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, BYPASS_CAPA);
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, BYPASS_CAPA);
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, BYPASS_CAPA);
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, BYPASS_CAPA);
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,
456 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 8, 53, 0)
458 rc = dt_xattr_set(env, obj, &tbuf,
459 XATTR_NAME_LFSCK_NAMESPACE_OLD,
460 LU_XATTR_CREATE, handle, BYPASS_CAPA);
466 dt_trans_stop(env, dev, handle);
470 CDEBUG(D_LFSCK, "%s: fail to store lfsck_namespace: rc = %d\n",
471 lfsck_lfsck2name(lfsck), rc);
475 static struct dt_object *
476 lfsck_namespace_load_one_trace_file(const struct lu_env *env,
477 struct lfsck_component *com,
478 struct dt_object *parent,
479 const char *name, bool reset)
481 struct lfsck_instance *lfsck = com->lc_lfsck;
482 struct dt_object *obj;
486 rc = local_object_unlink(env, lfsck->li_bottom, parent, name);
487 if (rc != 0 && rc != -ENOENT)
491 obj = local_index_find_or_create(env, lfsck->li_los, parent, name,
492 S_IFREG | S_IRUGO | S_IWUSR,
498 static int lfsck_namespace_load_sub_trace_files(const struct lu_env *env,
499 struct lfsck_component *com,
502 char *name = lfsck_env_info(env)->lti_key;
503 struct lfsck_sub_trace_obj *lsto;
504 struct dt_object *obj;
508 for (i = 0, lsto = &com->lc_sub_trace_objs[0];
509 i < LFSCK_STF_COUNT; i++, lsto++) {
510 snprintf(name, NAME_MAX, "%s_%02d", LFSCK_NAMESPACE, i);
511 if (lsto->lsto_obj != NULL) {
515 lfsck_object_put(env, lsto->lsto_obj);
516 lsto->lsto_obj = NULL;
519 obj = lfsck_namespace_load_one_trace_file(env, com,
520 com->lc_lfsck->li_lfsck_dir, name, reset);
524 lsto->lsto_obj = obj;
525 rc = obj->do_ops->do_index_try(env, obj, &dt_lfsck_features);
533 static int lfsck_namespace_init(const struct lu_env *env,
534 struct lfsck_component *com)
536 struct lfsck_namespace *ns = com->lc_file_ram;
539 memset(ns, 0, sizeof(*ns));
540 ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
541 ns->ln_status = LS_INIT;
542 down_write(&com->lc_sem);
543 rc = lfsck_namespace_store(env, com, true);
544 up_write(&com->lc_sem);
546 rc = lfsck_namespace_load_sub_trace_files(env, com, true);
552 * Update the namespace LFSCK trace file for the given @fid
554 * \param[in] env pointer to the thread context
555 * \param[in] com pointer to the lfsck component
556 * \param[in] fid the fid which flags to be updated in the lfsck
558 * \param[in] add true if add new flags, otherwise remove flags
560 * \retval 0 for succeed or nothing to be done
561 * \retval negative error number on failure
563 int lfsck_namespace_trace_update(const struct lu_env *env,
564 struct lfsck_component *com,
565 const struct lu_fid *fid,
566 const __u8 flags, bool add)
568 struct lfsck_instance *lfsck = com->lc_lfsck;
569 struct dt_object *obj;
570 struct lu_fid *key = &lfsck_env_info(env)->lti_fid3;
571 struct dt_device *dev;
572 struct thandle *th = NULL;
581 if (unlikely(!fid_is_sane(fid)))
584 idx = lfsck_sub_trace_file_fid2idx(fid);
585 obj = com->lc_sub_trace_objs[idx].lsto_obj;
586 dev = lfsck_obj2dev(obj);
587 mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
588 fid_cpu_to_be(key, fid);
589 rc = dt_lookup(env, obj, (struct dt_rec *)&old,
590 (const struct dt_key *)key, BYPASS_CAPA);
593 GOTO(unlock, rc = 0);
597 } else if (rc == 0) {
599 if ((old & flags) == flags)
600 GOTO(unlock, rc = 0);
604 if ((old & flags) == 0)
605 GOTO(unlock, rc = 0);
613 th = dt_trans_create(env, dev);
615 GOTO(log, rc = PTR_ERR(th));
618 rc = dt_declare_delete(env, obj,
619 (const struct dt_key *)key, th);
625 rc = dt_declare_insert(env, obj,
626 (const struct dt_rec *)&new,
627 (const struct dt_key *)key, th);
632 rc = dt_trans_start_local(env, dev, th);
637 rc = dt_delete(env, obj, (const struct dt_key *)key,
644 rc = dt_insert(env, obj, (const struct dt_rec *)&new,
645 (const struct dt_key *)key, th, BYPASS_CAPA, 1);
653 if (th != NULL && !IS_ERR(th))
654 dt_trans_stop(env, dev, th);
656 CDEBUG(D_LFSCK, "%s: namespace LFSCK %s flags for "DFID" in the "
657 "trace file, flags %x, old %x, new %x: rc = %d\n",
658 lfsck_lfsck2name(lfsck), add ? "add" : "del", PFID(fid),
659 (__u32)flags, (__u32)old, (__u32)new, rc);
662 mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
667 int lfsck_namespace_check_exist(const struct lu_env *env,
668 struct dt_object *dir,
669 struct dt_object *obj, const char *name)
671 struct lu_fid *fid = &lfsck_env_info(env)->lti_fid;
675 if (unlikely(lfsck_is_dead_obj(obj)))
676 RETURN(LFSCK_NAMEENTRY_DEAD);
678 rc = dt_lookup(env, dir, (struct dt_rec *)fid,
679 (const struct dt_key *)name, BYPASS_CAPA);
681 RETURN(LFSCK_NAMEENTRY_REMOVED);
686 if (!lu_fid_eq(fid, lfsck_dto2fid(obj)))
687 RETURN(LFSCK_NAMEENTRY_RECREATED);
692 static int lfsck_declare_namespace_exec_dir(const struct lu_env *env,
693 struct dt_object *obj,
694 struct thandle *handle)
698 /* For destroying all invalid linkEA entries. */
699 rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, handle);
703 /* For insert new linkEA entry. */
704 rc = dt_declare_xattr_set(env, obj,
705 lfsck_buf_get_const(env, NULL, DEFAULT_LINKEA_SIZE),
706 XATTR_NAME_LINK, 0, handle);
710 int __lfsck_links_read(const struct lu_env *env, struct dt_object *obj,
711 struct linkea_data *ldata)
715 if (ldata->ld_buf->lb_buf == NULL)
718 if (!dt_object_exists(obj))
721 rc = dt_xattr_get(env, obj, ldata->ld_buf, XATTR_NAME_LINK, BYPASS_CAPA);
723 /* Buf was too small, figure out what we need. */
724 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LINK,
729 lu_buf_realloc(ldata->ld_buf, rc);
730 if (ldata->ld_buf->lb_buf == NULL)
733 rc = dt_xattr_get(env, obj, ldata->ld_buf, XATTR_NAME_LINK,
738 rc = linkea_init(ldata);
744 * Remove linkEA for the given object.
746 * The caller should take the ldlm lock before the calling.
748 * \param[in] env pointer to the thread context
749 * \param[in] com pointer to the lfsck component
750 * \param[in] obj pointer to the dt_object to be handled
752 * \retval 0 for repaired cases
753 * \retval negative error number on failure
755 static int lfsck_namespace_links_remove(const struct lu_env *env,
756 struct lfsck_component *com,
757 struct dt_object *obj)
759 struct lfsck_instance *lfsck = com->lc_lfsck;
760 struct dt_device *dev = lfsck_obj2dev(obj);
761 struct thandle *th = NULL;
765 LASSERT(dt_object_remote(obj) == 0);
767 th = dt_trans_create(env, dev);
769 GOTO(log, rc = PTR_ERR(th));
771 rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, th);
775 rc = dt_trans_start_local(env, dev, th);
779 dt_write_lock(env, obj, 0);
780 if (unlikely(lfsck_is_dead_obj(obj)))
781 GOTO(unlock, rc = -ENOENT);
783 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
784 GOTO(unlock, rc = 0);
786 rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, th, BYPASS_CAPA);
791 dt_write_unlock(env, obj);
794 dt_trans_stop(env, dev, th);
797 CDEBUG(D_LFSCK, "%s: namespace LFSCK remove invalid linkEA "
798 "for the object "DFID": rc = %d\n",
799 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
802 struct lfsck_namespace *ns = com->lc_file_ram;
804 ns->ln_flags |= LF_INCONSISTENT;
810 static int lfsck_links_write(const struct lu_env *env, struct dt_object *obj,
811 struct linkea_data *ldata, struct thandle *handle)
813 const struct lu_buf *buf = lfsck_buf_get_const(env,
814 ldata->ld_buf->lb_buf,
815 ldata->ld_leh->leh_len);
817 return dt_xattr_set(env, obj, buf, XATTR_NAME_LINK, 0, handle,
821 static void lfsck_namespace_unpack_linkea_entry(struct linkea_data *ldata,
822 struct lu_name *cname,
826 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, cname, pfid);
827 /* To guarantee the 'name' is terminated with '0'. */
828 memcpy(buf, cname->ln_name, cname->ln_namelen);
829 buf[cname->ln_namelen] = 0;
830 cname->ln_name = buf;
833 static int lfsck_namespace_filter_linkea_entry(struct linkea_data *ldata,
834 struct lu_name *cname,
838 struct link_ea_entry *oldlee;
842 oldlee = ldata->ld_lee;
843 oldlen = ldata->ld_reclen;
844 linkea_next_entry(ldata);
845 while (ldata->ld_lee != NULL) {
846 ldata->ld_reclen = (ldata->ld_lee->lee_reclen[0] << 8) |
847 ldata->ld_lee->lee_reclen[1];
848 if (unlikely(ldata->ld_reclen == oldlen &&
849 memcmp(ldata->ld_lee, oldlee, oldlen) == 0)) {
854 linkea_del_buf(ldata, cname);
856 linkea_next_entry(ldata);
859 ldata->ld_lee = oldlee;
860 ldata->ld_reclen = oldlen;
866 * Insert orphan into .lustre/lost+found/MDTxxxx/ locally.
868 * Add the specified orphan MDT-object to the .lustre/lost+found/MDTxxxx/
869 * with the given type to generate the name, the detailed rules for name
870 * have been described as following.
872 * The function also generates the linkEA corresponding to the name entry
873 * under the .lustre/lost+found/MDTxxxx/ for the orphan MDT-object.
875 * \param[in] env pointer to the thread context
876 * \param[in] com pointer to the lfsck component
877 * \param[in] orphan pointer to the orphan MDT-object
878 * \param[in] infix additional information for the orphan name, such as
879 * the FID for original
880 * \param[in] type the type for describing why the orphan MDT-object is
881 * created. The rules are as following:
883 * type "D": The MDT-object is a directory, it may knows its parent
884 * but because there is no valid linkEA, the LFSCK cannot
885 * know where to put it back to the namespace.
886 * type "O": The MDT-object has no linkEA, and there is no name
887 * entry that references the MDT-object.
889 * type "S": The orphan MDT-object is a shard of a striped directory
891 * \see lfsck_layout_recreate_parent() for more types.
893 * The orphan name will be like:
894 * ${FID}-${infix}-${type}-${conflict_version}
896 * \param[out] count if some others inserted some linkEA entries by race,
897 * then return the linkEA entries count.
899 * \retval positive number for repaired cases
900 * \retval 0 if needs to repair nothing
901 * \retval negative error number on failure
903 static int lfsck_namespace_insert_orphan(const struct lu_env *env,
904 struct lfsck_component *com,
905 struct dt_object *orphan,
906 const char *infix, const char *type,
909 struct lfsck_thread_info *info = lfsck_env_info(env);
910 struct lu_name *cname = &info->lti_name;
911 struct dt_insert_rec *rec = &info->lti_dt_rec;
912 struct lu_attr *la = &info->lti_la3;
913 const struct lu_fid *cfid = lfsck_dto2fid(orphan);
914 const struct lu_fid *pfid;
916 struct lfsck_instance *lfsck = com->lc_lfsck;
917 struct dt_device *dev = lfsck_obj2dev(orphan);
918 struct dt_object *parent;
919 struct thandle *th = NULL;
920 struct lfsck_lock_handle *pllh = &info->lti_llh;
921 struct lustre_handle clh = { 0 };
922 struct linkea_data ldata = { NULL };
923 struct lu_buf linkea_buf;
930 cname->ln_name = NULL;
931 if (unlikely(lfsck->li_lpf_obj == NULL))
932 GOTO(log, rc = -ENXIO);
934 parent = lfsck->li_lpf_obj;
935 pfid = lfsck_dto2fid(parent);
939 namelen = snprintf(info->lti_key, NAME_MAX, DFID"%s-%s-%d",
940 PFID(cfid), infix, type, idx++);
941 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
942 (const struct dt_key *)info->lti_key,
944 if (rc != 0 && rc != -ENOENT)
947 if (unlikely(rc == 0 && lu_fid_eq(cfid, &tfid)))
949 } while (rc == 0 && !exist);
951 rc = lfsck_lock(env, lfsck, parent, info->lti_key, pllh,
952 MDS_INODELOCK_UPDATE, LCK_PW);
956 /* Re-check whether the name conflict with othrs after taken
958 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
959 (const struct dt_key *)info->lti_key, BYPASS_CAPA);
961 if (!lu_fid_eq(cfid, &tfid)) {
968 } else if (rc != -ENOENT) {
974 cname->ln_name = info->lti_key;
975 cname->ln_namelen = namelen;
976 rc = linkea_data_new(&ldata, &info->lti_linkea_buf2);
980 rc = linkea_add_buf(&ldata, cname, pfid);
984 rc = lfsck_ibits_lock(env, lfsck, orphan, &clh,
985 MDS_INODELOCK_UPDATE | MDS_INODELOCK_LOOKUP |
986 MDS_INODELOCK_XATTR, LCK_EX);
990 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
991 ldata.ld_leh->leh_len);
992 th = dt_trans_create(env, dev);
994 GOTO(log, rc = PTR_ERR(th));
996 if (S_ISDIR(lfsck_object_type(orphan))) {
997 rc = dt_declare_delete(env, orphan,
998 (const struct dt_key *)dotdot, th);
1002 rec->rec_type = S_IFDIR;
1003 rec->rec_fid = pfid;
1004 rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
1005 (const struct dt_key *)dotdot, th);
1010 rc = dt_declare_xattr_set(env, orphan, &linkea_buf,
1011 XATTR_NAME_LINK, 0, th);
1016 rec->rec_type = lfsck_object_type(orphan) & S_IFMT;
1017 rec->rec_fid = cfid;
1018 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1019 (const struct dt_key *)cname->ln_name,
1024 if (S_ISDIR(rec->rec_type)) {
1025 rc = dt_declare_ref_add(env, parent, th);
1031 memset(la, 0, sizeof(*la));
1032 la->la_ctime = cfs_time_current_sec();
1033 la->la_valid = LA_CTIME;
1034 rc = dt_declare_attr_set(env, orphan, la, th);
1038 rc = dt_trans_start_local(env, dev, th);
1042 dt_write_lock(env, orphan, 0);
1043 rc = lfsck_links_read(env, orphan, &ldata);
1044 if (likely((rc == -ENODATA) || (rc == -EINVAL) ||
1045 (rc == 0 && ldata.ld_leh->leh_reccount == 0))) {
1046 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1047 GOTO(unlock, rc = 1);
1049 if (S_ISDIR(lfsck_object_type(orphan))) {
1050 rc = dt_delete(env, orphan,
1051 (const struct dt_key *)dotdot, th,
1056 rec->rec_type = S_IFDIR;
1057 rec->rec_fid = pfid;
1058 rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1059 (const struct dt_key *)dotdot, th,
1065 rc = dt_xattr_set(env, orphan, &linkea_buf, XATTR_NAME_LINK, 0,
1068 if (rc == 0 && count != NULL)
1069 *count = ldata.ld_leh->leh_reccount;
1073 dt_write_unlock(env, orphan);
1075 if (rc == 0 && !exist) {
1076 rec->rec_type = lfsck_object_type(orphan) & S_IFMT;
1077 rec->rec_fid = cfid;
1078 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1079 (const struct dt_key *)cname->ln_name,
1080 th, BYPASS_CAPA, 1);
1081 if (rc == 0 && S_ISDIR(rec->rec_type)) {
1082 dt_write_lock(env, parent, 0);
1083 rc = dt_ref_add(env, parent, th);
1084 dt_write_unlock(env, parent);
1089 rc = dt_attr_set(env, orphan, la, th, BYPASS_CAPA);
1091 GOTO(stop, rc = (rc == 0 ? 1 : rc));
1094 dt_write_unlock(env, orphan);
1097 dt_trans_stop(env, dev, th);
1100 lfsck_ibits_unlock(&clh, LCK_EX);
1102 CDEBUG(D_LFSCK, "%s: namespace LFSCK insert orphan for the "
1103 "object "DFID", name = %s: rc = %d\n",
1104 lfsck_lfsck2name(lfsck), PFID(cfid),
1105 cname->ln_name != NULL ? cname->ln_name : "<NULL>", rc);
1108 struct lfsck_namespace *ns = com->lc_file_ram;
1110 ns->ln_flags |= LF_INCONSISTENT;
1117 * Add the specified name entry back to namespace.
1119 * If there is a linkEA entry that back references a name entry under
1120 * some parent directory, but such parent directory does not have the
1121 * claimed name entry. On the other hand, the linkEA entries count is
1122 * not larger than the MDT-object's hard link count. Under such case,
1123 * it is quite possible that the name entry is lost. Then the LFSCK
1124 * should add the name entry back to the namespace.
1126 * \param[in] env pointer to the thread context
1127 * \param[in] com pointer to the lfsck component
1128 * \param[in] parent pointer to the directory under which the name entry
1129 * will be inserted into
1130 * \param[in] child pointer to the object referenced by the name entry
1131 * that to be inserted into the parent
1132 * \param[in] name the name for the child in the parent directory
1134 * \retval positive number for repaired cases
1135 * \retval 0 if nothing to be repaired
1136 * \retval negative error number on failure
1138 static int lfsck_namespace_insert_normal(const struct lu_env *env,
1139 struct lfsck_component *com,
1140 struct dt_object *parent,
1141 struct dt_object *child,
1144 struct lfsck_thread_info *info = lfsck_env_info(env);
1145 struct lu_attr *la = &info->lti_la;
1146 struct dt_insert_rec *rec = &info->lti_dt_rec;
1147 struct lfsck_instance *lfsck = com->lc_lfsck;
1148 /* The child and its name may be on different MDTs. */
1149 struct dt_device *dev = lfsck->li_next;
1150 struct dt_object *pobj = NULL;
1151 struct dt_object *cobj = NULL;
1152 struct thandle *th = NULL;
1153 struct lfsck_lock_handle *llh = &info->lti_llh;
1157 /* @parent/@child may be based on lfsck->li_bottom,
1158 * but here we need the object based on the lfsck->li_next. */
1160 pobj = lfsck_object_locate(dev, parent);
1162 GOTO(log, rc = PTR_ERR(pobj));
1164 if (unlikely(!dt_try_as_dir(env, pobj)))
1165 GOTO(log, rc = -ENOTDIR);
1167 cobj = lfsck_object_locate(dev, child);
1169 GOTO(log, rc = PTR_ERR(cobj));
1171 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1174 rc = lfsck_lock(env, lfsck, parent, name, llh,
1175 MDS_INODELOCK_UPDATE, LCK_PW);
1179 th = dt_trans_create(env, dev);
1181 GOTO(unlock, rc = PTR_ERR(th));
1183 rec->rec_type = lfsck_object_type(cobj) & S_IFMT;
1184 rec->rec_fid = lfsck_dto2fid(cobj);
1185 rc = dt_declare_insert(env, pobj, (const struct dt_rec *)rec,
1186 (const struct dt_key *)name, th);
1190 if (S_ISDIR(rec->rec_type)) {
1191 rc = dt_declare_ref_add(env, pobj, th);
1196 memset(la, 0, sizeof(*la));
1197 la->la_ctime = cfs_time_current_sec();
1198 la->la_valid = LA_CTIME;
1199 rc = dt_declare_attr_set(env, pobj, la, th);
1203 rc = dt_declare_attr_set(env, cobj, la, th);
1207 rc = dt_trans_start_local(env, dev, th);
1211 rc = dt_insert(env, pobj, (const struct dt_rec *)rec,
1212 (const struct dt_key *)name, th, BYPASS_CAPA, 1);
1216 if (S_ISDIR(rec->rec_type)) {
1217 dt_write_lock(env, pobj, 0);
1218 rc = dt_ref_add(env, pobj, th);
1219 dt_write_unlock(env, pobj);
1224 la->la_ctime = cfs_time_current_sec();
1225 rc = dt_attr_set(env, pobj, la, th, BYPASS_CAPA);
1229 rc = dt_attr_set(env, cobj, la, th, BYPASS_CAPA);
1231 GOTO(stop, rc = (rc == 0 ? 1 : rc));
1234 dt_trans_stop(env, dev, th);
1240 CDEBUG(D_LFSCK, "%s: namespace LFSCK insert object "DFID" with "
1241 "the name %s and type %o to the parent "DFID": rc = %d\n",
1242 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(child)), name,
1243 lfsck_object_type(child) & S_IFMT,
1244 PFID(lfsck_dto2fid(parent)), rc);
1247 struct lfsck_namespace *ns = com->lc_file_ram;
1249 ns->ln_flags |= LF_INCONSISTENT;
1251 ns->ln_lost_dirent_repaired++;
1258 * Create the specified orphan directory.
