has no such linkEA entry or it has more other linkEA
6230 * entries, then re-generate the linkEA with the given information.
6232 * \param[in] env pointer to the thread context
6233 * \param[in] obj pointer to the dt_object to be handled
6234 * \param[in] cname the name for the child in the parent directory
6235 * \param[in] pfid the parent directory's FID for the linkEA
6237 * \retval 0 for success
6238 * \retval negative error number on failure
6240 int lfsck_verify_linkea(const struct lu_env *env, struct dt_object *obj,
6241 const struct lu_name *cname, const struct lu_fid *pfid)
6243 struct dt_device *dev = lfsck_obj2dev(obj);
6244 struct linkea_data ldata = { NULL };
6245 struct lu_buf linkea_buf;
6248 int fl = LU_XATTR_CREATE;
6252 LASSERT(S_ISDIR(lfsck_object_type(obj)));
6254 rc = lfsck_links_read(env, obj, &ldata);
6255 if (rc == -ENODATA) {
6257 } else if (rc == 0) {
6258 fl = LU_XATTR_REPLACE;
6259 if (ldata.ld_leh->leh_reccount != 1) {
6262 rc = linkea_links_find(&ldata, cname, pfid);
6271 rc = linkea_data_new(&ldata, &lfsck_env_info(env)->lti_linkea_buf);
6275 rc = linkea_add_buf(&ldata, cname, pfid);
6279 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
6280 ldata.ld_leh->leh_len);
6281 th = dt_trans_create(env, dev);
6283 RETURN(PTR_ERR(th));
6285 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
6286 XATTR_NAME_LINK, fl, th);
6290 rc = dt_trans_start_local(env, dev, th);
6294 dt_write_lock(env, obj, 0);
6295 rc = dt_xattr_set(env, obj, &linkea_buf,
6296 XATTR_NAME_LINK, fl, th);
6297 dt_write_unlock(env, obj);
6302 dt_trans_stop(env, dev, th);
6307 * Get the name and parent directory's FID from the first linkEA entry.
6309 * \param[in] env pointer to the thread context
6310 * \param[in] obj pointer to the object which get linkEA from
6311 * \param[out] name pointer to the buffer to hold the name
6312 * in the first linkEA entry
6313 * \param[out] pfid pointer to the buffer to hold the parent
6314 * directory's FID in the first linkEA entry
6316 * \retval 0 for success
6317 * \retval negative error number on failure
6319 int lfsck_links_get_first(const struct lu_env *env, struct dt_object *obj,
6320 char *name, struct lu_fid *pfid)
6322 struct lu_name *cname = &lfsck_env_info(env)->lti_name;
6323 struct linkea_data ldata = { NULL };
6326 rc = lfsck_links_read(env, obj, &ldata);
6330 linkea_first_entry(&ldata);
6331 if (ldata.ld_lee == NULL)
6334 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, cname, pfid);
6335 /* To guarantee the 'name' is terminated with '0'. */
6336 memcpy(name, cname->ln_name, cname->ln_namelen);
6337 name[cname->ln_namelen] = 0;
6343 * Update the object's name entry with the given FID.
6345 * \param[in] env pointer to the thread context
6346 * \param[in] lfsck pointer to the lfsck instance
6347 * \param[in] dir pointer to the directory that holds
6349 * \param[in] name the name for the entry to be updated
6350 * \param[in] fid the new FID for the name entry referenced
6351 * \param[in] type the type for the name entry to be updated
6353 * \retval 0 for success
6354 * \retval negative error number on failure
6356 int lfsck_update_name_entry(const struct lu_env *env,
6357 struct lfsck_instance *lfsck,
6358 struct dt_object *dir, const char *name,
6359 const struct lu_fid *fid, __u32 type)
6361 struct lfsck_thread_info *info = lfsck_env_info(env);
6362 struct dt_insert_rec *rec = &info->lti_dt_rec;
6363 struct lfsck_lock_handle *llh = &info->lti_llh;
6364 struct dt_device *dev = lfsck_obj2dev(dir);