1260 * For the case that the parent MDT-object stored in some MDT-object's
1261 * linkEA entry is lost, the LFSCK will re-create the parent object as
1262 * an orphan and insert it into .lustre/lost+found/MDTxxxx/ directory
1263 * with the name ${FID}-P-${conflict_version}.
1265 * \param[in] env pointer to the thread context
1266 * \param[in] com pointer to the lfsck component
1267 * \param[in] orphan pointer to the orphan MDT-object to be created
1268 * \param[in] lmv pointer to master LMV EA that will be set to the orphan
1270 * \retval positive number for repaired cases
1271 * \retval negative error number on failure
1273 static int lfsck_namespace_create_orphan_dir(const struct lu_env *env,
1274 struct lfsck_component *com,
1275 struct dt_object *orphan,
1276 struct lmv_mds_md_v1 *lmv)
1278 struct lfsck_thread_info *info = lfsck_env_info(env);
1279 struct lu_attr *la = &info->lti_la;
1280 struct dt_allocation_hint *hint = &info->lti_hint;
1281 struct dt_object_format *dof = &info->lti_dof;
1282 struct lu_name *cname = &info->lti_name2;
1283 struct dt_insert_rec *rec = &info->lti_dt_rec;
1284 struct lmv_mds_md_v1 *lmv2 = &info->lti_lmv2;
1285 const struct lu_fid *cfid = lfsck_dto2fid(orphan);
1287 struct lfsck_instance *lfsck = com->lc_lfsck;
1288 struct lfsck_namespace *ns = com->lc_file_ram;
1289 struct dt_device *dev = lfsck_obj2dev(orphan);
1290 struct dt_object *parent = NULL;
1291 struct thandle *th = NULL;
1292 struct lfsck_lock_handle *llh = &info->lti_llh;
1293 struct linkea_data ldata = { NULL };
1294 struct lu_buf linkea_buf;
1295 struct lu_buf lmv_buf;
1302 LASSERT(!dt_object_exists(orphan));
1304 cname->ln_name = NULL;
1305 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1308 if (dt_object_remote(orphan)) {
1309 LASSERT(lfsck->li_lpf_root_obj != NULL);
1311 idx = lfsck_find_mdt_idx_by_fid(env, lfsck, cfid);
1313 GOTO(log, rc = idx);
1315 snprintf(name, 8, "MDT%04x", idx);
1316 rc = dt_lookup(env, lfsck->li_lpf_root_obj,
1317 (struct dt_rec *)&tfid,
1318 (const struct dt_key *)name, BYPASS_CAPA);
1320 GOTO(log, rc = (rc == -ENOENT ? -ENXIO : rc));
1322 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
1324 GOTO(log, rc = PTR_ERR(parent));
1326 if (unlikely(!dt_try_as_dir(env, parent)))
1327 GOTO(log, rc = -ENOTDIR);
1329 if (unlikely(lfsck->li_lpf_obj == NULL))
1330 GOTO(log, rc = -ENXIO);
1332 parent = lfsck->li_lpf_obj;
1335 dev = lfsck_find_dev_by_fid(env, lfsck, cfid);
1337 GOTO(log, rc = PTR_ERR(dev));
1343 namelen = snprintf(name, 31, DFID"-P-%d",
1345 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
1346 (const struct dt_key *)name, BYPASS_CAPA);
1347 if (rc != 0 && rc != -ENOENT)
1351 rc = lfsck_lock(env, lfsck, parent, name, llh,
1352 MDS_INODELOCK_UPDATE, LCK_PW);
1356 /* Re-check whether the name conflict with othrs after taken
1358 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
1359 (const struct dt_key *)name, BYPASS_CAPA);
1360 if (unlikely(rc == 0)) {
1368 cname->ln_name = name;
1369 cname->ln_namelen = namelen;
1371 memset(la, 0, sizeof(*la));
1372 la->la_mode = S_IFDIR | 0700;
1373 la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
1374 LA_ATIME | LA_MTIME | LA_CTIME;
1376 orphan->do_ops->do_ah_init(env, hint, parent, orphan,
1377 la->la_mode & S_IFMT);
1379 memset(dof, 0, sizeof(*dof));
1380 dof->dof_type = dt_mode_to_dft(S_IFDIR);
1382 rc = linkea_data_new(&ldata, &info->lti_linkea_buf2);
1386 rc = linkea_add_buf(&ldata, cname, lfsck_dto2fid(parent));
1390 th = dt_trans_create(env, dev);
1392 GOTO(unlock1, rc = PTR_ERR(th));
1394 /* Sync the remote transaction to guarantee that the subsequent
1395 * lock against the @orphan can find the @orphan in time. */
1396 if (dt_object_remote(orphan))
1399 rc = dt_declare_create(env, orphan, la, hint, dof, th);
1403 if (unlikely(!dt_try_as_dir(env, orphan)))
1404 GOTO(stop, rc = -ENOTDIR);
1406 rec->rec_type = S_IFDIR;
1407 rec->rec_fid = cfid;
1408 rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
1409 (const struct dt_key *)dot, th);
1413 rec->rec_fid = lfsck_dto2fid(parent);
1414 rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
1415 (const struct dt_key *)dotdot, th);
1417 rc = dt_declare_ref_add(env, orphan, th);
1422 rc = dt_declare_ref_add(env, orphan, th);
1427 lmv->lmv_magic = LMV_MAGIC;
1428 lmv->lmv_master_mdt_index = lfsck_dev_idx(lfsck);
1429 lfsck_lmv_header_cpu_to_le(lmv2, lmv);
1430 lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
1431 rc = dt_declare_xattr_set(env, orphan, &lmv_buf,
1432 XATTR_NAME_LMV, 0, th);
1437 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
1438 ldata.ld_leh->leh_len);
1439 rc = dt_declare_xattr_set(env, orphan, &linkea_buf,
1440 XATTR_NAME_LINK, 0, th);
1444 rec->rec_fid = cfid;
1445 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1446 (const struct dt_key *)name, th);
1448 rc = dt_declare_ref_add(env, parent, th);
1453 rc = dt_trans_start_local(env, dev, th);
1457 dt_write_lock(env, orphan, 0);
1458 rc = dt_create(env, orphan, la, hint, dof, th);
1462 rec->rec_fid = cfid;
1463 rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1464 (const struct dt_key *)dot, th, BYPASS_CAPA, 1);
1468 rec->rec_fid = lfsck_dto2fid(parent);
1469 rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1470 (const struct dt_key *)dotdot, th,
1475 rc = dt_ref_add(env, orphan, th);
1480 rc = dt_xattr_set(env, orphan, &lmv_buf, XATTR_NAME_LMV, 0,
1486 rc = dt_xattr_set(env, orphan, &linkea_buf,
1487 XATTR_NAME_LINK, 0, th, BYPASS_CAPA);
1488 dt_write_unlock(env, orphan);
1492 rec->rec_fid = cfid;
1493 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1494 (const struct dt_key *)name, th, BYPASS_CAPA, 1);
1496 dt_write_lock(env, parent, 0);
1497 rc = dt_ref_add(env, parent, th);
1498 dt_write_unlock(env, parent);
1501 GOTO(stop, rc = (rc == 0 ? 1 : rc));
1504 dt_write_unlock(env, orphan);
1507 dt_trans_stop(env, dev, th);
1513 CDEBUG(D_LFSCK, "%s: namespace LFSCK create orphan dir for "
1514 "the object "DFID", name = %s: rc = %d\n",
1515 lfsck_lfsck2name(lfsck), PFID(cfid),
1516 cname->ln_name != NULL ? cname->ln_name : "<NULL>", rc);
1518 if (parent != NULL && !IS_ERR(parent) && parent != lfsck->li_lpf_obj)
1519 lfsck_object_put(env, parent);
1522 ns->ln_flags |= LF_INCONSISTENT;
1528 * Remove the specified entry from the linkEA.
1530 * Locate the linkEA entry with the given @cname and @pfid, then
1531 * remove this entry or the other entries those are repeated with
1534 * \param[in] env pointer to the thread context
1535 * \param[in] com pointer to the lfsck component
1536 * \param[in] obj pointer to the dt_object to be handled
1537 * \param[in,out]ldata pointer to the buffer that holds the linkEA
1538 * \param[in] cname the name for the child in the parent directory
1539 * \param[in] pfid the parent directory's FID for the linkEA
1540 * \param[in] next if true, then remove the first found linkEA
1541 * entry, and move the ldata->ld_lee to next entry
1543 * \retval positive number for repaired cases
1544 * \retval 0 if nothing to be repaired
1545 * \retval negative error number on failure
1547 static int lfsck_namespace_shrink_linkea(const struct lu_env *env,
1548 struct lfsck_component *com,
1549 struct dt_object *obj,
1550 struct linkea_data *ldata,
1551 struct lu_name *cname,
1552 struct lu_fid *pfid,
1555 struct lfsck_instance *lfsck = com->lc_lfsck;
1556 struct dt_device *dev = lfsck_obj2dev(obj);
1557 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
1558 struct thandle *th = NULL;
1559 struct lustre_handle lh = { 0 };
1560 struct linkea_data ldata_new = { NULL };
1561 struct lu_buf linkea_buf;
1565 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
1566 MDS_INODELOCK_UPDATE | MDS_INODELOCK_XATTR,
1572 linkea_del_buf(ldata, cname);
1574 lfsck_namespace_filter_linkea_entry(ldata, cname, pfid,
1576 lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
1577 ldata->ld_leh->leh_len);
1580 th = dt_trans_create(env, dev);
1582 GOTO(unlock1, rc = PTR_ERR(th));
1584 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
1585 XATTR_NAME_LINK, 0, th);
1589 rc = dt_trans_start_local(env, dev, th);
1593 dt_write_lock(env, obj, 0);
1594 if (unlikely(lfsck_is_dead_obj(obj)))
1595 GOTO(unlock2, rc = -ENOENT);
1597 rc = lfsck_links_read2(env, obj, &ldata_new);
1601 /* The specified linkEA entry has been removed by race. */
1602 rc = linkea_links_find(&ldata_new, cname, pfid);
1604 GOTO(unlock2, rc = 0);
1606 if (bk->lb_param & LPF_DRYRUN)
1607 GOTO(unlock2, rc = 1);
1610 linkea_del_buf(&ldata_new, cname);
1612 lfsck_namespace_filter_linkea_entry(&ldata_new, cname, pfid,
1615 if (linkea_buf.lb_len < ldata_new.ld_leh->leh_len) {
1616 dt_write_unlock(env, obj);
1617 dt_trans_stop(env, dev, th);
1618 lfsck_buf_init(&linkea_buf, ldata_new.ld_buf->lb_buf,
1619 ldata_new.ld_leh->leh_len);
1623 lfsck_buf_init(&linkea_buf, ldata_new.ld_buf->lb_buf,
1624 ldata_new.ld_leh->leh_len);
1625 rc = dt_xattr_set(env, obj, &linkea_buf,
1626 XATTR_NAME_LINK, 0, th, BYPASS_CAPA);
1628 GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
1631 dt_write_unlock(env, obj);
1634 dt_trans_stop(env, dev, th);
1637 lfsck_ibits_unlock(&lh, LCK_EX);
1640 CDEBUG(D_LFSCK, "%s: namespace LFSCK remove %s linkEA entry "
1641 "for the object: "DFID", parent "DFID", name %.*s\n",
1642 lfsck_lfsck2name(lfsck), next ? "invalid" : "redundant",
1643 PFID(lfsck_dto2fid(obj)), PFID(pfid), cname->ln_namelen,
1647 struct lfsck_namespace *ns = com->lc_file_ram;
1649 ns->ln_flags |= LF_INCONSISTENT;
1656 * Conditionally remove the specified entry from the linkEA.
1658 * Take the parent lock firstly, then check whether the specified
1659 * name entry exists or not: if yes, do nothing; otherwise, call
1660 * lfsck_namespace_shrink_linkea() to remove the linkea entry.
1662 * \param[in] env pointer to the thread context
1663 * \param[in] com pointer to the lfsck component
1664 * \param[in] parent pointer to the parent directory
1665 * \param[in] child pointer to the child object that holds the linkEA
1666 * \param[in,out]ldata pointer to the buffer that holds the linkEA
1667 * \param[in] cname the name for the child in the parent directory
1668 * \param[in] pfid the parent directory's FID for the linkEA
1670 * \retval positive number for repaired cases
1671 * \retval 0 if nothing to be repaired
1672 * \retval negative error number on failure
1674 static int lfsck_namespace_shrink_linkea_cond(const struct lu_env *env,
1675 struct lfsck_component *com,
1676 struct dt_object *parent,
1677 struct dt_object *child,
1678 struct linkea_data *ldata,
1679 struct lu_name *cname,
1680 struct lu_fid *pfid)
1682 struct lfsck_thread_info *info = lfsck_env_info(env);
1683 struct lu_fid *cfid = &info->lti_fid3;
1684 struct lfsck_lock_handle *llh = &info->lti_llh;
1688 rc = lfsck_lock(env, com->lc_lfsck, parent, cname->ln_name, llh,
1689 MDS_INODELOCK_UPDATE, LCK_PR);
1693 dt_read_lock(env, parent, 0);
1694 if (unlikely(lfsck_is_dead_obj(parent))) {
1695 dt_read_unlock(env, parent);
1697 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
1703 rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
1704 (const struct dt_key *)cname->ln_name,
1706 dt_read_unlock(env, parent);
1708 /* It is safe to release the ldlm lock, because when the logic come
1709 * here, we have got all the needed information above whether the
1710 * linkEA entry is valid or not. It is not important that others
1711 * may add new linkEA entry after the ldlm lock released. If other
1712 * has removed the specified linkEA entry by race, then it is OK,
1713 * because the subsequent lfsck_namespace_shrink_linkea() can handle
1716 if (rc == -ENOENT) {
1717 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
1726 /* The LFSCK just found some internal status of cross-MDTs
1727 * create operation. That is normal. */
1728 if (lu_fid_eq(cfid, lfsck_dto2fid(child))) {
1729 linkea_next_entry(ldata);
1734 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata, cname,
1741 * Conditionally replace name entry in the parent.
1743 * As required, the LFSCK may re-create the lost MDT-object for dangling
1744 * name entry, but such repairing may be wrong because of bad FID in the
1745 * name entry. As the LFSCK processing, the real MDT-object may be found,
1746 * then the LFSCK should check whether the former re-created MDT-object
1747 * has been modified or not, if not, then destroy it and update the name
1748 * entry in the parent to reference the real MDT-object.
1750 * \param[in] env pointer to the thread context
1751 * \param[in] com pointer to the lfsck component
1752 * \param[in] parent pointer to the parent directory
1753 * \param[in] child pointer to the MDT-object that may be the real
1754 * MDT-object corresponding to the name entry in parent
1755 * \param[in] cfid the current FID in the name entry
1756 * \param[in] cname contains the name of the child in the parent directory
1758 * \retval positive number for repaired cases
1759 * \retval 0 if nothing to be repaired
1760 * \retval negative error number on failure
1762 static int lfsck_namespace_replace_cond(const struct lu_env *env,
1763 struct lfsck_component *com,
1764 struct dt_object *parent,
1765 struct dt_object *child,
1766 const struct lu_fid *cfid,
1767 const struct lu_name *cname)
1769 struct lfsck_thread_info *info = lfsck_env_info(env);
1770 struct lu_attr *la = &info->lti_la;
1771 struct dt_insert_rec *rec = &info->lti_dt_rec;
1773 struct lfsck_instance *lfsck = com->lc_lfsck;
1774 /* The child and its name may be on different MDTs. */
1775 struct dt_device *dev = lfsck->li_next;
1776 const char *name = cname->ln_name;
1777 struct dt_object *pobj = NULL;
1778 struct dt_object *cobj = NULL;
1779 struct lfsck_lock_handle *pllh = &info->lti_llh;
1780 struct lustre_handle clh = { 0 };
1781 struct linkea_data ldata = { NULL };
1782 struct thandle *th = NULL;
1787 /* @parent/@child may be based on lfsck->li_bottom,
1788 * but here we need the object based on the lfsck->li_next. */
1790 pobj = lfsck_object_locate(dev, parent);
1792 GOTO(log, rc = PTR_ERR(pobj));
1794 if (unlikely(!dt_try_as_dir(env, pobj)))
1795 GOTO(log, rc = -ENOTDIR);
1797 rc = lfsck_lock(env, lfsck, parent, name, pllh,
1798 MDS_INODELOCK_UPDATE, LCK_PW);
1802 if (!fid_is_sane(cfid)) {
1807 cobj = lfsck_object_find_by_dev(env, dev, cfid);
1810 if (rc == -ENOENT) {
1818 if (!dt_object_exists(cobj)) {
1823 rc = dt_lookup(env, pobj, (struct dt_rec *)&tfid,
1824 (const struct dt_key *)name, BYPASS_CAPA);
1825 if (rc == -ENOENT) {
1833 /* Someone changed the name entry, cannot replace it. */
1834 if (!lu_fid_eq(cfid, &tfid))
1837 /* lock the object to be destroyed. */
1838 rc = lfsck_ibits_lock(env, lfsck, cobj, &clh,
1839 MDS_INODELOCK_UPDATE |
1840 MDS_INODELOCK_UPDATE | MDS_INODELOCK_XATTR,
1845 if (unlikely(lfsck_is_dead_obj(cobj))) {
1850 rc = dt_attr_get(env, cobj, la, BYPASS_CAPA);
1854 /* The object has been modified by other(s), or it is not created by
1855 * LFSCK, the two cases are indistinguishable. So cannot replace it. */
1856 if (la->la_ctime != 0)
1859 if (S_ISREG(la->la_mode)) {
1860 rc = dt_xattr_get(env, cobj, &LU_BUF_NULL, XATTR_NAME_LOV,
1862 /* If someone has created related OST-object(s),
1864 if ((rc > 0) || (rc < 0 && rc != -ENODATA))
1865 GOTO(log, rc = (rc > 0 ? 0 : rc));
1869 dt_read_lock(env, child, 0);
1870 rc = lfsck_links_read2(env, child, &ldata);
1871 dt_read_unlock(env, child);
1873 /* Someone changed the child, no need to replace. */
1880 rc = linkea_links_find(&ldata, cname, lfsck_dto2fid(pobj));
1881 /* Someone moved the child, no need to replace. */
1885 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1888 th = dt_trans_create(env, dev);
1890 GOTO(log, rc = PTR_ERR(th));
1893 rc = dt_declare_destroy(env, cobj, th);
1898 rc = dt_declare_delete(env, pobj, (const struct dt_key *)name, th);
1902 rec->rec_type = S_IFDIR;
1903 rec->rec_fid = lfsck_dto2fid(child);
1904 rc = dt_declare_insert(env, pobj, (const struct dt_rec *)rec,
1905 (const struct dt_key *)name, th);
1909 rc = dt_trans_start_local(env, dev, th);
1914 rc = dt_destroy(env, cobj, th);
1919 /* The old name entry maybe not exist. */
1920 rc = dt_delete(env, pobj, (const struct dt_key *)name, th, BYPASS_CAPA);
1921 if (rc != 0 && rc != -ENOENT)
1924 rc = dt_insert(env, pobj, (const struct dt_rec *)rec,
1925 (const struct dt_key *)name, th, BYPASS_CAPA, 1);
1927 GOTO(stop, rc = (rc == 0 ? 1 : rc));
1930 dt_trans_stop(env, dev, th);
1933 lfsck_ibits_unlock(&clh, LCK_EX);
1936 if (cobj != NULL && !IS_ERR(cobj))
1937 lfsck_object_put(env, cobj);
1939 CDEBUG(D_LFSCK, "%s: namespace LFSCK conditionally destroy the "
1940 "object "DFID" because of conflict with the object "DFID
1941 " under the parent "DFID" with name %s: rc = %d\n",
1942 lfsck_lfsck2name(lfsck), PFID(cfid),
1943 PFID(lfsck_dto2fid(child)), PFID(lfsck_dto2fid(parent)),
1950 * Overwrite the linkEA for the object with the given ldata.
1952 * The caller should take the ldlm lock before the calling.
1954 * \param[in] env pointer to the thread context
1955 * \param[in] com pointer to the lfsck component
1956 * \param[in] obj pointer to the dt_object to be handled
1957 * \param[in] ldata pointer to the new linkEA data
1959 * \retval positive number for repaired cases
1960 * \retval 0 if nothing to be repaired
1961 * \retval negative error number on failure
1963 int lfsck_namespace_rebuild_linkea(const struct lu_env *env,
1964 struct lfsck_component *com,
1965 struct dt_object *obj,
1966 struct linkea_data *ldata)
1968 struct lfsck_instance *lfsck = com->lc_lfsck;
1969 struct dt_device *dev = lfsck_obj2dev(obj);
1970 struct thandle *th = NULL;
1971 struct lu_buf linkea_buf;
1975 th = dt_trans_create(env, dev);
1977 GOTO(log, rc = PTR_ERR(th));
1979 lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
1980 ldata->ld_leh->leh_len);
1981 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
1982 XATTR_NAME_LINK, 0, th);
1986 rc = dt_trans_start_local(env, dev, th);
1990 dt_write_lock(env, obj, 0);
1991 if (unlikely(lfsck_is_dead_obj(obj)))
1992 GOTO(unlock, rc = 0);
1994 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1995 GOTO(unlock, rc = 1);
1997 rc = dt_xattr_set(env, obj, &linkea_buf,
1998 XATTR_NAME_LINK, 0, th, BYPASS_CAPA);
2000 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2003 dt_write_unlock(env, obj);
2006 dt_trans_stop(env, dev, th);
2009 CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild linkEA for the "
2010 "object "DFID": rc = %d\n",
2011 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
2014 struct lfsck_namespace *ns = com->lc_file_ram;
2016 ns->ln_flags |= LF_INCONSISTENT;
2023 * Repair invalid name entry.
2025 * If the name entry contains invalid information, such as bad file type
2026 * or (and) corrupted object FID, then either remove the name entry or
2027 * udpate the name entry with the given (right) information.
2029 * \param[in] env pointer to the thread context
2030 * \param[in] com pointer to the lfsck component
2031 * \param[in] parent pointer to the parent directory
2032 * \param[in] child pointer to the object referenced by the name entry
2033 * \param[in] name the old name of the child under the parent directory
2034 * \param[in] name2 the new name of the child under the parent directory
2035 * \param[in] type the type claimed by the name entry
2036 * \param[in] update update the name entry if true; otherwise, remove it
2037 * \param[in] dec decrease the parent nlink count if true
2039 * \retval positive number for repaired successfully
2040 * \retval 0 if nothing to be repaired
2041 * \retval negative error number on failure
2043 int lfsck_namespace_repair_dirent(const struct lu_env *env,
2044 struct lfsck_component *com,
2045 struct dt_object *parent,
2046 struct dt_object *child,
2047 const char *name, const char *name2,
2048 __u16 type, bool update, bool dec)
2050 struct lfsck_thread_info *info = lfsck_env_info(env);
2051 struct dt_insert_rec *rec = &info->lti_dt_rec;
2052 const struct lu_fid *cfid = lfsck_dto2fid(child);
2054 struct lfsck_instance *lfsck = com->lc_lfsck;
2055 struct dt_device *dev = lfsck->li_next;
2056 struct thandle *th = NULL;
2057 struct lfsck_lock_handle *llh = &info->lti_llh;
2058 struct lustre_handle lh = { 0 };
2062 if (unlikely(!dt_try_as_dir(env, parent)))
2063 GOTO(log, rc = -ENOTDIR);
2065 if (!update || strcmp(name, name2) == 0)
2066 rc = lfsck_lock(env, lfsck, parent, name, llh,
2067 MDS_INODELOCK_UPDATE, LCK_PW);
2069 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
2070 MDS_INODELOCK_UPDATE, LCK_PW);
2074 th = dt_trans_create(env, dev);
2076 GOTO(unlock1, rc = PTR_ERR(th));
2078 rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
2083 rec->rec_type = lfsck_object_type(child) & S_IFMT;
2084 rec->rec_fid = cfid;
2085 rc = dt_declare_insert(env, parent,
2086 (const struct dt_rec *)rec,
2087 (const struct dt_key *)name2, th);
2093 rc = dt_declare_ref_del(env, parent, th);
2098 rc = dt_trans_start_local(env, dev, th);
2102 dt_write_lock(env, parent, 0);
2103 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
2104 (const struct dt_key *)name, BYPASS_CAPA);
2105 /* Someone has removed the bad name entry by race. */
2107 GOTO(unlock2, rc = 0);
2112 /* Someone has removed the bad name entry and reused it for other
2113 * object by race. */
2114 if (!lu_fid_eq(&tfid, cfid))
2115 GOTO(unlock2, rc = 0);
2117 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2118 GOTO(unlock2, rc = 1);
2120 rc = dt_delete(env, parent, (const struct dt_key *)name, th,
2126 rc = dt_insert(env, parent,
2127 (const struct dt_rec *)rec,
2128 (const struct dt_key *)name2, th,
2135 rc = dt_ref_del(env, parent, th);
2140 GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
2143 dt_write_unlock(env, parent);
2146 dt_trans_stop(env, dev, th);
2148 /* We are not sure whether the child will become orphan or not.
2149 * Record it in the LFSCK trace file for further checking in
2150 * the second-stage scanning. */
2151 if (!update && !dec && rc == 0)
2152 lfsck_namespace_trace_update(env, com, cfid,
2153 LNTF_CHECK_LINKEA, true);
2156 /* It is harmless even if unlock the unused lock_handle */
2157 lfsck_ibits_unlock(&lh, LCK_PW);
2161 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found bad name "
2162 "entry for: parent "DFID", child "DFID", name %s, type "
2163 "in name entry %o, type claimed by child %o. repair it "
2164 "by %s with new name2 %s: rc = %d\n", lfsck_lfsck2name(lfsck),
2165 PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)),
2166 name, type, update ? lfsck_object_type(child) : 0,
2167 update ? "updating" : "removing", name2, rc);
2170 struct lfsck_namespace *ns = com->lc_file_ram;
2172 ns->ln_flags |= LF_INCONSISTENT;
2179 * Update the ".." name entry for the given object.
2181 * The object's ".." is corrupted, this function will update the ".." name
2182 * entry with the given pfid, and the linkEA with the given ldata.
2184 * The caller should take the ldlm lock before the calling.
2186 * \param[in] env pointer to the thread context
2187 * \param[in] com pointer to the lfsck component
2188 * \param[in] obj pointer to the dt_object to be handled
2189 * \param[in] pfid the new fid for the object's ".." name entry
2190 * \param[in] cname the name for the @obj in the parent directory
2192 * \retval positive number for repaired cases
2193 * \retval 0 if nothing to be repaired
2194 * \retval negative error number on failure
2196 static int lfsck_namespace_repair_unmatched_pairs(const struct lu_env *env,
2197 struct lfsck_component *com,
2198 struct dt_object *obj,
2199 const struct lu_fid *pfid,
2200 struct lu_name *cname)
2202 struct lfsck_thread_info *info = lfsck_env_info(env);
2203 struct dt_insert_rec *rec = &info->lti_dt_rec;
2204 struct lfsck_instance *lfsck = com->lc_lfsck;
2205 struct dt_device *dev = lfsck_obj2dev(obj);
2206 struct thandle *th = NULL;
2207 struct linkea_data ldata = { NULL };
2208 struct lu_buf linkea_buf;
2212 LASSERT(!dt_object_remote(obj));
2213 LASSERT(S_ISDIR(lfsck_object_type(obj)));
2215 rc = linkea_data_new(&ldata, &info->lti_big_buf);
2219 rc = linkea_add_buf(&ldata, cname, pfid);
2223 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
2224 ldata.ld_leh->leh_len);
2226 th = dt_trans_create(env, dev);
2228 GOTO(log, rc = PTR_ERR(th));
2230 rc = dt_declare_delete(env, obj, (const struct dt_key *)dotdot, th);
2234 rec->rec_type = S_IFDIR;
2235 rec->rec_fid = pfid;
2236 rc = dt_declare_insert(env, obj, (const struct dt_rec *)rec,
2237 (const struct dt_key *)dotdot, th);
2241 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
2242 XATTR_NAME_LINK, 0, th);
2246 rc = dt_trans_start_local(env, dev, th);
2250 dt_write_lock(env, obj, 0);
2251 if (unlikely(lfsck_is_dead_obj(obj)))
2252 GOTO(unlock, rc = 0);
2254 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2255 GOTO(unlock, rc = 1);
2257 /* The old ".." name entry maybe not exist. */
2258 dt_delete(env, obj, (const struct dt_key *)dotdot, th,
2261 rc = dt_insert(env, obj, (const struct dt_rec *)rec,
2262 (const struct dt_key *)dotdot, th, BYPASS_CAPA, 1);
2266 rc = dt_xattr_set(env, obj, &linkea_buf,
2267 XATTR_NAME_LINK, 0, th, BYPASS_CAPA);
2269 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2272 dt_write_unlock(env, obj);
2275 dt_trans_stop(env, dev, th);
2278 CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild dotdot name entry for "
2279 "the object "DFID", new parent "DFID": rc = %d\n",
2280 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)),
2284 struct lfsck_namespace *ns = com->lc_file_ram;
2286 ns->ln_flags |= LF_INCONSISTENT;
2293 * Handle orphan @obj during Double Scan Directory.
2295 * Remove the @obj's current (invalid) linkEA entries, and insert
2296 * it in the directory .lustre/lost+found/MDTxxxx/ with the name:
2297 * ${FID}-${PFID}-D-${conflict_version}
2299 * The caller should take the ldlm lock before the calling.
2301 * \param[in] env pointer to the thread context
2302 * \param[in] com pointer to the lfsck component
2303 * \param[in] obj pointer to the orphan object to be handled
2304 * \param[in] pfid the new fid for the object's ".." name entry
2305 * \param[in,out] lh ldlm lock handler for the given @obj
2306 * \param[out] type to tell the caller what the inconsistency is
2308 * \retval positive number for repaired cases
2309 * \retval 0 if nothing to be repaired
2310 * \retval negative error number on failure
2313 lfsck_namespace_dsd_orphan(const struct lu_env *env,
2314 struct lfsck_component *com,
2315 struct dt_object *obj,
2316 const struct lu_fid *pfid,
2317 struct lustre_handle *lh,
2318 enum lfsck_namespace_inconsistency_type *type)
2320 struct lfsck_thread_info *info = lfsck_env_info(env);
2321 struct lfsck_namespace *ns = com->lc_file_ram;
2325 /* Remove the unrecognized linkEA. */
2326 rc = lfsck_namespace_links_remove(env, com, obj);
2327 lfsck_ibits_unlock(lh, LCK_EX);
2328 if (rc < 0 && rc != -ENODATA)
2331 *type = LNIT_MUL_REF;
2333 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
2334 * ever tried to verify some remote MDT-object that resides on this
2335 * MDT, but this MDT failed to respond such request. So means there
2336 * may be some remote name entry on other MDT that references this
2337 * object with another name, so we cannot know whether this linkEA
2338 * is valid or not. So keep it there and maybe resolved when next
2340 if (ns->ln_flags & LF_INCOMPLETE)
2343 /* The unique linkEA is invalid, even if the ".." name entry may be
2344 * valid, we still cannot know via which name entry this directory
2345 * will be referenced. Then handle it as pure orphan. */
2346 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2347 "-"DFID, PFID(pfid));
2348 rc = lfsck_namespace_insert_orphan(env, com, obj,
2349 info->lti_tmpbuf, "D", NULL);
2355 * Double Scan Directory object for single linkEA entry case.
2357 * The given @child has unique linkEA entry. If the linkEA entry is valid,
2358 * then check whether the name is in the namespace or not, if not, add the
2359 * missing name entry back to namespace. If the linkEA entry is invalid,
2360 * then remove it and insert the @child in the .lustre/lost+found/MDTxxxx/
2363 * \param[in] env pointer to the thread context
2364 * \param[in] com pointer to the lfsck component
2365 * \param[in] child pointer to the directory to be double scanned
2366 * \param[in] pfid the FID corresponding to the ".." entry
2367 * \param[in] ldata pointer to the linkEA data for the given @child
2368 * \param[in,out] lh ldlm lock handler for the given @child
2369 * \param[out] type to tell the caller what the inconsistency is
2370 * \param[in] retry if found inconsistency, but the caller does not hold
2371 * ldlm lock on the @child, then set @retry as true
2373 * \retval positive number for repaired cases
2374 * \retval 0 if nothing to be repaired
2375 * \retval negative error number on failure
2378 lfsck_namespace_dsd_single(const struct lu_env *env,
2379 struct lfsck_component *com,
2380 struct dt_object *child,
2381 const struct lu_fid *pfid,
2382 struct linkea_data *ldata,
2383 struct lustre_handle *lh,
2384 enum lfsck_namespace_inconsistency_type *type,
2387 struct lfsck_thread_info *info = lfsck_env_info(env);
2388 struct lu_name *cname = &info->lti_name;
2389 const struct lu_fid *cfid = lfsck_dto2fid(child);
2391 struct lfsck_namespace *ns = com->lc_file_ram;
2392 struct lfsck_instance *lfsck = com->lc_lfsck;
2393 struct dt_object *parent = NULL;
2394 struct lmv_mds_md_v1 *lmv;
2398 lfsck_namespace_unpack_linkea_entry(ldata, cname, &tfid, info->lti_key);
2399 /* The unique linkEA entry with bad parent will be handled as orphan. */
2400 if (!fid_is_sane(&tfid)) {
2401 if (!lustre_handle_is_used(lh) && retry != NULL)
2404 rc = lfsck_namespace_dsd_orphan(env, com, child,
2410 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
2412 GOTO(out, rc = PTR_ERR(parent));
2414 /* We trust the unique linkEA entry in spite of whether it matches the
2415 * ".." name entry or not. Because even if the linkEA entry is wrong
2416 * and the ".." name entry is right, we still cannot know via which
2417 * name entry the child will be referenced, since all known entries
2418 * have been verified during the first-stage scanning. */
2419 if (!dt_object_exists(parent)) {
2420 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2421 * has ever tried to verify some remote MDT-object that resides
2422 * on this MDT, but this MDT failed to respond such request. So
2423 * means there may be some remote name entry on other MDT that
2424 * references this object with another name, so we cannot know
2425 * whether this linkEA is valid or not. So keep it there and
2426 * maybe resolved when next LFSCK run. */
2427 if (ns->ln_flags & LF_INCOMPLETE)
2430 if (!lustre_handle_is_used(lh) && retry != NULL) {
2436 lfsck_ibits_unlock(lh, LCK_EX);
2439 lmv = &info->lti_lmv;
2440 rc = lfsck_read_stripe_lmv(env, child, lmv);
2441 if (rc != 0 && rc != -ENODATA)
2444 if (rc == -ENODATA || lmv->lmv_magic != LMV_MAGIC_STRIPE) {
2446 } else if (lfsck_shard_name_to_index(env,
2447 cname->ln_name, cname->ln_namelen,
2448 S_IFDIR, cfid) < 0) {
2449 /* It is an invalid name entry, we
2450 * cannot trust the parent also. */
2451 rc = lfsck_namespace_shrink_linkea(env, com, child,
2452 ldata, cname, &tfid, true);
2456 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2457 "-"DFID, PFID(pfid));
2458 rc = lfsck_namespace_insert_orphan(env, com, child,
2459 info->lti_tmpbuf, "S", NULL);
2464 /* Create the lost parent as an orphan. */
2465 rc = lfsck_namespace_create_orphan_dir(env, com, parent, lmv);
2467 /* Add the missing name entry to the parent. */
2468 rc = lfsck_namespace_insert_normal(env, com, parent,
2469 child, cname->ln_name);
2470 if (unlikely(rc == -EEXIST)) {
2471 /* Unfortunately, someone reused the name
2472 * under the parent by race. So we have
2473 * to remove the linkEA entry from
2474 * current child object. It means that the
2475 * LFSCK cannot recover the system
2476 * totally back to its original status,
2477 * but it is necessary to make the
2478 * current system to be consistent. */
2479 rc = lfsck_namespace_shrink_linkea(env,
2481 cname, &tfid, true);
2483 snprintf(info->lti_tmpbuf,
2484 sizeof(info->lti_tmpbuf),
2485 "-"DFID, PFID(pfid));
2486 rc = lfsck_namespace_insert_orphan(env,
2487 com, child, info->lti_tmpbuf,
2496 /* The unique linkEA entry with bad parent will be handled as orphan. */
2497 if (unlikely(!dt_try_as_dir(env, parent))) {
2498 if (!lustre_handle_is_used(lh) && retry != NULL)
2501 rc = lfsck_namespace_dsd_orphan(env, com, child,
2507 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
2508 (const struct dt_key *)cname->ln_name, BYPASS_CAPA);
2509 if (rc == -ENOENT) {
2510 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2511 * has ever tried to verify some remote MDT-object that resides
2512 * on this MDT, but this MDT failed to respond such request. So
2513 * means there may be some remote name entry on other MDT that
2514 * references this object with another name, so we cannot know
2515 * whether this linkEA is valid or not. So keep it there and
2516 * maybe resolved when next LFSCK run. */
2517 if (ns->ln_flags & LF_INCOMPLETE)
2520 if (!lustre_handle_is_used(lh) && retry != NULL) {
2526 lfsck_ibits_unlock(lh, LCK_EX);
2527 rc = lfsck_namespace_check_name(env, parent, child, cname);
2534 /* It is an invalid name entry, drop it. */
2535 if (unlikely(rc > 0)) {
2536 rc = lfsck_namespace_shrink_linkea(env, com, child,
2537 ldata, cname, &tfid, true);
2539 snprintf(info->lti_tmpbuf,
2540 sizeof(info->lti_tmpbuf),
2541 "-"DFID, PFID(pfid));
2542 rc = lfsck_namespace_insert_orphan(env, com,
2543 child, info->lti_tmpbuf, "D", NULL);
2549 /* Add the missing name entry back to the namespace. */
2550 rc = lfsck_namespace_insert_normal(env, com, parent, child,
2552 if (unlikely(rc == -ESTALE))
2553 /* It may happen when the remote object has been
2554 * removed, but the local MDT is not aware of that. */
2557 if (unlikely(rc == -EEXIST)) {
2558 /* Unfortunately, someone reused the name under the
2559 * parent by race. So we have to remove the linkEA
2560 * entry from current child object. It means that the
2561 * LFSCK cannot recover the system totally back to
2562 * its original status, but it is necessary to make
2563 * the current system to be consistent.
2565 * It also may be because of the LFSCK found some
2566 * internal status of create operation. Under such
2567 * case, nothing to be done. */
2568 rc = lfsck_namespace_shrink_linkea_cond(env, com,
2569 parent, child, ldata, cname, &tfid);
2571 snprintf(info->lti_tmpbuf,
2572 sizeof(info->lti_tmpbuf),
2573 "-"DFID, PFID(pfid));
2574 rc = lfsck_namespace_insert_orphan(env, com,
2575 child, info->lti_tmpbuf, "D", NULL);
2585 if (!lu_fid_eq(&tfid, cfid)) {
2586 if (!lustre_handle_is_used(lh) && retry != NULL) {
2592 lfsck_ibits_unlock(lh, LCK_EX);
2593 /* The name entry references another MDT-object that
2594 * may be created by the LFSCK for repairing dangling
2595 * name entry. Try to replace it. */
2596 rc = lfsck_namespace_replace_cond(env, com, parent, child,
2599 rc = lfsck_namespace_dsd_orphan(env, com, child,
2605 if (fid_is_zero(pfid))
2608 /* The ".." name entry is wrong, update it. */
2609 if (!lu_fid_eq(pfid, lfsck_dto2fid(parent))) {
2610 if (!lustre_handle_is_used(lh) && retry != NULL) {
2616 *type = LNIT_UNMATCHED_PAIRS;
2617 rc = lfsck_namespace_repair_unmatched_pairs(env, com, child,
2618 lfsck_dto2fid(parent), cname);
2624 if (parent != NULL && !IS_ERR(parent))
2625 lfsck_object_put(env, parent);
2631 * Double Scan Directory object for multiple linkEA entries case.
2633 * The given @child has multiple linkEA entries. There is at most one linkEA
2634 * entry will be valid, all the others will be removed. Firstly, the function
2635 * will try to find out the linkEA entry for which the name entry exists under
2636 * the given parent (@pfid). If there is no linkEA entry that matches the given
2637 * ".." name entry, then tries to find out the first linkEA entry that both the
2638 * parent and the name entry exist to rebuild a new ".." name entry.
2640 * \param[in] env pointer to the thread context
2641 * \param[in] com pointer to the lfsck component
2642 * \param[in] child pointer to the directory to be double scanned
2643 * \param[in] pfid the FID corresponding to the ".." entry
2644 * \param[in] ldata pointer to the linkEA data for the given @child
2645 * \param[in,out] lh ldlm lock handler for the given @child
2646 * \param[out] type to tell the caller what the inconsistency is
2647 * \param[in] lpf true if the ".." entry is under lost+found/MDTxxxx/
2649 * \retval positive number for repaired cases
2650 * \retval 0 if nothing to be repaired
2651 * \retval negative error number on failure
2654 lfsck_namespace_dsd_multiple(const struct lu_env *env,
2655 struct lfsck_component *com,
2656 struct dt_object *child,
2657 const struct lu_fid *pfid,
2658 struct linkea_data *ldata,
2659 struct lustre_handle *lh,
2660 enum lfsck_namespace_inconsistency_type *type,
2663 struct lfsck_thread_info *info = lfsck_env_info(env);
2664 struct lu_name *cname = &info->lti_name;
2665 const struct lu_fid *cfid = lfsck_dto2fid(child);
2666 struct lu_fid *pfid2 = &info->lti_fid3;
2668 struct lfsck_namespace *ns = com->lc_file_ram;
2669 struct lfsck_instance *lfsck = com->lc_lfsck;
2670 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
2671 struct dt_object *parent = NULL;
2672 struct linkea_data ldata_new = { NULL };
2673 int dirent_count = 0;
2674 int linkea_count = 0;
2680 while (ldata->ld_lee != NULL) {
2681 lfsck_namespace_unpack_linkea_entry(ldata, cname, &tfid,
2683 /* Drop repeated linkEA entries. */
2684 lfsck_namespace_filter_linkea_entry(ldata, cname, &tfid, true);
2685 /* Drop invalid linkEA entry. */
2686 if (!fid_is_sane(&tfid)) {
2687 linkea_del_buf(ldata, cname);
2692 /* If current dotdot is the .lustre/lost+found/MDTxxxx/,
2693 * then it is possible that: the directry object has ever
2694 * been lost, but its name entry was there. In the former
2695 * LFSCK run, during the first-stage scanning, the LFSCK
2696 * found the dangling name entry, but it did not recreate
2697 * the lost object, and when moved to the second-stage
2698 * scanning, some children objects of the lost directory
2699 * object were found, then the LFSCK recreated such lost
2700 * directory object as an orphan.