6370 rc = lfsck_lock(env, lfsck, dir, name, llh,
6371 MDS_INODELOCK_UPDATE, LCK_PW);
6375 th = dt_trans_create(env, dev);
6377 GOTO(unlock, rc = PTR_ERR(th));
6379 rc = dt_declare_delete(env, dir, (const struct dt_key *)name, th);
6383 rec->rec_type = type;
6385 rc = dt_declare_insert(env, dir, (const struct dt_rec *)rec,
6386 (const struct dt_key *)name, th);
6390 rc = dt_declare_ref_add(env, dir, th);
6394 rc = dt_trans_start_local(env, dev, th);
6398 rc = dt_delete(env, dir, (const struct dt_key *)name, th);
6399 if (rc == -ENOENT) {
6407 rc = dt_insert(env, dir, (const struct dt_rec *)rec,
6408 (const struct dt_key *)name, th, 1);
6409 if (rc == 0 && S_ISDIR(type) && !exists) {
6410 dt_write_lock(env, dir, 0);
6411 rc = dt_ref_add(env, dir, th);
6412 dt_write_unlock(env, dir);
6418 dt_trans_stop(env, dev, th);
6422 CDEBUG(D_LFSCK, "%s: update name entry "DFID"/%s with the FID "DFID
6423 " and the type %o: rc = %d\n", lfsck_lfsck2name(lfsck),
6424 PFID(lfsck_dto2fid(dir)), name, PFID(fid), type, rc);
6429 int lfsck_namespace_setup(const struct lu_env *env,
6430 struct lfsck_instance *lfsck)
6432 struct lfsck_component *com;
6433 struct lfsck_namespace *ns;
6434 struct dt_object *root = NULL;
6435 struct dt_object *obj;
6440 LASSERT(lfsck->li_master);
6446 INIT_LIST_HEAD(&com->lc_link);
6447 INIT_LIST_HEAD(&com->lc_link_dir);
6448 init_rwsem(&com->lc_sem);
6449 atomic_set(&com->lc_ref, 1);
6450 com->lc_lfsck = lfsck;
6451 com->lc_type = LFSCK_TYPE_NAMESPACE;
6452 com->lc_ops = &lfsck_namespace_ops;
6453 com->lc_data = lfsck_assistant_data_init(
6454 &lfsck_namespace_assistant_ops,
6456 if (com->lc_data == NULL)
6457 GOTO(out, rc = -ENOMEM);
6459 com->lc_file_size = sizeof(struct lfsck_namespace);
6460 OBD_ALLOC(com->lc_file_ram, com->lc_file_size);
6461 if (com->lc_file_ram == NULL)
6462 GOTO(out, rc = -ENOMEM);
6464 OBD_ALLOC(com->lc_file_disk, com->lc_file_size);
6465 if (com->lc_file_disk == NULL)
6466 GOTO(out, rc = -ENOMEM);
6468 for (i = 0; i < LFSCK_STF_COUNT; i++)
6469 mutex_init(&com->lc_sub_trace_objs[i].lsto_mutex);
6471 root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
6473 GOTO(out, rc = PTR_ERR(root));
6475 if (unlikely(!dt_try_as_dir(env, root)))
6476 GOTO(out, rc = -ENOTDIR);
6478 obj = local_index_find_or_create(env, lfsck->li_los, root,
6480 S_IFREG | S_IRUGO | S_IWUSR,
6481 &dt_lfsck_features);
6483 GOTO(out, rc = PTR_ERR(obj));
6486 rc = lfsck_namespace_load(env, com);
6488 rc = lfsck_namespace_init(env, com);
6490 rc = lfsck_namespace_reset(env, com, true);
6492 rc = lfsck_namespace_load_sub_trace_files(env, com, false);
6496 ns = com->lc_file_ram;
6497 switch (ns->ln_status) {
6502 spin_lock(&lfsck->li_lock);
6503 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
6504 spin_unlock(&lfsck->li_lock);
6507 CERROR("%s: unknown lfsck_namespace status %d\n",
6508 lfsck_lfsck2name(lfsck), ns->ln_status);
6510 case LS_SCANNING_PHASE1:
6511 case LS_SCANNING_PHASE2:
6512 /* No need to store the status to disk right now.
6513 * If the system crashed before the status stored,
6514 * it will be loaded back when next time. */
6515 ns->ln_status = LS_CRASHED;
6519 spin_lock(&lfsck->li_lock);
6520 list_add_tail(&com->lc_link, &lfsck->li_list_scan);
6521 list_add_tail(&com->lc_link_dir, &lfsck->li_list_dir);
6522 spin_unlock(&lfsck->li_lock);
6529 if (root != NULL && !IS_ERR(root))
6530 lfsck_object_put(env, root);
6532 lfsck_component_cleanup(env, com);
6533 CERROR("%s: fail to init namespace LFSCK component: rc = %d\n",
6534 lfsck_lfsck2name(lfsck), rc);