2702 * When the LFSCK runs again, if the dangling name is still
2703 * there, the LFSCK should move the orphan directory object
2704 * back to the normal namespace. */
2705 if (!lpf && !lu_fid_eq(pfid, &tfid) && once) {
2706 linkea_next_entry(ldata);
2710 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
2712 RETURN(PTR_ERR(parent));
2714 if (!dt_object_exists(parent)) {
2715 lfsck_object_put(env, parent);
2716 if (ldata->ld_leh->leh_reccount > 1) {
2717 /* If it is NOT the last linkEA entry, then
2718 * there is still other chance to make the
2719 * child to be visible via other parent, then
2720 * remove this linkEA entry. */
2721 linkea_del_buf(ldata, cname);
2729 /* The linkEA entry with bad parent will be removed. */
2730 if (unlikely(!dt_try_as_dir(env, parent))) {
2731 lfsck_object_put(env, parent);
2732 linkea_del_buf(ldata, cname);
2737 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
2738 (const struct dt_key *)cname->ln_name,
2740 *pfid2 = *lfsck_dto2fid(parent);
2741 if (rc == -ENOENT) {
2742 lfsck_object_put(env, parent);
2743 linkea_next_entry(ldata);
2748 lfsck_object_put(env, parent);
2753 if (lu_fid_eq(&tfid, cfid)) {
2754 lfsck_object_put(env, parent);
2755 if (!lu_fid_eq(pfid, pfid2)) {
2756 *type = LNIT_UNMATCHED_PAIRS;
2757 rc = lfsck_namespace_repair_unmatched_pairs(env,
2758 com, child, pfid2, cname);
2764 /* It is the most common case that we find the
2765 * name entry corresponding to the linkEA entry
2766 * that matches the ".." name entry. */
2767 rc = linkea_data_new(&ldata_new, &info->lti_big_buf);
2771 rc = linkea_add_buf(&ldata_new, cname, pfid2);
2775 rc = lfsck_namespace_rebuild_linkea(env, com, child,
2780 linkea_del_buf(ldata, cname);
2782 linkea_first_entry(ldata);
2783 /* There may be some invalid dangling name entries under
2784 * other parent directories, remove all of them. */
2785 while (ldata->ld_lee != NULL) {
2786 lfsck_namespace_unpack_linkea_entry(ldata,
2787 cname, &tfid, info->lti_key);
2788 if (!fid_is_sane(&tfid))
2791 parent = lfsck_object_find_bottom(env, lfsck,
2793 if (IS_ERR(parent)) {
2794 rc = PTR_ERR(parent);
2795 if (rc != -ENOENT &&
2796 bk->lb_param & LPF_FAILOUT)
2802 if (!dt_object_exists(parent)) {
2803 lfsck_object_put(env, parent);
2807 rc = lfsck_namespace_repair_dirent(env, com,
2808 parent, child, cname->ln_name,
2809 cname->ln_name, S_IFDIR, false, true);
2810 lfsck_object_put(env, parent);
2812 if (bk->lb_param & LPF_FAILOUT)
2821 linkea_del_buf(ldata, cname);
2824 ns->ln_dirent_repaired += dirent_count;
2829 lfsck_ibits_unlock(lh, LCK_EX);
2830 /* The name entry references another MDT-object that may be
2831 * created by the LFSCK for repairing dangling name entry.
2832 * Try to replace it. */
2833 rc = lfsck_namespace_replace_cond(env, com, parent, child,
2835 lfsck_object_put(env, parent);
2842 linkea_del_buf(ldata, cname);
2845 linkea_first_entry(ldata);
2846 if (ldata->ld_leh->leh_reccount == 1) {
2847 rc = lfsck_namespace_dsd_single(env, com, child, pfid, ldata,
2850 if (rc == 0 && fid_is_zero(pfid) && linkea_count > 0)
2851 rc = lfsck_namespace_rebuild_linkea(env, com, child,
2857 /* All linkEA entries are invalid and removed, then handle the @child
2859 if (ldata->ld_leh->leh_reccount == 0) {
2860 rc = lfsck_namespace_dsd_orphan(env, com, child, pfid, lh,
2866 /* If the dangling name entry for the orphan directory object has
2867 * been remvoed, then just check whether the directory object is
2868 * still under the .lustre/lost+found/MDTxxxx/ or not. */
2874 /* There is no linkEA entry that matches the ".." name entry. Find
2875 * the first linkEA entry that both parent and name entry exist to
2876 * rebuild a new ".." name entry. */
2886 * Repair the object's nlink attribute.
2888 * If all the known name entries have been verified, then the object's hard
2889 * link attribute should match the object's linkEA entries count unless the
2890 * object's has too much hard link to be recorded in the linkEA. Such cases
2891 * should have been marked in the LFSCK trace file. Otherwise, trust the
2892 * linkEA to update the object's nlink attribute.
2894 * \param[in] env pointer to the thread context
2895 * \param[in] com pointer to the lfsck component
2896 * \param[in] obj pointer to the dt_object to be handled
2897 * \param[in,out] la pointer to buffer to object's attribute before
2898 * and after the repairing
2900 * \retval positive number for repaired cases
2901 * \retval 0 if nothing to be repaired
2902 * \retval negative error number on failure
2904 static int lfsck_namespace_repair_nlink(const struct lu_env *env,
2905 struct lfsck_component *com,
2906 struct dt_object *obj,
2909 struct lfsck_thread_info *info = lfsck_env_info(env);
2910 struct lu_fid *tfid = &info->lti_fid3;
2911 struct lfsck_namespace *ns = com->lc_file_ram;
2912 struct lfsck_instance *lfsck = com->lc_lfsck;
2913 struct dt_device *dev = lfsck_obj2dev(obj);
2914 const struct lu_fid *cfid = lfsck_dto2fid(obj);
2915 struct thandle *th = NULL;
2916 struct linkea_data ldata = { NULL };
2917 struct lustre_handle lh = { 0 };
2918 __u32 old = la->la_nlink;
2924 LASSERT(!dt_object_remote(obj));
2925 LASSERT(S_ISREG(lfsck_object_type(obj)));
2927 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
2928 MDS_INODELOCK_UPDATE, LCK_PW);
2932 th = dt_trans_create(env, dev);
2934 GOTO(log, rc = PTR_ERR(th));
2936 la->la_valid = LA_NLINK;
2937 rc = dt_declare_attr_set(env, obj, la, th);
2941 rc = dt_trans_start_local(env, dev, th);
2945 dt_write_lock(env, obj, 0);
2946 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
2947 * ever tried to verify some remote MDT-object that resides on this
2948 * MDT, but this MDT failed to respond such request. So means there
2949 * may be some remote name entry on other MDT that references this
2950 * object with another name, so we cannot know whether this linkEA
2951 * is valid or not. So keep it there and maybe resolved when next
2953 if (ns->ln_flags & LF_INCOMPLETE)
2954 GOTO(unlock, rc = 0);
2956 fid_cpu_to_be(tfid, cfid);
2957 idx = lfsck_sub_trace_file_fid2idx(cfid);
2958 rc = dt_lookup(env, com->lc_sub_trace_objs[idx].lsto_obj,
2959 (struct dt_rec *)&flags, (const struct dt_key *)tfid,
2964 if (flags & LNTF_SKIP_NLINK)
2965 GOTO(unlock, rc = 0);
2967 rc = dt_attr_get(env, obj, la, BYPASS_CAPA);
2969 GOTO(unlock, rc = (rc == -ENOENT ? 0 : rc));
2971 rc = lfsck_links_read2(env, obj, &ldata);
2973 GOTO(unlock, rc = (rc == -ENODATA ? 0 : rc));
2975 if (la->la_nlink == ldata.ld_leh->leh_reccount ||
2976 unlikely(la->la_nlink == 0))
2977 GOTO(unlock, rc = 0);
2979 la->la_nlink = ldata.ld_leh->leh_reccount;
2980 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2981 GOTO(unlock, rc = 1);
2983 rc = dt_attr_set(env, obj, la, th, BYPASS_CAPA);
2985 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2988 dt_write_unlock(env, obj);
2991 dt_trans_stop(env, dev, th);
2994 lfsck_ibits_unlock(&lh, LCK_PW);
2995 CDEBUG(D_LFSCK, "%s: namespace LFSCK repaired the object "DFID"'s "
2996 "nlink count from %u to %u: rc = %d\n",
2997 lfsck_lfsck2name(lfsck), PFID(cfid), old, la->la_nlink, rc);
3000 ns->ln_flags |= LF_INCONSISTENT;
3006 * Double scan the directory object for namespace LFSCK.
3008 * This function will verify the <parent, child> pairs in the namespace tree:
3009 * the parent references the child via some name entry that should be in the
3010 * child's linkEA entry, the child should back references the parent via its
3013 * The LFSCK will scan every linkEA entry in turn until find out the first
3014 * matched pairs. If found, then all other linkEA entries will be dropped.
3015 * If all the linkEA entries cannot match the ".." name entry, then there
3016 * are serveral possible cases:
3018 * 1) If there is only one linkEA entry, then trust it as long as the PFID
3019 * in the linkEA entry is valid.
3021 * 2) If there are multiple linkEA entries, then try to find the linkEA
3022 * that matches the ".." name entry. If found, then all other entries
3023 * are invalid; otherwise, it is quite possible that the ".." name entry
3024 * is corrupted. Under such case, the LFSCK will rebuild the ".." name
3025 * entry according to the first valid linkEA entry (both the parent and
3026 * the name entry should exist).
3028 * 3) If the directory object has no (valid) linkEA entry, then the
3029 * directory object will be handled as pure orphan and inserted
3030 * in the .lustre/lost+found/MDTxxxx/ with the name:
3031 * ${self_FID}-${PFID}-D-${conflict_version}
3033 * \param[in] env pointer to the thread context
3034 * \param[in] com pointer to the lfsck component
3035 * \param[in] child pointer to the directory object to be handled
3036 * \param[in] flags to indicate the specical checking on the @child
3038 * \retval positive number for repaired cases
3039 * \retval 0 if nothing to be repaired
3040 * \retval negative error number on failure
3042 static int lfsck_namespace_double_scan_dir(const struct lu_env *env,
3043 struct lfsck_component *com,
3044 struct dt_object *child, __u8 flags)
3046 struct lfsck_thread_info *info = lfsck_env_info(env);
3047 const struct lu_fid *cfid = lfsck_dto2fid(child);
3048 struct lu_fid *pfid = &info->lti_fid2;
3049 struct lfsck_namespace *ns = com->lc_file_ram;
3050 struct lfsck_instance *lfsck = com->lc_lfsck;
3051 struct lustre_handle lh = { 0 };
3052 struct linkea_data ldata = { NULL };
3053 bool unknown = false;
3056 enum lfsck_namespace_inconsistency_type type = LNIT_BAD_LINKEA;
3060 LASSERT(!dt_object_remote(child));
3062 if (flags & LNTF_UNCERTAIN_LMV) {
3063 if (flags & LNTF_RECHECK_NAME_HASH) {
3064 rc = lfsck_namespace_scan_shard(env, com, child);
3068 ns->ln_striped_shards_scanned++;
3070 ns->ln_striped_shards_skipped++;
3074 flags &= ~(LNTF_RECHECK_NAME_HASH | LNTF_UNCERTAIN_LMV);
3078 if (flags & (LNTF_CHECK_LINKEA | LNTF_CHECK_PARENT) &&
3079 !(lfsck->li_bookmark_ram.lb_param & LPF_ALL_TGT)) {
3080 CDEBUG(D_LFSCK, "%s: some MDT(s) maybe NOT take part in the"
3081 "the namespace LFSCK, then the LFSCK cannot guarantee"
3082 "all the name entries have been verified in first-stage"
3083 "scanning. So have to skip orphan related handling for"
3084 "the directory object "DFID" with remote name entry\n",
3085 lfsck_lfsck2name(lfsck), PFID(cfid));
3090 if (unlikely(!dt_try_as_dir(env, child)))
3091 GOTO(out, rc = -ENOTDIR);
3093 /* We only take ldlm lock on the @child when required. When the
3094 * logic comes here for the first time, it is always false. */
3098 rc = lfsck_ibits_lock(env, lfsck, child, &lh,
3099 MDS_INODELOCK_UPDATE |
3100 MDS_INODELOCK_XATTR, LCK_EX);
3105 dt_read_lock(env, child, 0);
3106 if (unlikely(lfsck_is_dead_obj(child))) {
3107 dt_read_unlock(env, child);
3112 rc = dt_lookup(env, child, (struct dt_rec *)pfid,
3113 (const struct dt_key *)dotdot, BYPASS_CAPA);
3115 if (rc != -ENOENT && rc != -ENODATA && rc != -EINVAL) {
3116 dt_read_unlock(env, child);
3121 if (!lustre_handle_is_used(&lh)) {
3122 dt_read_unlock(env, child);
3127 } else if (lfsck->li_lpf_obj != NULL &&
3128 lu_fid_eq(pfid, lfsck_dto2fid(lfsck->li_lpf_obj))) {
3130 } else if (unlikely(!fid_is_sane(pfid))) {
3134 rc = lfsck_links_read(env, child, &ldata);
3135 dt_read_unlock(env, child);
3137 if (rc != -ENODATA && rc != -EINVAL)
3140 if (!lustre_handle_is_used(&lh))
3143 if (rc == -EINVAL && !fid_is_zero(pfid)) {
3144 /* Remove the corrupted linkEA. */
3145 rc = lfsck_namespace_links_remove(env, com, child);
3147 /* Here, because of the crashed linkEA, we
3148 * cannot know whether there is some parent
3149 * that references the child directory via
3150 * some name entry or not. So keep it there,
3151 * when the LFSCK run next time, if there is
3152 * some parent that references this object,
3153 * then the LFSCK can rebuild the linkEA;
3154 * otherwise, this object will be handled
3155 * as orphan as above. */
3158 /* 1. If we have neither ".." nor linkEA,
3159 * then it is an orphan.
3161 * 2. If we only have the ".." name entry,
3162 * but no parent references this child
3163 * directory, then handle it as orphan. */
3164 lfsck_ibits_unlock(&lh, LCK_EX);
3165 type = LNIT_MUL_REF;
3167 /* If the LFSCK is marked as LF_INCOMPLETE,
3168 * then means some MDT has ever tried to
3169 * verify some remote MDT-object that resides
3170 * on this MDT, but this MDT failed to respond
3171 * such request. So means there may be some
3172 * remote name entry on other MDT that
3173 * references this object with another name,
3174 * so we cannot know whether this linkEA is
3175 * valid or not. So keep it there and maybe
3176 * resolved when next LFSCK run. */
3177 if (ns->ln_flags & LF_INCOMPLETE)
3180 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
3181 "-"DFID, PFID(pfid));
3182 rc = lfsck_namespace_insert_orphan(env, com, child,
3183 info->lti_tmpbuf, "D", NULL);
3189 linkea_first_entry(&ldata);
3190 /* This is the most common case: the object has unique linkEA entry. */
3191 if (ldata.ld_leh->leh_reccount == 1) {
3192 rc = lfsck_namespace_dsd_single(env, com, child, pfid, &ldata,
3193 &lh, &type, &retry);
3195 LASSERT(!lustre_handle_is_used(&lh));
3204 if (!lustre_handle_is_used(&lh))
3207 if (unlikely(ldata.ld_leh->leh_reccount == 0)) {
3208 rc = lfsck_namespace_dsd_orphan(env, com, child, pfid, &lh,
3214 /* When we come here, the cases usually like that:
3215 * 1) The directory object has a corrupted linkEA entry. During the
3216 * first-stage scanning, the LFSCK cannot know such corruption,
3217 * then it appends the right linkEA entry according to the found
3218 * name entry after the bad one.
3220 * 2) The directory object has a right linkEA entry. During the
3221 * first-stage scanning, the LFSCK finds some bad name entry,
3222 * but the LFSCK cannot aware that at that time, then it adds
3223 * the bad linkEA entry for further processing. */
3224 rc = lfsck_namespace_dsd_multiple(env, com, child, pfid, &ldata,
3230 lfsck_ibits_unlock(&lh, LCK_EX);
3233 case LNIT_BAD_LINKEA:
3234 ns->ln_linkea_repaired++;
3236 case LNIT_UNMATCHED_PAIRS:
3237 ns->ln_unmatched_pairs_repaired++;
3240 ns->ln_mul_ref_repaired++;
3248 ns->ln_unknown_inconsistency++;
3254 * Double scan the MDT-object for namespace LFSCK.
3256 * If the MDT-object contains invalid or repeated linkEA entries, then drop
3257 * those entries from the linkEA; if the linkEA becomes empty or the object
3258 * has no linkEA, then it is an orphan and will be added into the directory
3259 * .lustre/lost+found/MDTxxxx/; if the remote parent is lost, then recreate
3260 * the remote parent; if the name entry corresponding to some linkEA entry
3261 * is lost, then add the name entry back to the namespace.
3263 * \param[in] env pointer to the thread context
3264 * \param[in] com pointer to the lfsck component
3265 * \param[in] child pointer to the dt_object to be handled
3266 * \param[in] flags some hints to indicate how the @child should be handled
3268 * \retval positive number for repaired cases
3269 * \retval 0 if nothing to be repaired
3270 * \retval negative error number on failure
3272 static int lfsck_namespace_double_scan_one(const struct lu_env *env,
3273 struct lfsck_component *com,
3274 struct dt_object *child, __u8 flags)
3276 struct lfsck_thread_info *info = lfsck_env_info(env);
3277 struct lu_attr *la = &info->lti_la;
3278 struct lu_name *cname = &info->lti_name;
3279 struct lu_fid *pfid = &info->lti_fid;
3280 struct lu_fid *cfid = &info->lti_fid2;
3281 struct lfsck_instance *lfsck = com->lc_lfsck;
3282 struct lfsck_namespace *ns = com->lc_file_ram;
3283 struct dt_object *parent = NULL;
3284 struct linkea_data ldata = { NULL };
3285 bool repaired = false;
3290 dt_read_lock(env, child, 0);
3291 if (unlikely(lfsck_is_dead_obj(child))) {
3292 dt_read_unlock(env, child);
3297 if (S_ISDIR(lfsck_object_type(child))) {
3298 dt_read_unlock(env, child);
3299 rc = lfsck_namespace_double_scan_dir(env, com, child, flags);
3304 rc = lfsck_links_read(env, child, &ldata);
3305 dt_read_unlock(env, child);
3309 linkea_first_entry(&ldata);
3310 while (ldata.ld_lee != NULL) {
3311 lfsck_namespace_unpack_linkea_entry(&ldata, cname, pfid,
3313 rc = lfsck_namespace_filter_linkea_entry(&ldata, cname, pfid,
3315 /* Found repeated linkEA entries */
3317 rc = lfsck_namespace_shrink_linkea(env, com, child,
3318 &ldata, cname, pfid, false);
3330 /* Invalid PFID in the linkEA entry. */
3331 if (!fid_is_sane(pfid)) {
3332 rc = lfsck_namespace_shrink_linkea(env, com, child,
3333 &ldata, cname, pfid, true);
3343 parent = lfsck_object_find_bottom(env, lfsck, pfid);
3345 GOTO(out, rc = PTR_ERR(parent));
3347 if (!dt_object_exists(parent)) {
3350 if (ldata.ld_leh->leh_reccount > 1) {
3351 /* If it is NOT the last linkEA entry, then
3352 * there is still other chance to make the
3353 * child to be visible via other parent, then
3354 * remove this linkEA entry. */
3355 rc = lfsck_namespace_shrink_linkea(env, com,
3356 child, &ldata, cname, pfid, true);
3358 /* If the LFSCK is marked as LF_INCOMPLETE,
3359 * then means some MDT has ever tried to
3360 * verify some remote MDT-object that resides
3361 * on this MDT, but this MDT failed to respond
3362 * such request. So means there may be some
3363 * remote name entry on other MDT that
3364 * references this object with another name,
3365 * so we cannot know whether this linkEA is
3366 * valid or not. So keep it there and maybe
3367 * resolved when next LFSCK run. */
3368 if (ns->ln_flags & LF_INCOMPLETE) {
3369 lfsck_object_put(env, parent);
3374 /* Create the lost parent as an orphan. */
3375 rc = lfsck_namespace_create_orphan_dir(env, com,
3378 lfsck_object_put(env, parent);
3386 /* Add the missing name entry to the parent. */
3387 rc = lfsck_namespace_insert_normal(env, com,
3388 parent, child, cname->ln_name);
3389 if (unlikely(rc == -EEXIST))
3390 /* Unfortunately, someone reused the
3391 * name under the parent by race. So we
3392 * have to remove the linkEA entry from
3393 * current child object. It means that
3394 * the LFSCK cannot recover the system
3395 * totally back to its original status,
3396 * but it is necessary to make the
3397 * current system to be consistent. */
3398 rc = lfsck_namespace_shrink_linkea(env,
3402 linkea_next_entry(&ldata);
3405 lfsck_object_put(env, parent);
3415 /* The linkEA entry with bad parent will be removed. */
3416 if (unlikely(!dt_try_as_dir(env, parent))) {
3417 lfsck_object_put(env, parent);
3418 rc = lfsck_namespace_shrink_linkea(env, com, child,
3419 &ldata, cname, pfid, true);
3429 rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
3430 (const struct dt_key *)cname->ln_name,
3432 if (rc != 0 && rc != -ENOENT) {
3433 lfsck_object_put(env, parent);
3439 if (lu_fid_eq(cfid, lfsck_dto2fid(child))) {
3440 /* It is the most common case that we
3441 * find the name entry corresponding
3442 * to the linkEA entry. */
3443 lfsck_object_put(env, parent);
3444 linkea_next_entry(&ldata);
3446 /* The name entry references another
3447 * MDT-object that may be created by
3448 * the LFSCK for repairing dangling
3449 * name entry. Try to replace it. */
3450 rc = lfsck_namespace_replace_cond(env, com,
3451 parent, child, cfid, cname);
3452 lfsck_object_put(env, parent);
3458 linkea_next_entry(&ldata);
3460 rc = lfsck_namespace_shrink_linkea(env,
3474 rc = dt_attr_get(env, child, la, BYPASS_CAPA);
3478 /* If there is no name entry in the parent dir and the object
3479 * link count is less than the linkea entries count, then the
3480 * linkea entry should be removed. */
3481 if (ldata.ld_leh->leh_reccount > la->la_nlink) {
3482 rc = lfsck_namespace_shrink_linkea_cond(env, com,
3483 parent, child, &ldata, cname, pfid);
3484 lfsck_object_put(env, parent);
3494 /* If the LFSCK is marked as LF_INCOMPLETE, then means some
3495 * MDT has ever tried to verify some remote MDT-object that
3496 * resides on this MDT, but this MDT failed to respond such
3497 * request. So means there may be some remote name entry on
3498 * other MDT that references this object with another name,
3499 * so we cannot know whether this linkEA is valid or not.
3500 * So keep it there and maybe resolved when next LFSCK run. */
3501 if (ns->ln_flags & LF_INCOMPLETE) {
3502 lfsck_object_put(env, parent);
3507 rc = lfsck_namespace_check_name(env, parent, child, cname);
3512 lfsck_object_put(env, parent);
3517 /* It is an invalid name entry, drop it. */
3518 if (unlikely(rc > 0)) {
3519 lfsck_object_put(env, parent);
3520 rc = lfsck_namespace_shrink_linkea(env, com, child,
3521 &ldata, cname, pfid, true);
3531 /* Add the missing name entry back to the namespace. */
3532 rc = lfsck_namespace_insert_normal(env, com, parent, child,
3534 if (unlikely(rc == -ESTALE))
3535 /* It may happen when the remote object has been
3536 * removed, but the local MDT is not aware of that. */
3539 if (unlikely(rc == -EEXIST))
3540 /* Unfortunately, someone reused the name under the
3541 * parent by race. So we have to remove the linkEA
3542 * entry from current child object. It means that the
3543 * LFSCK cannot recover the system totally back to
3544 * its original status, but it is necessary to make
3545 * the current system to be consistent.
3547 * It also may be because of the LFSCK found some
3548 * internal status of create operation. Under such
3549 * case, nothing to be done. */
3550 rc = lfsck_namespace_shrink_linkea_cond(env, com,
3551 parent, child, &ldata, cname, pfid);
3553 linkea_next_entry(&ldata);
3555 lfsck_object_put(env, parent);
3566 if (rc < 0 && rc != -ENODATA)
3570 LASSERT(ldata.ld_leh != NULL);
3572 count = ldata.ld_leh->leh_reccount;
3576 /* If the LFSCK is marked as LF_INCOMPLETE, then means some
3577 * MDT has ever tried to verify some remote MDT-object that
3578 * resides on this MDT, but this MDT failed to respond such
3579 * request. So means there may be some remote name entry on
3580 * other MDT that references this object with another name,
3581 * so we cannot know whether this linkEA is valid or not.
3582 * So keep it there and maybe resolved when next LFSCK run. */
3583 if (!(ns->ln_flags & LF_INCOMPLETE)) {
3584 /* If the child becomes orphan, then insert it into
3585 * the global .lustre/lost+found/MDTxxxx directory. */
3586 rc = lfsck_namespace_insert_orphan(env, com, child,
3592 ns->ln_mul_ref_repaired++;
3597 rc = dt_attr_get(env, child, la, BYPASS_CAPA);
3601 if (la->la_nlink != 0 && la->la_nlink != count) {
3602 rc = lfsck_namespace_repair_nlink(env, com, child, la);
3604 ns->ln_objs_nlink_repaired++;
3611 if (la->la_nlink > 1)
3612 ns->ln_mul_linked_repaired++;
3621 static void lfsck_namespace_dump_statistics(struct seq_file *m,
3622 struct lfsck_namespace *ns,
3623 __u64 checked_phase1,
3624 __u64 checked_phase2,
3628 seq_printf(m, "checked_phase1: "LPU64"\n"
3629 "checked_phase2: "LPU64"\n"
3630 "updated_phase1: "LPU64"\n"
3631 "updated_phase2: "LPU64"\n"
3632 "failed_phase1: "LPU64"\n"
3633 "failed_phase2: "LPU64"\n"
3634 "directories: "LPU64"\n"
3635 "dirent_repaired: "LPU64"\n"
3636 "linkea_repaired: "LPU64"\n"
3637 "nlinks_repaired: "LPU64"\n"
3638 "multiple_linked_checked: "LPU64"\n"
3639 "multiple_linked_repaired: "LPU64"\n"
3640 "unknown_inconsistency: "LPU64"\n"
3641 "unmatched_pairs_repaired: "LPU64"\n"
3642 "dangling_repaired: "LPU64"\n"
3643 "multiple_referenced_repaired: "LPU64"\n"
3644 "bad_file_type_repaired: "LPU64"\n"
3645 "lost_dirent_repaired: "LPU64"\n"
3646 "local_lost_found_scanned: "LPU64"\n"
3647 "local_lost_found_moved: "LPU64"\n"
3648 "local_lost_found_skipped: "LPU64"\n"
3649 "local_lost_found_failed: "LPU64"\n"
3650 "striped_dirs_scanned: "LPU64"\n"
3651 "striped_dirs_repaired: "LPU64"\n"
3652 "striped_dirs_failed: "LPU64"\n"
3653 "striped_dirs_disabled: "LPU64"\n"
3654 "striped_dirs_skipped: "LPU64"\n"
3655 "striped_shards_scanned: "LPU64"\n"
3656 "striped_shards_repaired: "LPU64"\n"
3657 "striped_shards_failed: "LPU64"\n"
3658 "striped_shards_skipped: "LPU64"\n"
3659 "name_hash_repaired: "LPU64"\n"
3660 "success_count: %u\n"
3661 "run_time_phase1: %u seconds\n"
3662 "run_time_phase2: %u seconds\n",
3665 ns->ln_items_repaired,
3666 ns->ln_objs_repaired_phase2,
3667 ns->ln_items_failed,
3668 ns->ln_objs_failed_phase2,
3669 ns->ln_dirs_checked,
3670 ns->ln_dirent_repaired,
3671 ns->ln_linkea_repaired,
3672 ns->ln_objs_nlink_repaired,
3673 ns->ln_mul_linked_checked,
3674 ns->ln_mul_linked_repaired,
3675 ns->ln_unknown_inconsistency,
3676 ns->ln_unmatched_pairs_repaired,
3677 ns->ln_dangling_repaired,
3678 ns->ln_mul_ref_repaired,
3679 ns->ln_bad_type_repaired,
3680 ns->ln_lost_dirent_repaired,
3681 ns->ln_local_lpf_scanned,
3682 ns->ln_local_lpf_moved,
3683 ns->ln_local_lpf_skipped,
3684 ns->ln_local_lpf_failed,
3685 ns->ln_striped_dirs_scanned,
3686 ns->ln_striped_dirs_repaired,
3687 ns->ln_striped_dirs_failed,
3688 ns->ln_striped_dirs_disabled,
3689 ns->ln_striped_dirs_skipped,
3690 ns->ln_striped_shards_scanned,
3691 ns->ln_striped_shards_repaired,
3692 ns->ln_striped_shards_failed,
3693 ns->ln_striped_shards_skipped,
3694 ns->ln_name_hash_repaired,
3695 ns->ln_success_count,
3700 static void lfsck_namespace_release_lmv(const struct lu_env *env,
3701 struct lfsck_component *com)
3703 struct lfsck_instance *lfsck = com->lc_lfsck;
3704 struct lfsck_namespace *ns = com->lc_file_ram;
3706 while (!list_empty(&lfsck->li_list_lmv)) {
3707 struct lfsck_lmv_unit *llu;
3708 struct lfsck_lmv *llmv;
3710 llu = list_entry(lfsck->li_list_lmv.next,
3711 struct lfsck_lmv_unit, llu_link);
3712 llmv = &llu->llu_lmv;
3714 LASSERTF(atomic_read(&llmv->ll_ref) == 1,
3715 "still in using: %u\n",
3716 atomic_read(&llmv->ll_ref));
3718 ns->ln_striped_dirs_skipped++;
3719 lfsck_lmv_put(env, llmv);
3723 static int lfsck_namespace_check_for_double_scan(const struct lu_env *env,
3724 struct lfsck_component *com,
3725 struct dt_object *obj)
3727 struct lu_attr *la = &lfsck_env_info(env)->lti_la;
3730 rc = dt_attr_get(env, obj, la, BYPASS_CAPA);
3734 /* zero-linkEA object may be orphan, but it also maybe because
3735 * of upgrading. Currently, we cannot record it for double scan.
3736 * Because it may cause the LFSCK trace file to be too large. */
3738 /* "la_ctime" == 1 means that it has ever been removed from
3739 * backend /lost+found directory but not been added back to
3740 * the normal namespace yet. */
3742 if ((S_ISREG(lfsck_object_type(obj)) && la->la_nlink > 1) ||
3743 unlikely(la->la_ctime == 1))
3744 rc = lfsck_namespace_trace_update(env, com, lfsck_dto2fid(obj),
3745 LNTF_CHECK_LINKEA, true);
3750 /* namespace APIs */
3752 static int lfsck_namespace_reset(const struct lu_env *env,
3753 struct lfsck_component *com, bool init)
3755 struct lfsck_instance *lfsck = com->lc_lfsck;
3756 struct lfsck_namespace *ns = com->lc_file_ram;
3757 struct lfsck_assistant_data *lad = com->lc_data;
3758 struct dt_object *root;
3759 struct dt_object *dto;
3763 root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
3765 GOTO(log, rc = PTR_ERR(root));
3767 if (unlikely(!dt_try_as_dir(env, root)))
3768 GOTO(put, rc = -ENOTDIR);
3770 down_write(&com->lc_sem);
3772 memset(ns, 0, sizeof(*ns));
3774 __u32 count = ns->ln_success_count;
3775 __u64 last_time = ns->ln_time_last_complete;
3777 memset(ns, 0, sizeof(*ns));
3778 ns->ln_success_count = count;
3779 ns->ln_time_last_complete = last_time;
3781 ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
3782 ns->ln_status = LS_INIT;
3784 lfsck_object_put(env, com->lc_obj);
3786 dto = lfsck_namespace_load_one_trace_file(env, com, root,
3787 LFSCK_NAMESPACE, true);
3789 GOTO(out, rc = PTR_ERR(dto));
3792 rc = lfsck_namespace_load_sub_trace_files(env, com, true);
3796 lad->lad_incomplete = 0;
3797 CFS_RESET_BITMAP(lad->lad_bitmap);
3799 rc = lfsck_namespace_store(env, com, true);
3804 up_write(&com->lc_sem);
3807 lfsck_object_put(env, root);
3809 CDEBUG(D_LFSCK, "%s: namespace LFSCK reset: rc = %d\n",
3810 lfsck_lfsck2name(lfsck), rc);
3815 lfsck_namespace_fail(const struct lu_env *env, struct lfsck_component *com,
3818 struct lfsck_namespace *ns = com->lc_file_ram;
3820 down_write(&com->lc_sem);
3822 com->lc_new_checked++;
3823 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
3824 up_write(&com->lc_sem);
3827 static void lfsck_namespace_close_dir(const struct lu_env *env,
3828 struct lfsck_component *com)
3830 struct lfsck_namespace *ns = com->lc_file_ram;
3831 struct lfsck_assistant_data *lad = com->lc_data;
3832 struct lfsck_instance *lfsck = com->lc_lfsck;
3833 struct lfsck_lmv *llmv = lfsck->li_lmv;
3834 struct lfsck_namespace_req *lnr;
3836 sizeof(*lnr) + LFSCK_TMPBUF_LEN;
3837 bool wakeup = false;
3843 OBD_ALLOC(lnr, size);
3845 ns->ln_striped_dirs_skipped++;
3850 /* Generate a dummy request to indicate that all shards' name entry
3851 * in this striped directory has been scanned for the first time. */
3852 INIT_LIST_HEAD(&lnr->lnr_lar.lar_list);
3853 lnr->lnr_lar.lar_fid = *lfsck_dto2fid(lfsck->li_obj_dir);
3854 lnr->lnr_lmv = lfsck_lmv_get(llmv);
3855 lnr->lnr_fid = *lfsck_dto2fid(lfsck->li_obj_dir);
3856 lnr->lnr_oit_cookie = lfsck->li_pos_current.lp_oit_cookie;
3857 lnr->lnr_dir_cookie = MDS_DIR_END_OFF;
3858 lnr->lnr_size = size;
3860 spin_lock(&lad->lad_lock);
3861 if (lad->lad_assistant_status < 0) {
3862 spin_unlock(&lad->lad_lock);
3863 lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar);
3864 ns->ln_striped_dirs_skipped++;
3869 list_add_tail(&lnr->lnr_lar.lar_list, &lad->lad_req_list);
3870 if (lad->lad_prefetched == 0)
3873 lad->lad_prefetched++;
3874 spin_unlock(&lad->lad_lock);
3876 wake_up_all(&lad->lad_thread.t_ctl_waitq);
3881 static int lfsck_namespace_open_dir(const struct lu_env *env,
3882 struct lfsck_component *com)
3884 struct lfsck_instance *lfsck = com->lc_lfsck;
3885 struct lfsck_namespace *ns = com->lc_file_ram;
3886 struct lfsck_lmv *llmv = lfsck->li_lmv;
3893 if (llmv->ll_lmv_master) {
3894 struct lmv_mds_md_v1 *lmv = &llmv->ll_lmv;
3896 if (lmv->lmv_master_mdt_index != lfsck_dev_idx(lfsck)) {
3897 lmv->lmv_master_mdt_index =
3898 lfsck_dev_idx(lfsck);
3899 ns->ln_flags |= LF_INCONSISTENT;
3900 llmv->ll_lmv_updated = 1;
3903 rc = lfsck_namespace_verify_stripe_slave(env, com,
3904 lfsck->li_obj_dir, llmv);
3907 RETURN(rc > 0 ? 0 : rc);
3910 static int lfsck_namespace_checkpoint(const struct lu_env *env,
3911 struct lfsck_component *com, bool init)
3913 struct lfsck_instance *lfsck = com->lc_lfsck;
3914 struct lfsck_namespace *ns = com->lc_file_ram;
3918 rc = lfsck_checkpoint_generic(env, com);
3923 down_write(&com->lc_sem);
3925 ns->ln_pos_latest_start = lfsck->li_pos_checkpoint;
3927 ns->ln_pos_last_checkpoint = lfsck->li_pos_checkpoint;
3928 ns->ln_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
3929 HALF_SEC - lfsck->li_time_last_checkpoint);
3930 ns->ln_time_last_checkpoint = cfs_time_current_sec();
3931 ns->ln_items_checked += com->lc_new_checked;
3932 com->lc_new_checked = 0;
3935 rc = lfsck_namespace_store(env, com, false);
3936 up_write(&com->lc_sem);
3939 CDEBUG(D_LFSCK, "%s: namespace LFSCK checkpoint at the pos ["LPU64
3940 ", "DFID", "LPX64"]: rc = %d\n", lfsck_lfsck2name(lfsck),
3941 lfsck->li_pos_current.lp_oit_cookie,
3942 PFID(&lfsck->li_pos_current.lp_dir_parent),
3943 lfsck->li_pos_current.lp_dir_cookie, rc);
3945 return rc > 0 ? 0 : rc;
3948 static int lfsck_namespace_prep(const struct lu_env *env,
3949 struct lfsck_component *com,
3950 struct lfsck_start_param *lsp)
3952 struct lfsck_instance *lfsck = com->lc_lfsck;
3953 struct lfsck_namespace *ns = com->lc_file_ram;
3954 struct lfsck_position *pos = &com->lc_pos_start;
3957 rc = lfsck_namespace_load_bitmap(env, com);
3958 if (rc != 0 || ns->ln_status == LS_COMPLETED) {
3959 rc = lfsck_namespace_reset(env, com, false);
3961 rc = lfsck_set_param(env, lfsck, lsp->lsp_start, true);
3964 CDEBUG(D_LFSCK, "%s: namespace LFSCK prep failed: "
3965 "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
3971 down_write(&com->lc_sem);
3972 ns->ln_time_latest_start = cfs_time_current_sec();
3973 spin_lock(&lfsck->li_lock);
3975 if (ns->ln_flags & LF_SCANNED_ONCE) {
3976 if (!lfsck->li_drop_dryrun ||
3977 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent)) {
3978 ns->ln_status = LS_SCANNING_PHASE2;
3979 list_move_tail(&com->lc_link,
3980 &lfsck->li_list_double_scan);
3981 if (!list_empty(&com->lc_link_dir))
3982 list_del_init(&com->lc_link_dir);
3983 lfsck_pos_set_zero(pos);
3985 ns->ln_status = LS_SCANNING_PHASE1;
3986 ns->ln_run_time_phase1 = 0;
3987 ns->ln_run_time_phase2 = 0;
3988 ns->ln_items_checked = 0;
3989 ns->ln_items_repaired = 0;
3990 ns->ln_items_failed = 0;
3991 ns->ln_dirs_checked = 0;
3992 ns->ln_objs_checked_phase2 = 0;
3993 ns->ln_objs_repaired_phase2 = 0;
3994 ns->ln_objs_failed_phase2 = 0;
3995 ns->ln_objs_nlink_repaired = 0;
3996 ns->ln_dirent_repaired = 0;
3997 ns->ln_linkea_repaired = 0;
3998 ns->ln_mul_linked_checked = 0;
3999 ns->ln_mul_linked_repaired = 0;
4000 ns->ln_unknown_inconsistency = 0;
4001 ns->ln_unmatched_pairs_repaired = 0;
4002 ns->ln_dangling_repaired = 0;
4003 ns->ln_mul_ref_repaired = 0;
4004 ns->ln_bad_type_repaired = 0;
4005 ns->ln_lost_dirent_repaired = 0;
4006 ns->ln_striped_dirs_scanned = 0;
4007 ns->ln_striped_dirs_repaired = 0;
4008 ns->ln_striped_dirs_failed = 0;
4009 ns->ln_striped_dirs_disabled = 0;
4010 ns->ln_striped_dirs_skipped = 0;
4011 ns->ln_striped_shards_scanned = 0;
4012 ns->ln_striped_shards_repaired = 0;
4013 ns->ln_striped_shards_failed = 0;
4014 ns->ln_striped_shards_skipped = 0;
4015 ns->ln_name_hash_repaired = 0;
4016 fid_zero(&ns->ln_fid_latest_scanned_phase2);
4017 if (list_empty(&com->lc_link_dir))
4018 list_add_tail(&com->lc_link_dir,
4019 &lfsck->li_list_dir);
4020 *pos = ns->ln_pos_first_inconsistent;
4023 ns->ln_status = LS_SCANNING_PHASE1;
4024 if (list_empty(&com->lc_link_dir))
4025 list_add_tail(&com->lc_link_dir,
4026 &lfsck->li_list_dir);
4027 if (!lfsck->li_drop_dryrun ||
4028 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent)) {
4029 *pos = ns->ln_pos_last_checkpoint;
4030 pos->lp_oit_cookie++;
4032 *pos = ns->ln_pos_first_inconsistent;
4036 spin_unlock(&lfsck->li_lock);
4037 up_write(&com->lc_sem);
4039 rc = lfsck_start_assistant(env, com, lsp);
4041 CDEBUG(D_LFSCK, "%s: namespace LFSCK prep done, start pos ["LPU64", "
4042 DFID", "LPX64"]: rc = %d\n",
4043 lfsck_lfsck2name(lfsck), pos->lp_oit_cookie,
4044 PFID(&pos->lp_dir_parent), pos->lp_dir_cookie, rc);
4049 static int lfsck_namespace_exec_oit(const struct lu_env *env,
4050 struct lfsck_component *com,
4051 struct dt_object *obj)
4053 struct lfsck_thread_info *info = lfsck_env_info(env);
4054 struct lfsck_namespace *ns = com->lc_file_ram;
4055 struct lfsck_instance *lfsck = com->lc_lfsck;
4056 const struct lu_fid *fid = lfsck_dto2fid(obj);
4057 struct lu_fid *pfid = &info->lti_fid2;
4058 struct lu_name *cname = &info->lti_name;
4059 struct lu_seq_range *range = &info->lti_range;
4060 struct seq_server_site *ss = lfsck_dev_site(lfsck);
4061 struct linkea_data ldata = { NULL };
4062 __u32 idx = lfsck_dev_idx(lfsck);
4066 rc = lfsck_links_read(env, obj, &ldata);
4070 /* -EINVAL means crashed linkEA, should be verified. */
4071 if (rc == -EINVAL) {
4072 rc = lfsck_namespace_trace_update(env, com, fid,
4073 LNTF_CHECK_LINKEA, true);
4075 struct lustre_handle lh = { 0 };
4077 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
4078 MDS_INODELOCK_UPDATE |
4079 MDS_INODELOCK_XATTR, LCK_EX);
4081 rc = lfsck_namespace_links_remove(env, com,
4083 lfsck_ibits_unlock(&lh, LCK_EX);
4087 GOTO(out, rc = (rc == -ENOENT ? 0 : rc));
4090 if (rc == -ENODATA) {
4091 rc = lfsck_namespace_check_for_double_scan(env, com, obj);
4099 /* Record multiple-linked object. */
4100 if (ldata.ld_leh->leh_reccount > 1) {
4101 rc = lfsck_namespace_trace_update(env, com, fid,
4102 LNTF_CHECK_LINKEA, true);
4107 linkea_first_entry(&ldata);
4108 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, cname, pfid);
4109 if (!fid_is_sane(pfid)) {
4110 rc = lfsck_namespace_trace_update(env, com, fid,
4111 LNTF_CHECK_PARENT, true);
4113 fld_range_set_mdt(range);
4114 rc = fld_local_lookup(env, ss->ss_server_fld,
4115 fid_seq(pfid), range);
4116 if ((rc == -ENOENT) ||
4117 (rc == 0 && range->lsr_index != idx))
4118 rc = lfsck_namespace_trace_update(env, com, fid,
4119 LNTF_CHECK_LINKEA, true);
4121 rc = lfsck_namespace_check_for_double_scan(env, com,
4128 down_write(&com->lc_sem);
4129 if (S_ISDIR(lfsck_object_type(obj)))
4130 ns->ln_dirs_checked++;
4132 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
4133 up_write(&com->lc_sem);
4138 static int lfsck_namespace_exec_dir(const struct lu_env *env,
4139 struct lfsck_component *com,
4140 struct lu_dirent *ent, __u16 type)
4142 struct lfsck_assistant_data *lad = com->lc_data;
4143 struct lfsck_instance *lfsck = com->lc_lfsck;
4144 struct lfsck_namespace_req *lnr;
4145 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4146 struct ptlrpc_thread *mthread = &lfsck->li_thread;
4147 struct ptlrpc_thread *athread = &lad->lad_thread;
4148 struct l_wait_info lwi = { 0 };
4149 bool wakeup = false;
4151 l_wait_event(mthread->t_ctl_waitq,
4152 lad->lad_prefetched < bk->lb_async_windows ||
4153 !thread_is_running(mthread) ||
4154 thread_is_stopped(athread),
4157 if (unlikely(!thread_is_running(mthread)) ||
4158 thread_is_stopped(athread))
4161 if (unlikely(lfsck_is_dead_obj(lfsck->li_obj_dir)))
4164 lnr = lfsck_namespace_assistant_req_init(com->lc_lfsck, ent, type);
4166 struct lfsck_namespace *ns = com->lc_file_ram;
4168 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
4169 return PTR_ERR(lnr);
4172 spin_lock(&lad->lad_lock);
4173 if (lad->lad_assistant_status < 0) {
4174 spin_unlock(&lad->lad_lock);
4175 lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar);
4176 return lad->lad_assistant_status;
4179 list_add_tail(&lnr->lnr_lar.lar_list, &lad->lad_req_list);
4180 if (lad->lad_prefetched == 0)
4183 lad->lad_prefetched++;
4184 spin_unlock(&lad->lad_lock);
4186 wake_up_all(&lad->lad_thread.t_ctl_waitq);
4188 down_write(&com->lc_sem);
4189 com->lc_new_checked++;
4190 up_write(&com->lc_sem);
4195 static int lfsck_namespace_post(const struct lu_env *env,
4196 struct lfsck_component *com,
4197 int result, bool init)
4199 struct lfsck_instance *lfsck = com->lc_lfsck;
4200 struct lfsck_namespace *ns = com->lc_file_ram;
4204 lfsck_post_generic(env, com, &result);
4206 down_write(&com->lc_sem);
4207 lfsck_namespace_release_lmv(env, com);
4209 spin_lock(&lfsck->li_lock);
4211 ns->ln_pos_last_checkpoint = lfsck->li_pos_checkpoint;
4213 ns->ln_status = LS_SCANNING_PHASE2;
4214 ns->ln_flags |= LF_SCANNED_ONCE;
4215 ns->ln_flags &= ~LF_UPGRADE;
4216 list_del_init(&com->lc_link_dir);
4217 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
4218 } else if (result == 0) {
4219 if (lfsck->li_status != 0)
4220 ns->ln_status = lfsck->li_status;
4222 ns->ln_status = LS_STOPPED;
4223 if (ns->ln_status != LS_PAUSED) {
4224 list_del_init(&com->lc_link_dir);
4225 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4228 ns->ln_status = LS_FAILED;
4229 list_del_init(&com->lc_link_dir);
4230 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4232 spin_unlock(&lfsck->li_lock);
4235 ns->ln_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
4236 HALF_SEC - lfsck->li_time_last_checkpoint);
4237 ns->ln_time_last_checkpoint = cfs_time_current_sec();
4238 ns->ln_items_checked += com->lc_new_checked;
4239 com->lc_new_checked = 0;
4242 rc = lfsck_namespace_store(env, com, false);
4243 up_write(&com->lc_sem);
4245 CDEBUG(D_LFSCK, "%s: namespace LFSCK post done: rc = %d\n",
4246 lfsck_lfsck2name(lfsck), rc);
4252 lfsck_namespace_dump(const struct lu_env *env, struct lfsck_component *com,
4255 struct lfsck_instance *lfsck = com->lc_lfsck;
4256 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4257 struct lfsck_namespace *ns = com->lc_file_ram;
4260 down_read(&com->lc_sem);
4261 seq_printf(m, "name: lfsck_namespace\n"
4267 lfsck_status2names(ns->ln_status));
4269 rc = lfsck_bits_dump(m, ns->ln_flags, lfsck_flags_names, "flags");
4273 rc = lfsck_bits_dump(m, bk->lb_param, lfsck_param_names, "param");
4277 rc = lfsck_time_dump(m, ns->ln_time_last_complete,
4282 rc = lfsck_time_dump(m, ns->ln_time_latest_start,
4287 rc = lfsck_time_dump(m, ns->ln_time_last_checkpoint,
4292 rc = lfsck_pos_dump(m, &ns->ln_pos_latest_start,
4293 "latest_start_position");
4297 rc = lfsck_pos_dump(m, &ns->ln_pos_last_checkpoint,
4298 "last_checkpoint_position");
4302 rc = lfsck_pos_dump(m, &ns->ln_pos_first_inconsistent,
4303 "first_failure_position");
4307 if (ns->ln_status == LS_SCANNING_PHASE1) {
4308 struct lfsck_position pos;
4309 const struct dt_it_ops *iops;
4310 cfs_duration_t duration = cfs_time_current() -
4311 lfsck->li_time_last_checkpoint;
4312 __u64 checked = ns->ln_items_checked + com->lc_new_checked;
4313 __u64 speed = checked;
4314 __u64 new_checked = com->lc_new_checked *
4315 msecs_to_jiffies(MSEC_PER_SEC);
4316 __u32 rtime = ns->ln_run_time_phase1 +
4317 cfs_duration_sec(duration + HALF_SEC);
4320 do_div(new_checked, duration);
4323 do_div(speed, rtime);
4325 lfsck_namespace_dump_statistics(m, ns, checked, 0, rtime, 0);
4326 seq_printf(m, "average_speed_phase1: "LPU64" items/sec\n"
4327 "average_speed_phase2: N/A\n"
4328 "average_speed_total: "LPU64" items/sec\n"
4329 "real_time_speed_phase1: "LPU64" items/sec\n"
4330 "real_time_speed_phase2: N/A\n",
4335 LASSERT(lfsck->li_di_oit != NULL);
4337 iops = &lfsck->li_obj_oit->do_index_ops->dio_it;
4339 /* The low layer otable-based iteration position may NOT
4340 * exactly match the namespace-based directory traversal
4341 * cookie. Generally, it is not a serious issue. But the
4342 * caller should NOT make assumption on that. */
4343 pos.lp_oit_cookie = iops->store(env, lfsck->li_di_oit);
4344 if (!lfsck->li_current_oit_processed)
4345 pos.lp_oit_cookie--;
4347 spin_lock(&lfsck->li_lock);
4348 if (lfsck->li_di_dir != NULL) {
4349 pos.lp_dir_cookie = lfsck->li_cookie_dir;
4350 if (pos.lp_dir_cookie >= MDS_DIR_END_OFF) {
4351 fid_zero(&pos.lp_dir_parent);
4352 pos.lp_dir_cookie = 0;
4355 *lfsck_dto2fid(lfsck->li_obj_dir);
4358 fid_zero(&pos.lp_dir_parent);
4359 pos.lp_dir_cookie = 0;
4361 spin_unlock(&lfsck->li_lock);
4362 lfsck_pos_dump(m, &pos, "current_position");
4363 } else if (ns->ln_status == LS_SCANNING_PHASE2) {
4364 cfs_duration_t duration = cfs_time_current() -
4365 lfsck->li_time_last_checkpoint;
4366 __u64 checked = ns->ln_objs_checked_phase2 +
4367 com->lc_new_checked;
4368 __u64 speed1 = ns->ln_items_checked;
4369 __u64 speed2 = checked;
4370 __u64 speed0 = speed1 + speed2;
4371 __u64 new_checked = com->lc_new_checked *
4372 msecs_to_jiffies(MSEC_PER_SEC);
4373 __u32 rtime = ns->ln_run_time_phase2 +
4374 cfs_duration_sec(duration + HALF_SEC);
4375 __u32 time0 = ns->ln_run_time_phase1 + rtime;
4378 do_div(new_checked, duration);
4380 if (ns->ln_run_time_phase1 != 0)
4381 do_div(speed1, ns->ln_run_time_phase1);
4382 else if (ns->ln_items_checked != 0)
4386 do_div(speed2, rtime);
4387 else if (checked != 0)
4391 do_div(speed0, time0);
4393 lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked,
4395 ns->ln_run_time_phase1, rtime);
4396 seq_printf(m, "average_speed_phase1: "LPU64" items/sec\n"
4397 "average_speed_phase2: "LPU64" objs/sec\n"
4398 "average_speed_total: "LPU64" items/sec\n"
4399 "real_time_speed_phase1: N/A\n"
4400 "real_time_speed_phase2: "LPU64" objs/sec\n"
4401 "current_position: "DFID"\n",
4406 PFID(&ns->ln_fid_latest_scanned_phase2));
4408 __u64 speed1 = ns->ln_items_checked;
4409 __u64 speed2 = ns->ln_objs_checked_phase2;
4410 __u64 speed0 = speed1 + speed2;
4411 __u32 time0 = ns->ln_run_time_phase1 + ns->ln_run_time_phase2;
4413 if (ns->ln_run_time_phase1 != 0)
4414 do_div(speed1, ns->ln_run_time_phase1);
4415 else if (ns->ln_items_checked != 0)
4418 if (ns->ln_run_time_phase2 != 0)
4419 do_div(speed2, ns->ln_run_time_phase2);
4420 else if (ns->ln_objs_checked_phase2 != 0)
4424 do_div(speed0, time0);
4426 lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked,
4427 ns->ln_objs_checked_phase2,
4428 ns->ln_run_time_phase1,
4429 ns->ln_run_time_phase2);
4430 seq_printf(m, "average_speed_phase1: "LPU64" items/sec\n"
4431 "average_speed_phase2: "LPU64" objs/sec\n"
4432 "average_speed_total: "LPU64" items/sec\n"
4433 "real_time_speed_phase1: N/A\n"
4434 "real_time_speed_phase2: N/A\n"
4435 "current_position: N/A\n",
4441 up_read(&com->lc_sem);
4445 static int lfsck_namespace_double_scan(const struct lu_env *env,
4446 struct lfsck_component *com)
4448 struct lfsck_namespace *ns = com->lc_file_ram;
4449 struct lfsck_assistant_data *lad = com->lc_data;
4450 struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
4451 struct lfsck_tgt_desc *ltd;
4452 struct lfsck_tgt_desc *next;
4455 rc = lfsck_double_scan_generic(env, com, ns->ln_status);
4456 if (thread_is_stopped(&lad->lad_thread)) {
4457 LASSERT(list_empty(&lad->lad_req_list));
4458 LASSERT(list_empty(&lad->lad_mdt_phase1_list));
4460 spin_lock(<ds->ltd_lock);
4461 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4462 ltd_namespace_phase_list) {
4463 list_del_init(<d->ltd_namespace_phase_list);
4465 spin_unlock(<ds->ltd_lock);
4471 static void lfsck_namespace_data_release(const struct lu_env *env,
4472 struct lfsck_component *com)
4474 struct lfsck_assistant_data *lad = com->lc_data;
4475 struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
4476 struct lfsck_tgt_desc *ltd;
4477 struct lfsck_tgt_desc *next;
4479 LASSERT(lad != NULL);
4480 LASSERT(thread_is_init(&lad->lad_thread) ||
4481 thread_is_stopped(&lad->lad_thread));
4482 LASSERT(list_empty(&lad->lad_req_list));
4484 com->lc_data = NULL;
4485 lfsck_namespace_release_lmv(env, com);
4487 spin_lock(<ds->ltd_lock);
4488 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
4489 ltd_namespace_phase_list) {
4490 list_del_init(<d->ltd_namespace_phase_list);
4492 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4493 ltd_namespace_phase_list) {
4494 list_del_init(<d->ltd_namespace_phase_list);
4496 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_list,
4497 ltd_namespace_list) {
4498 list_del_init(<d->ltd_namespace_list);
4500 spin_unlock(<ds->ltd_lock);
4502 if (likely(lad->lad_bitmap != NULL))
4503 CFS_FREE_BITMAP(lad->lad_bitmap);
4508 static void lfsck_namespace_quit(const struct lu_env *env,
4509 struct lfsck_component *com)
4511 struct lfsck_assistant_data *lad = com->lc_data;
4512 struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
4513 struct lfsck_tgt_desc *ltd;
4514 struct lfsck_tgt_desc *next;
4516 LASSERT(lad != NULL);
4518 lfsck_quit_generic(env, com);
4520 LASSERT(thread_is_init(&lad->lad_thread) ||
4521 thread_is_stopped(&lad->lad_thread));
4522 LASSERT(list_empty(&lad->lad_req_list));
4524 lfsck_namespace_release_lmv(env, com);
4526 spin_lock(<ds->ltd_lock);
4527 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
4528 ltd_namespace_phase_list) {
4529 list_del_init(<d->ltd_namespace_phase_list);
4531 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4532 ltd_namespace_phase_list) {
4533 list_del_init(<d->ltd_namespace_phase_list);
4535 spin_unlock(<ds->ltd_lock);
4538 static int lfsck_namespace_in_notify(const struct lu_env *env,
4539 struct lfsck_component *com,
4540 struct lfsck_request *lr,
4543 struct lfsck_instance *lfsck = com->lc_lfsck;
4544 struct lfsck_namespace *ns = com->lc_file_ram;
4545 struct lfsck_assistant_data *lad = com->lc_data;
4546 struct lfsck_tgt_descs *ltds = &lfsck->li_mdt_descs;
4547 struct lfsck_tgt_desc *ltd;
4552 switch (lr->lr_event) {
4553 case LE_SKIP_NLINK_DECLARE: {
4554 struct dt_object *obj;
4555 struct lu_fid *key = &lfsck_env_info(env)->lti_fid3;
4559 LASSERT(th != NULL);
4561 idx = lfsck_sub_trace_file_fid2idx(&lr->lr_fid);
4562 obj = com->lc_sub_trace_objs[idx].lsto_obj;
4563 fid_cpu_to_be(key, &lr->lr_fid);
4564 mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
4565 rc = dt_declare_delete(env, obj,
4566 (const struct dt_key *)key, th);
4568 rc = dt_declare_insert(env, obj,
4569 (const struct dt_rec *)&flags,
4570 (const struct dt_key *)key, th);
4571 mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
4575 case LE_SKIP_NLINK: {
4576 struct dt_object *obj;
4577 struct lu_fid *key = &lfsck_env_info(env)->lti_fid3;
4583 LASSERT(th != NULL);
4585 idx = lfsck_sub_trace_file_fid2idx(&lr->lr_fid);
4586 obj = com->lc_sub_trace_objs[idx].lsto_obj;
4587 fid_cpu_to_be(key, &lr->lr_fid);
4588 mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
4589 rc = dt_lookup(env, obj, (struct dt_rec *)&flags,
4590 (const struct dt_key *)key, BYPASS_CAPA);
4592 if (flags & LNTF_SKIP_NLINK) {
4594 &com->lc_sub_trace_objs[idx].lsto_mutex);
4600 } else if (rc != -ENOENT) {
4604 flags |= LNTF_SKIP_NLINK;
4606 rc = dt_delete(env, obj, (const struct dt_key *)key,
4612 rc = dt_insert(env, obj, (const struct dt_rec *)&flags,
4613 (const struct dt_key *)key, th, BYPASS_CAPA, 1);
4618 mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
4619 CDEBUG(D_LFSCK, "%s: RPC service thread mark the "DFID
4620 " to be skipped for namespace double scan: rc = %d\n",
4621 lfsck_lfsck2name(com->lc_lfsck), PFID(&lr->lr_fid), rc);
4624 /* If we cannot record this object in the LFSCK tracing,
4625 * we have to mark the LFSC as LF_INCOMPLETE, then the
4626 * LFSCK will skip nlink attribute verification for
4628 ns->ln_flags |= LF_INCOMPLETE;
4632 case LE_SET_LMV_MASTER: {
4633 struct dt_object *obj;
4635 obj = lfsck_object_find_bottom(env, lfsck, &lr->lr_fid);
4637 RETURN(PTR_ERR(obj));
4639 rc = lfsck_namespace_notify_lmv_master_local(env, com, obj);
4640 lfsck_object_put(env, obj);
4642 RETURN(rc > 0 ? 0 : rc);
4644 case LE_SET_LMV_SLAVE: {
4645 if (!(lr->lr_flags & LEF_RECHECK_NAME_HASH))
4646 ns->ln_striped_shards_repaired++;
4648 rc = lfsck_namespace_trace_update(env, com, &lr->lr_fid,
4649 LNTF_RECHECK_NAME_HASH, true);
4651 RETURN(rc > 0 ? 0 : rc);
4653 case LE_PHASE1_DONE:
4654 case LE_PHASE2_DONE:
4661 CDEBUG(D_LFSCK, "%s: namespace LFSCK handles notify %u from MDT %x, "
4662 "status %d, flags %x\n", lfsck_lfsck2name(lfsck), lr->lr_event,
4663 lr->lr_index, lr->lr_status, lr->lr_flags2);
4665 spin_lock(<ds->ltd_lock);
4666 ltd = LTD_TGT(ltds, lr->lr_index);
4668 spin_unlock(<ds->ltd_lock);
4673 list_del_init(<d->ltd_namespace_phase_list);
4674 switch (lr->lr_event) {
4675 case LE_PHASE1_DONE:
4676 if (lr->lr_status <= 0) {
4677 ltd->ltd_namespace_done = 1;
4678 list_del_init(<d->ltd_namespace_list);
4679 CDEBUG(D_LFSCK, "%s: MDT %x failed/stopped at "
4680 "phase1 for namespace LFSCK: rc = %d.\n",
4681 lfsck_lfsck2name(lfsck),
4682 ltd->ltd_index, lr->lr_status);
4683 ns->ln_flags |= LF_INCOMPLETE;
4688 if (lr->lr_flags2 & LF_INCOMPLETE)
4689 ns->ln_flags |= LF_INCOMPLETE;
4691 if (list_empty(<d->ltd_namespace_list))
4692 list_add_tail(<d->ltd_namespace_list,
4693 &lad->lad_mdt_list);
4694 list_add_tail(<d->ltd_namespace_phase_list,
4695 &lad->lad_mdt_phase2_list);
4697 case LE_PHASE2_DONE:
4698 ltd->ltd_namespace_done = 1;
4699 list_del_init(<d->ltd_namespace_list);
4703 ltd->ltd_namespace_done = 1;
4704 list_del_init(<d->ltd_namespace_list);
4705 if (!(lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT)) {
4707 "%s: the peer MDT %x exit namespace LFSCK\n",
4708 lfsck_lfsck2name(lfsck), ltd->ltd_index);
4709 ns->ln_flags |= LF_INCOMPLETE;
4715 spin_unlock(<ds->ltd_lock);
4717 if (fail && lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) {
4718 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
4720 memset(stop, 0, sizeof(*stop));
4721 stop->ls_status = lr->lr_status;
4722 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
4723 lfsck_stop(env, lfsck->li_bottom, stop);
4724 } else if (lfsck_phase2_next_ready(lad)) {
4725 wake_up_all(&lad->lad_thread.t_ctl_waitq);
4731 static int lfsck_namespace_query(const struct lu_env *env,
4732 struct lfsck_component *com)
4734 struct lfsck_namespace *ns = com->lc_file_ram;
4736 return ns->ln_status;
4739 static struct lfsck_operations lfsck_namespace_ops = {
4740 .lfsck_reset = lfsck_namespace_reset,
4741 .lfsck_fail = lfsck_namespace_fail,
4742 .lfsck_close_dir = lfsck_namespace_close_dir,
4743 .lfsck_open_dir = lfsck_namespace_open_dir,
4744 .lfsck_checkpoint = lfsck_namespace_checkpoint,
4745 .lfsck_prep = lfsck_namespace_prep,
4746 .lfsck_exec_oit = lfsck_namespace_exec_oit,
4747 .lfsck_exec_dir = lfsck_namespace_exec_dir,
4748 .lfsck_post = lfsck_namespace_post,
4749 .lfsck_dump = lfsck_namespace_dump,
4750 .lfsck_double_scan = lfsck_namespace_double_scan,
4751 .lfsck_data_release = lfsck_namespace_data_release,
4752 .lfsck_quit = lfsck_namespace_quit,
4753 .lfsck_in_notify = lfsck_namespace_in_notify,
4754 .lfsck_query = lfsck_namespace_query,
4758 * Repair dangling name entry.
4760 * For the name entry with dangling reference, we need to repare the
4761 * inconsistency according to the LFSCK sponsor's requirement:
4763 * 1) Keep the inconsistency there and report the inconsistency case,
4764 * then give the chance to the application to find related issues,
4765 * and the users can make the decision about how to handle it with
4766 * more human knownledge. (by default)
4768 * 2) Re-create the missing MDT-object with the FID information.
4770 * \param[in] env pointer to the thread context
4771 * \param[in] com pointer to the lfsck component
4772 * \param[in] parent pointer to the dir object that contains the dangling
4774 * \param[in] child pointer to the object corresponding to the dangling
4776 * \param[in] lnr pointer to the namespace request that contains the
4777 * name's name, parent object, parent's LMV, and ect.
4779 * \retval positive number if no need to repair
4780 * \retval zero for repaired successfully
4781 * \retval negative error number on failure
4783 int lfsck_namespace_repair_dangling(const struct lu_env *env,
4784 struct lfsck_component *com,
4785 struct dt_object *parent,
4786 struct dt_object *child,
4787 struct lfsck_namespace_req *lnr)
4789 struct lfsck_thread_info *info = lfsck_env_info(env);
4790 struct lu_attr *la = &info->lti_la;
4791 struct dt_allocation_hint *hint = &info->lti_hint;
4792 struct dt_object_format *dof = &info->lti_dof;
4793 struct dt_insert_rec *rec = &info->lti_dt_rec;
4794 struct lmv_mds_md_v1 *lmv2 = &info->lti_lmv2;
4795 struct dt_object *pobj = NULL;
4796 struct dt_object *cobj = NULL;
4797 const struct lu_name *cname;
4798 struct linkea_data ldata = { NULL };
4799 struct lfsck_lock_handle *llh = &info->lti_llh;
4800 struct lu_buf linkea_buf;
4801 struct lu_buf lmv_buf;
4802 struct lfsck_instance *lfsck = com->lc_lfsck;
4803 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4804 struct dt_device *dev = lfsck->li_next;
4805 struct thandle *th = NULL;
4807 __u16 type = lnr->lnr_type;
4811 cname = lfsck_name_get_const(env, lnr->lnr_name, lnr->lnr_namelen);
4812 if (bk->lb_param & LPF_CREATE_MDTOBJ)
4817 if (!create || bk->lb_param & LPF_DRYRUN)
4820 /* We may need to create the sub-objects of the @child via LOD,
4821 * so make the modification based on lfsck->li_next. */
4823 pobj = lfsck_object_locate(dev, parent);
4825 GOTO(log, rc = PTR_ERR(pobj));
4827 if (unlikely(!dt_try_as_dir(env, pobj)))
4828 GOTO(log, rc = -ENOTDIR);
4830 cobj = lfsck_object_locate(dev, child);
4832 GOTO(log, rc = PTR_ERR(cobj));
4834 rc = linkea_data_new(&ldata, &info->lti_linkea_buf2);
4838 rc = linkea_add_buf(&ldata, cname, lfsck_dto2fid(pobj));
4842 rc = lfsck_lock(env, lfsck, parent, lnr->lnr_name, llh,
4843 MDS_INODELOCK_UPDATE, LCK_PR);
4847 rc = lfsck_namespace_check_exist(env, pobj, cobj, lnr->lnr_name);
4851 /* Set the ctime as zero, then others can know it is created for
4852 * repairing dangling name entry by LFSCK. And if the LFSCK made
4853 * wrong decision and the real MDT-object has been found later,
4854 * then the LFSCK has chance to fix the incosistency properly. */
4855 memset(la, 0, sizeof(*la));
4856 la->la_mode = (type & S_IFMT) | 0600;
4857 la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
4858 LA_ATIME | LA_MTIME | LA_CTIME;
4860 cobj->do_ops->do_ah_init(env, hint, pobj, cobj,
4861 la->la_mode & S_IFMT);
4863 memset(dof, 0, sizeof(*dof));
4864 dof->dof_type = dt_mode_to_dft(type);
4865 /* If the target is a regular file, then the LFSCK will only create
4866 * the MDT-object without stripes (dof->dof_reg.striped = 0). related
4867 * OST-objects will be created when write open. */
4869 th = dt_trans_create(env, dev);
4871 GOTO(log, rc = PTR_ERR(th));
4873 /* 1a. create child. */
4874 rc = dt_declare_create(env, cobj, la, hint, dof, th);
4878 if (S_ISDIR(type)) {
4879 if (unlikely(!dt_try_as_dir(env, cobj)))
4880 GOTO(stop, rc = -ENOTDIR);
4882 /* 2a. insert dot into child dir */
4883 rec->rec_type = S_IFDIR;
4884 rec->rec_fid = lfsck_dto2fid(cobj);
4885 rc = dt_declare_insert(env, cobj,
4886 (const struct dt_rec *)rec,
4887 (const struct dt_key *)dot, th);
4891 /* 3a. insert dotdot into child dir */
4892 rec->rec_fid = lfsck_dto2fid(pobj);
4893 rc = dt_declare_insert(env, cobj,
4894 (const struct dt_rec *)rec,
4895 (const struct dt_key *)dotdot, th);
4899 /* 4a. increase child nlink */
4900 rc = dt_declare_ref_add(env, cobj, th);
4904 /* 5a. generate slave LMV EA. */
4905 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
4908 idx = lfsck_shard_name_to_index(env,
4909 lnr->lnr_name, lnr->lnr_namelen,
4910 type, lfsck_dto2fid(cobj));
4911 if (unlikely(idx < 0))
4912 GOTO(stop, rc = idx);
4914 *lmv2 = lnr->lnr_lmv->ll_lmv;
4915 lmv2->lmv_magic = LMV_MAGIC_STRIPE;
4916 lmv2->lmv_master_mdt_index = idx;
4918 lfsck_lmv_header_cpu_to_le(lmv2, lmv2);
4919 lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
4920 rc = dt_declare_xattr_set(env, cobj, &lmv_buf,
4921 XATTR_NAME_LMV, 0, th);
4927 /* 6a. insert linkEA for child */
4928 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
4929 ldata.ld_leh->leh_len);
4930 rc = dt_declare_xattr_set(env, cobj, &linkea_buf,
4931 XATTR_NAME_LINK, 0, th);
4935 rc = dt_trans_start_local(env, dev, th);
4937 GOTO(stop, rc = (rc == -EEXIST ? 1 : rc));
4939 dt_write_lock(env, cobj, 0);
4940 /* 1b. create child */
4941 rc = dt_create(env, cobj, la, hint, dof, th);
4943 GOTO(unlock, rc = (rc == -EEXIST ? 1 : rc));
4945 if (S_ISDIR(type)) {
4946 /* 2b. insert dot into child dir */
4947 rec->rec_type = S_IFDIR;
4948 rec->rec_fid = lfsck_dto2fid(cobj);
4949 rc = dt_insert(env, cobj, (const struct dt_rec *)rec,
4950 (const struct dt_key *)dot, th, BYPASS_CAPA, 1);
4954 /* 3b. insert dotdot into child dir */
4955 rec->rec_fid = lfsck_dto2fid(pobj);
4956 rc = dt_insert(env, cobj, (const struct dt_rec *)rec,
4957 (const struct dt_key *)dotdot, th,
4962 /* 4b. increase child nlink */
4963 rc = dt_ref_add(env, cobj, th);
4967 /* 5b. generate slave LMV EA. */
4968 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
4969 rc = dt_xattr_set(env, cobj, &lmv_buf, XATTR_NAME_LMV,
4970 0, th, BYPASS_CAPA);
4976 /* 6b. insert linkEA for child. */
4977 rc = dt_xattr_set(env, cobj, &linkea_buf,
4978 XATTR_NAME_LINK, 0, th, BYPASS_CAPA);
4983 dt_write_unlock(env, cobj);
4986 dt_trans_stop(env, dev, th);
4990 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found dangling "
4991 "reference for: parent "DFID", child "DFID", type %u, "
4992 "name %s. %s: rc = %d\n", lfsck_lfsck2name(lfsck),
4993 PFID(&lnr->lnr_lar.lar_fid), PFID(lfsck_dto2fid(child)),
4994 type, cname->ln_name,
4995 create ? "Create the lost OST-object as required" :
4996 "Keep the MDT-object there by default", rc);
4999 struct lfsck_namespace *ns = com->lc_file_ram;
5001 ns->ln_flags |= LF_INCONSISTENT;
5007 static int lfsck_namespace_assistant_handler_p1(const struct lu_env *env,
5008 struct lfsck_component *com,
5009 struct lfsck_assistant_req *lar)
5011 struct lfsck_thread_info *info = lfsck_env_info(env);
5012 struct lu_attr *la = &info->lti_la;
5013 struct lfsck_instance *lfsck = com->lc_lfsck;
5014 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5015 struct lfsck_namespace *ns = com->lc_file_ram;
5016 struct linkea_data ldata = { NULL };
5017 const struct lu_name *cname;
5018 struct thandle *handle = NULL;
5019 struct lfsck_namespace_req *lnr =
5020 container_of0(lar, struct lfsck_namespace_req, lnr_lar);
5021 struct dt_object *dir = NULL;
5022 struct dt_object *obj = NULL;
5023 const struct lu_fid *pfid;
5024 struct dt_device *dev = NULL;
5025 struct lustre_handle lh = { 0 };
5026 bool repaired = false;
5027 bool dtlocked = false;
5031 bool bad_hash = false;
5035 enum lfsck_namespace_inconsistency_type type = LNIT_NONE;
5038 dir = lfsck_object_find_bottom(env, lfsck, &lar->lar_fid);
5040 RETURN(PTR_ERR(dir));
5042 if (unlikely(lfsck_is_dead_obj(dir)))
5043 GOTO(put_dir, rc = 0);
5045 if (unlikely(!dt_try_as_dir(env, dir)))
5046 GOTO(put_dir, rc = -ENOTDIR);
5048 pfid = lfsck_dto2fid(dir);
5050 if (lnr->lnr_attr & LUDA_UPGRADE) {
5051 ns->ln_flags |= LF_UPGRADE;
5052 ns->ln_dirent_repaired++;
5054 } else if (lnr->lnr_attr & LUDA_REPAIR) {
5055 ns->ln_flags |= LF_INCONSISTENT;
5056 ns->ln_dirent_repaired++;
5060 if (unlikely(fid_is_zero(&lnr->lnr_fid))) {
5061 if (strcmp(lnr->lnr_name, dotdot) != 0)
5064 rc = lfsck_namespace_trace_update(env, com, pfid,
5065 LNTF_CHECK_PARENT, true);
5070 if (unlikely(!fid_is_sane(&lnr->lnr_fid))) {
5071 CDEBUG(D_LFSCK, "%s: dir scan find invalid FID "DFID
5072 " for the name entry %.*s under "DFID"\n",
5073 lfsck_lfsck2name(lfsck), PFID(&lnr->lnr_fid),
5074 lnr->lnr_namelen, lnr->lnr_name, PFID(pfid));
5076 if (strcmp(lnr->lnr_name, dotdot) != 0)
5077 /* invalid FID means bad name entry, remove it. */
5078 type = LNIT_BAD_DIRENT;
5080 /* If the parent FID is invalid, we cannot remove
5081 * the ".." entry directly. */
5082 rc = lfsck_namespace_trace_update(env, com, pfid,
5083 LNTF_CHECK_PARENT, true);
5088 if (unlikely(lnr->lnr_dir_cookie == MDS_DIR_END_OFF)) {
5089 rc = lfsck_namespace_striped_dir_rescan(env, com, dir, lnr);
5094 if (fid_seq_is_dot(fid_seq(&lnr->lnr_fid)))
5097 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
5098 rc = lfsck_namespace_handle_striped_master(env, com, dir, lnr);
5103 idx = lfsck_find_mdt_idx_by_fid(env, lfsck, &lnr->lnr_fid);
5105 GOTO(out, rc = idx);
5107 if (idx == lfsck_dev_idx(lfsck)) {
5108 if (unlikely(strcmp(lnr->lnr_name, dotdot) == 0))
5111 dev = lfsck->li_bottom;
5113 struct lfsck_tgt_desc *ltd;
5115 /* Usually, some local filesystem consistency verification
5116 * tools can guarantee the local namespace tree consistenct.
5117 * So the LFSCK will only verify the remote directory. */
5118 if (unlikely(strcmp(lnr->lnr_name, dotdot) == 0)) {
5119 rc = lfsck_namespace_trace_update(env, com, pfid,
5120 LNTF_CHECK_PARENT, true);
5125 ltd = LTD_TGT(&lfsck->li_mdt_descs, idx);
5126 if (unlikely(ltd == NULL)) {
5127 CDEBUG(D_LFSCK, "%s: cannot talk with MDT %x which "
5128 "did not join the namespace LFSCK\n",
5129 lfsck_lfsck2name(lfsck), idx);
5130 lfsck_lad_set_bitmap(env, com, idx);
5132 GOTO(out, rc = -ENODEV);
5138 obj = lfsck_object_find_by_dev(env, dev, &lnr->lnr_fid);
5140 GOTO(out, rc = PTR_ERR(obj));
5142 cname = lfsck_name_get_const(env, lnr->lnr_name, lnr->lnr_namelen);
5143 if (dt_object_exists(obj) == 0) {
5146 rc = lfsck_namespace_check_exist(env, dir, obj, lnr->lnr_name);
5148 if (!lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5149 lnr->lnr_name, lnr->lnr_namelen)) {
5150 type = LNIT_BAD_DIRENT;
5155 type = LNIT_DANGLING;
5156 rc = lfsck_namespace_repair_dangling(env, com, dir,
5165 if (!(bk->lb_param & LPF_DRYRUN) && repaired) {
5168 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
5169 MDS_INODELOCK_UPDATE |
5170 MDS_INODELOCK_XATTR, LCK_EX);
5174 handle = dt_trans_create(env, dev);
5176 GOTO(out, rc = PTR_ERR(handle));
5178 rc = lfsck_declare_namespace_exec_dir(env, obj, handle);
5182 rc = dt_trans_start_local(env, dev, handle);
5186 dt_write_lock(env, obj, 0);
5190 rc = lfsck_namespace_check_exist(env, dir, obj, lnr->lnr_name);
5194 rc = lfsck_links_read(env, obj, &ldata);
5195 if (unlikely(rc == -ENOENT)) {
5196 if (handle != NULL) {
5197 dt_write_unlock(env, obj);
5200 dt_trans_stop(env, dev, handle);
5203 lfsck_ibits_unlock(&lh, LCK_EX);
5206 /* It may happen when the remote object has been removed,
5207 * but the local MDT is not aware of that. */
5209 } else if (rc == 0) {
5210 count = ldata.ld_leh->leh_reccount;
5211 rc = linkea_links_find(&ldata, cname, pfid);
5213 (count == 1 || !S_ISDIR(lfsck_object_type(obj)))) {
5214 if ((lfsck_object_type(obj) & S_IFMT) !=
5216 ns->ln_flags |= LF_INCONSISTENT;
5217 type = LNIT_BAD_TYPE;
5223 ns->ln_flags |= LF_INCONSISTENT;
5225 /* If the name entry hash does not match the slave striped
5226 * directory, and the name entry does not match also, then
5227 * it is quite possible that name entry is corrupted. */
5228 if (!lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5229 lnr->lnr_name, lnr->lnr_namelen)) {
5230 type = LNIT_BAD_DIRENT;
5235 /* If the file type stored in the name entry does not match
5236 * the file type claimed by the object, and the object does
5237 * not recognize the name entry, then it is quite possible
5238 * that the name entry is corrupted. */
5239 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type) {
5240 type = LNIT_BAD_DIRENT;
5245 /* For sub-dir object, we cannot make sure whether the sub-dir
5246 * back references the parent via ".." name entry correctly or
5247 * not in the LFSCK first-stage scanning. It may be that the
5248 * (remote) sub-dir ".." name entry has no parent FID after
5249 * file-level backup/restore and its linkEA may be wrong.
5250 * So under such case, we should replace the linkEA according
5251 * to current name entry. But this needs to be done during the
5252 * LFSCK second-stage scanning. The LFSCK will record the name
5253 * entry for further possible using. */
5257 } else if (unlikely(rc == -EINVAL)) {
5258 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type)
5259 type = LNIT_BAD_TYPE;
5262 ns->ln_flags |= LF_INCONSISTENT;
5263 /* The magic crashed, we are not sure whether there are more
5264 * corrupt data in the linkea, so remove all linkea entries. */
5268 } else if (rc == -ENODATA) {
5269 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type)
5270 type = LNIT_BAD_TYPE;
5273 ns->ln_flags |= LF_UPGRADE;
5278 if (bk->lb_param & LPF_DRYRUN) {
5279 ns->ln_linkea_repaired++;
5285 if (!lustre_handle_is_used(&lh))
5291 rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, handle,
5298 rc = linkea_data_new(&ldata,
5299 &lfsck_env_info(env)->lti_linkea_buf);
5304 rc = linkea_add_buf(&ldata, cname, pfid);
5308 rc = lfsck_links_write(env, obj, &ldata, handle);
5309 if (unlikely(rc == -ENOSPC) &&
5310 S_ISREG(lfsck_object_type(obj)) && !dt_object_remote(obj)) {
5311 if (handle != NULL) {
5312 LASSERT(dt_write_locked(env, obj));
5314 dt_write_unlock(env, obj);
5317 dt_trans_stop(env, dev, handle);
5320 lfsck_ibits_unlock(&lh, LCK_EX);
5323 rc = lfsck_namespace_trace_update(env, com,
5324 &lnr->lnr_fid, LNTF_SKIP_NLINK, true);
5326 /* If we cannot record this object in the
5327 * LFSCK tracing, we have to mark the LFSCK
5328 * as LF_INCOMPLETE, then the LFSCK will
5329 * skip nlink attribute verification for
5331 ns->ln_flags |= LF_INCOMPLETE;
5339 count = ldata.ld_leh->leh_reccount;
5340 if (!S_ISDIR(lfsck_object_type(obj)) ||
5341 !dt_object_remote(obj)) {
5342 ns->ln_linkea_repaired++;
5352 dt_write_unlock(env, obj);
5354 if (handle != NULL && !IS_ERR(handle))
5355 dt_trans_stop(env, dev, handle);
5358 lfsck_ibits_unlock(&lh, LCK_EX);
5360 if (!name_is_dot_or_dotdot(lnr->lnr_name, lnr->lnr_namelen) &&
5361 !lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5362 lnr->lnr_name, lnr->lnr_namelen) &&
5363 type != LNIT_BAD_DIRENT) {
5364 ns->ln_flags |= LF_INCONSISTENT;
5367 rc = lfsck_namespace_repair_bad_name_hash(env, com, dir,
5368 lnr->lnr_lmv, lnr->lnr_name);
5377 rc = lfsck_namespace_repair_dirent(env, com, dir,
5378 obj, lnr->lnr_name, lnr->lnr_name,
5379 lnr->lnr_type, true, false);
5383 case LNIT_BAD_DIRENT:
5385 /* XXX: This is a bad dirent, we do not know whether
5386 * the original name entry reference a regular
5387 * file or a directory, then keep the parent's
5388 * nlink count unchanged here. */
5389 rc = lfsck_namespace_repair_dirent(env, com, dir,
5390 obj, lnr->lnr_name, lnr->lnr_name,
5391 lnr->lnr_type, false, false);
5399 if (count == 1 && S_ISREG(lfsck_object_type(obj)))
5400 dt_attr_get(env, obj, la, BYPASS_CAPA);
5403 down_write(&com->lc_sem);
5405 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail to handle "
5406 "the entry: "DFID", parent "DFID", name %.*s: rc = %d\n",
5407 lfsck_lfsck2name(lfsck), PFID(&lnr->lnr_fid),
5408 PFID(lfsck_dto2fid(dir)),
5409 lnr->lnr_namelen, lnr->lnr_name, rc);
5411 lfsck_namespace_record_failure(env, lfsck, ns);
5412 if ((rc == -ENOTCONN || rc == -ESHUTDOWN || rc == -EREMCHG ||
5413 rc == -ETIMEDOUT || rc == -EHOSTDOWN ||
5414 rc == -EHOSTUNREACH || rc == -EINPROGRESS) &&
5415 dev != NULL && dev != lfsck->li_bottom)
5416 lfsck_lad_set_bitmap(env, com, idx);
5418 if (!(bk->lb_param & LPF_FAILOUT))
5422 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant "
5423 "repaired the entry: "DFID", parent "DFID
5424 ", name %.*s\n", lfsck_lfsck2name(lfsck),
5425 PFID(&lnr->lnr_fid),
5426 PFID(lfsck_dto2fid(dir)),
5427 lnr->lnr_namelen, lnr->lnr_name);
5430 ns->ln_items_repaired++;
5434 ns->ln_dangling_repaired++;
5437 ns->ln_bad_type_repaired++;
5439 case LNIT_BAD_DIRENT:
5440 ns->ln_dirent_repaired++;
5446 if (bk->lb_param & LPF_DRYRUN &&
5447 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent))
5448 lfsck_pos_fill(env, lfsck,
5449 &ns->ln_pos_first_inconsistent,
5454 ns->ln_name_hash_repaired++;
5456 /* Not count repeatedly. */
5458 ns->ln_items_repaired++;
5460 if (bk->lb_param & LPF_DRYRUN &&
5461 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent))
5462 lfsck_pos_fill(env, lfsck,
5463 &ns->ln_pos_first_inconsistent,
5470 if (count > 1 || la->la_nlink > 1)
5471 ns->ln_mul_linked_checked++;
5473 up_write(&com->lc_sem);
5475 if (obj != NULL && !IS_ERR(obj))
5476 lfsck_object_put(env, obj);
5479 lfsck_object_put(env, dir);
5485 * Handle one orphan under the backend /lost+found directory
5487 * Insert the orphan FID into the namespace LFSCK trace file for further
5488 * processing (via the subsequent namespace LFSCK second-stage scanning).
5489 * At the same time, remove the orphan name entry from backend /lost+found
5490 * directory. There is an interval between the orphan name entry removed
5491 * from the backend /lost+found directory and the orphan FID in the LFSCK
5492 * trace file handled. In such interval, the LFSCK can be reset, then
5493 * all the FIDs recorded in the namespace LFSCK trace file will be dropped.
5494 * To guarantee that the orphans can be found when LFSCK run next time
5495 * without e2fsck again, when remove the orphan name entry, the LFSCK
5496 * will set the orphan's ctime attribute as 1. Since normal applications
5497 * cannot change the object's ctime attribute as 1. Then when LFSCK run
5498 * next time, it can record the object (that ctime is 1) in the namespace
5499 * LFSCK trace file during the first-stage scanning.
5501 * \param[in] env pointer to the thread context
5502 * \param[in] com pointer to the lfsck component
5503 * \param[in] parent pointer to the object for the backend /lost+found
5504 * \param[in] ent pointer to the name entry for the target under the
5505 * backend /lost+found
5507 * \retval positive for repaired
5508 * \retval 0 if needs to repair nothing
5509 * \retval negative error number on failure
5511 static int lfsck_namespace_scan_local_lpf_one(const struct lu_env *env,
5512 struct lfsck_component *com,
5513 struct dt_object *parent,
5514 struct lu_dirent *ent)
5516 struct lfsck_thread_info *info = lfsck_env_info(env);
5517 struct lu_fid *key = &info->lti_fid;
5518 struct lu_attr *la = &info->lti_la;
5519 struct lfsck_instance *lfsck = com->lc_lfsck;
5520 struct dt_object *obj;
5521 struct dt_device *dev = lfsck->li_bottom;
5522 struct dt_object *child = NULL;
5523 struct thandle *th = NULL;
5530 child = lfsck_object_find_by_dev(env, dev, &ent->lde_fid);
5532 RETURN(PTR_ERR(child));
5534 LASSERT(dt_object_exists(child));
5535 LASSERT(!dt_object_remote(child));
5537 idx = lfsck_sub_trace_file_fid2idx(&ent->lde_fid);
5538 obj = com->lc_sub_trace_objs[idx].lsto_obj;
5539 fid_cpu_to_be(key, &ent->lde_fid);
5540 rc = dt_lookup(env, obj, (struct dt_rec *)&flags,
5541 (const struct dt_key *)key, BYPASS_CAPA);
5544 flags |= LNTF_CHECK_ORPHAN;
5545 } else if (rc == -ENOENT) {
5546 flags = LNTF_CHECK_ORPHAN;
5551 th = dt_trans_create(env, dev);
5553 GOTO(out, rc = PTR_ERR(th));
5555 /* a1. remove name entry from backend /lost+found */
5556 rc = dt_declare_delete(env, parent,
5557 (const struct dt_key *)ent->lde_name, th);
5561 if (S_ISDIR(lfsck_object_type(child))) {
5562 /* a2. decrease parent's nlink */
5563 rc = dt_declare_ref_del(env, parent, th);
5569 /* a3. remove child's FID from the LFSCK trace file. */
5570 rc = dt_declare_delete(env, obj,
5571 (const struct dt_key *)key, th);
5575 /* a4. set child's ctime as 1 */
5576 memset(la, 0, sizeof(*la));
5578 la->la_valid = LA_CTIME;
5579 rc = dt_declare_attr_set(env, child, la, th);
5584 /* a5. insert child's FID into the LFSCK trace file. */
5585 rc = dt_declare_insert(env, obj, (const struct dt_rec *)&flags,
5586 (const struct dt_key *)key, th);
5590 rc = dt_trans_start_local(env, dev, th);
5594 /* b1. remove name entry from backend /lost+found */
5595 rc = dt_delete(env, parent, (const struct dt_key *)ent->lde_name, th,
5600 if (S_ISDIR(lfsck_object_type(child))) {
5601 /* b2. decrease parent's nlink */
5602 dt_write_lock(env, parent, 0);
5603 rc = dt_ref_del(env, parent, th);
5604 dt_write_unlock(env, parent);
5610 /* a3. remove child's FID from the LFSCK trace file. */
5611 rc = dt_delete(env, obj, (const struct dt_key *)key, th,
5616 /* b4. set child's ctime as 1 */
5617 rc = dt_attr_set(env, child, la, th, BYPASS_CAPA);
5622 /* b5. insert child's FID into the LFSCK trace file. */
5623 rc = dt_insert(env, obj, (const struct dt_rec *)&flags,
5624 (const struct dt_key *)key, th, BYPASS_CAPA, 1);
5626 GOTO(stop, rc = (rc == 0 ? 1 : rc));
5629 dt_trans_stop(env, dev, th);
5632 lfsck_object_put(env, child);
5638 * Handle orphans under the backend /lost+found directory
5640 * Some backend checker, such as e2fsck for ldiskfs may find some orphans
5641 * and put them under the backend /lost+found directory that is invisible
5642 * to client. The LFSCK will scan such directory, for the original client
5643 * visible orphans, add their fids into the namespace LFSCK trace file,
5644 * then the subsenquent namespace LFSCK second-stage scanning can handle
5645 * them as other objects to be double scanned: either move back to normal
5646 * namespace, or to the global visible orphan directory:
5647 * /ROOT/.lustre/lost+found/MDTxxxx/
5649 * \param[in] env pointer to the thread context
5650 * \param[in] com pointer to the lfsck component
5652 static void lfsck_namespace_scan_local_lpf(const struct lu_env *env,
5653 struct lfsck_component *com)
5655 struct lfsck_thread_info *info = lfsck_env_info(env);
5656 struct lu_dirent *ent =
5657 (struct lu_dirent *)info->lti_key;
5658 struct lu_seq_range *range = &info->lti_range;
5659 struct lfsck_instance *lfsck = com->lc_lfsck;
5660 struct ptlrpc_thread *thread = &lfsck->li_thread;
5661 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5662 struct lfsck_namespace *ns = com->lc_file_ram;
5663 struct dt_object *parent;
5664 const struct dt_it_ops *iops;
5666 struct seq_server_site *ss = lfsck_dev_site(lfsck);
5668 __u32 idx = lfsck_dev_idx(lfsck);
5673 parent = lfsck_object_find_by_dev(env, lfsck->li_bottom,
5674 &LU_BACKEND_LPF_FID);
5675 if (IS_ERR(parent)) {
5676 CERROR("%s: fail to find backend /lost+found: rc = %ld\n",
5677 lfsck_lfsck2name(lfsck), PTR_ERR(parent));
5681 /* It is normal that the /lost+found does not exist for ZFS backend. */
5682 if (!dt_object_exists(parent))
5685 if (unlikely(!dt_try_as_dir(env, parent)))
5686 GOTO(out, rc = -ENOTDIR);
5688 CDEBUG(D_LFSCK, "%s: start to scan backend /lost+found\n",
5689 lfsck_lfsck2name(lfsck));
5691 com->lc_new_scanned = 0;
5692 iops = &parent->do_index_ops->dio_it;
5693 di = iops->init(env, parent, LUDA_64BITHASH | LUDA_TYPE, BYPASS_CAPA);
5695 GOTO(out, rc = PTR_ERR(di));
5697 rc = iops->load(env, di, 0);
5699 rc = iops->next(env, di);
5704 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
5705 unlikely(!thread_is_running(thread)))
5708 rc = iops->rec(env, di, (struct dt_rec *)ent,
5709 LUDA_64BITHASH | LUDA_TYPE);
5711 rc = lfsck_unpack_ent(ent, &cookie, &type);
5713 if (unlikely(rc != 0)) {
5714 CDEBUG(D_LFSCK, "%s: fail to iterate backend "
5715 "/lost+found: rc = %d\n",
5716 lfsck_lfsck2name(lfsck), rc);
5721 /* skip dot and dotdot entries */
5722 if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen))
5725 if (!fid_seq_in_fldb(fid_seq(&ent->lde_fid)))
5728 if (fid_is_norm(&ent->lde_fid)) {
5729 fld_range_set_mdt(range);
5730 rc = fld_local_lookup(env, ss->ss_server_fld,
5731 fid_seq(&ent->lde_fid), range);
5734 } else if (idx != 0) {
5735 /* If the returned FID is IGIF, then there are three
5738 * 1) The object is upgraded from old Lustre-1.8 with
5739 * IGIF assigned to such object.
5740 * 2) The object is a backend local object and is
5741 * invisible to client.
5742 * 3) The object lost its LMV EA, and since there is
5743 * no FID-in-dirent for the orphan in the backend
5744 * /lost+found directory, then the low layer will
5745 * return IGIF for such object.
5747 * For MDTx (x != 0), it is either case 2) or case 3),
5748 * but from the LFSCK view, they are indistinguishable.
5749 * To be safe, the LFSCK will keep it there and report
5750 * some message, then the adminstrator can handle that
5753 * For MDT0, it is more possible the case 1). The LFSCK
5754 * will handle the orphan as an upgraded object. */
5755 CDEBUG(D_LFSCK, "%s: the orphan %.*s with IGIF "DFID
5756 "in the backend /lost+found on the MDT %04x, "
5757 "to be safe, skip it.\n",
5758 lfsck_lfsck2name(lfsck), ent->lde_namelen,
5759 ent->lde_name, PFID(&ent->lde_fid), idx);
5763 rc = lfsck_namespace_scan_local_lpf_one(env, com, parent, ent);
5766 down_write(&com->lc_sem);
5767 com->lc_new_scanned++;
5768 ns->ln_local_lpf_scanned++;
5770 ns->ln_local_lpf_moved++;
5772 ns->ln_local_lpf_skipped++;
5774 ns->ln_local_lpf_failed++;
5775 up_write(&com->lc_sem);
5777 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
5781 lfsck_control_speed_by_self(com);
5782 if (unlikely(!thread_is_running(thread))) {
5787 rc = iops->next(env, di);
5791 iops->fini(env, di);
5796 CDEBUG(D_LFSCK, "%s: stop to scan backend /lost+found: rc = %d\n",
5797 lfsck_lfsck2name(lfsck), rc);
5799 lfsck_object_put(env, parent);
5803 * Rescan the striped directory after the master LMV EA reset.
5805 * Sometimes, the master LMV EA of the striped directory maybe lost, so when
5806 * the namespace LFSCK engine scan the striped directory for the first time,
5807 * it will be regarded as a normal directory. As the LFSCK processing, some
5808 * other LFSCK instance on other MDT will find the shard of this striped dir,
5809 * and find that the master MDT-object of the striped directory lost its LMV
5810 * EA, then such remote LFSCK instance will regenerate the master LMV EA and
5811 * notify the LFSCK instance on this MDT to rescan the striped directory.
5813 * \param[in] env pointer to the thread context
5814 * \param[in] com pointer to the lfsck component
5815 * \param[in] llu the lfsck_lmv_unit that contains the striped directory
5818 * \retval positive number for success
5819 * \retval 0 for LFSCK stopped/paused
5820 * \retval negative error number on failure
5822 static int lfsck_namespace_rescan_striped_dir(const struct lu_env *env,
5823 struct lfsck_component *com,
5824 struct lfsck_lmv_unit *llu)
5826 struct lfsck_thread_info *info = lfsck_env_info(env);
5827 struct lfsck_instance *lfsck = com->lc_lfsck;
5828 struct lfsck_assistant_data *lad = com->lc_data;
5829 struct dt_object *dir;
5830 const struct dt_it_ops *iops;
5832 struct lu_dirent *ent =
5833 (struct lu_dirent *)info->lti_key;
5834 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5835 struct ptlrpc_thread *thread = &lfsck->li_thread;
5836 struct lfsck_namespace_req *lnr;
5837 struct lfsck_assistant_req *lar;
5842 LASSERT(list_empty(&lad->lad_req_list));
5844 lfsck->li_lmv = &llu->llu_lmv;
5845 lfsck->li_obj_dir = lfsck_object_get(llu->llu_obj);
5846 rc = lfsck_open_dir(env, lfsck, 0);
5850 dir = lfsck->li_obj_dir;
5851 di = lfsck->li_di_dir;
5852 iops = &dir->do_index_ops->dio_it;
5854 rc = iops->rec(env, di, (struct dt_rec *)ent,
5855 lfsck->li_args_dir);
5857 rc = lfsck_unpack_ent(ent, &lfsck->li_cookie_dir,
5861 if (bk->lb_param & LPF_FAILOUT)
5867 if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen))
5870 lnr = lfsck_namespace_assistant_req_init(lfsck, ent, type);
5872 if (bk->lb_param & LPF_FAILOUT)
5873 GOTO(out, rc = PTR_ERR(lnr));
5878 lar = &lnr->lnr_lar;
5879 rc = lfsck_namespace_assistant_handler_p1(env, com, lar);
5880 lfsck_namespace_assistant_req_fini(env, lar);
5881 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
5884 if (unlikely(!thread_is_running(thread)))
5888 rc = iops->next(env, di);
5892 lfsck_close_dir(env, lfsck, rc);
5896 /* The close_dir() may insert a dummy lnr in the lad->lad_req_list. */
5897 if (list_empty(&lad->lad_req_list))
5900 spin_lock(&lad->lad_lock);
5901 lar = list_entry(lad->lad_req_list.next, struct lfsck_assistant_req,
5903 list_del_init(&lar->lar_list);
5904 spin_unlock(&lad->lad_lock);
5906 rc = lfsck_namespace_assistant_handler_p1(env, com, lar);
5907 lfsck_namespace_assistant_req_fini(env, lar);
5909 RETURN(rc == 0 ? 1 : rc);
5913 lfsck_namespace_double_scan_one_trace_file(const struct lu_env *env,
5914 struct lfsck_component *com,
5915 struct dt_object *obj, bool first)
5917 struct lfsck_instance *lfsck = com->lc_lfsck;
5918 struct ptlrpc_thread *thread = &lfsck->li_thread;
5919 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5920 struct lfsck_namespace *ns = com->lc_file_ram;
5921 const struct dt_it_ops *iops = &obj->do_index_ops->dio_it;
5922 struct dt_object *target;
5930 di = iops->init(env, obj, 0, BYPASS_CAPA);
5932 RETURN(PTR_ERR(di));
5935 fid_cpu_to_be(&fid, &ns->ln_fid_latest_scanned_phase2);
5938 rc = iops->get(env, di, (const struct dt_key *)&fid);
5943 /* The start one either has been processed or does not exist,
5945 rc = iops->next(env, di);
5951 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
5952 unlikely(!thread_is_running(thread)))
5955 key = iops->key(env, di);
5964 fid_be_to_cpu(&fid, (const struct lu_fid *)key);
5965 if (!fid_is_sane(&fid)) {
5970 target = lfsck_object_find_bottom(env, lfsck, &fid);
5971 if (IS_ERR(target)) {
5972 rc = PTR_ERR(target);
5976 if (dt_object_exists(target)) {
5977 rc = iops->rec(env, di, (struct dt_rec *)&flags, 0);
5979 rc = lfsck_namespace_double_scan_one(env, com,
5986 lfsck_object_put(env, target);
5989 down_write(&com->lc_sem);
5990 com->lc_new_checked++;
5991 com->lc_new_scanned++;
5992 if (rc >= 0 && fid_is_sane(&fid))
5993 ns->ln_fid_latest_scanned_phase2 = fid;
5995 ns->ln_objs_repaired_phase2++;
5997 ns->ln_objs_failed_phase2++;
5998 up_write(&com->lc_sem);
6000 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
6003 if (unlikely(cfs_time_beforeq(com->lc_time_next_checkpoint,
6004 cfs_time_current())) &&
6005 com->lc_new_checked != 0) {
6006 down_write(&com->lc_sem);
6007 ns->ln_run_time_phase2 +=
6008 cfs_duration_sec(cfs_time_current() +
6009 HALF_SEC - com->lc_time_last_checkpoint);
6010 ns->ln_time_last_checkpoint = cfs_time_current_sec();
6011 ns->ln_objs_checked_phase2 += com->lc_new_checked;
6012 com->lc_new_checked = 0;
6013 rc = lfsck_namespace_store(env, com, false);
6014 up_write(&com->lc_sem);
6018 com->lc_time_last_checkpoint = cfs_time_current();
6019 com->lc_time_next_checkpoint =
6020 com->lc_time_last_checkpoint +
6021 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
6024 lfsck_control_speed_by_self(com);
6025 if (unlikely(!thread_is_running(thread)))
6028 rc = iops->next(env, di);
6037 iops->fini(env, di);
6042 static int lfsck_namespace_assistant_handler_p2(const struct lu_env *env,
6043 struct lfsck_component *com)
6045 struct lfsck_instance *lfsck = com->lc_lfsck;
6046 struct lfsck_namespace *ns = com->lc_file_ram;
6051 while (!list_empty(&lfsck->li_list_lmv)) {
6052 struct lfsck_lmv_unit *llu;
6054 spin_lock(&lfsck->li_lock);
6055 llu = list_entry(lfsck->li_list_lmv.next,
6056 struct lfsck_lmv_unit, llu_link);
6057 list_del_init(&llu->llu_link);
6058 spin_unlock(&lfsck->li_lock);
6060 rc = lfsck_namespace_rescan_striped_dir(env, com, llu);
6065 CDEBUG(D_LFSCK, "%s: namespace LFSCK phase2 scan start\n",
6066 lfsck_lfsck2name(lfsck));
6068 lfsck_namespace_scan_local_lpf(env, com);
6070 com->lc_new_checked = 0;
6071 com->lc_new_scanned = 0;
6072 com->lc_time_last_checkpoint = cfs_time_current();
6073 com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
6074 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
6076 i = lfsck_sub_trace_file_fid2idx(&ns->ln_fid_latest_scanned_phase2);
6077 rc = lfsck_namespace_double_scan_one_trace_file(env, com,
6078 com->lc_sub_trace_objs[i].lsto_obj, true);
6079 while (rc > 0 && ++i < LFSCK_STF_COUNT)
6080 rc = lfsck_namespace_double_scan_one_trace_file(env, com,
6081 com->lc_sub_trace_objs[i].lsto_obj, false);
6083 CDEBUG(D_LFSCK, "%s: namespace LFSCK phase2 scan stop at the No. %d "
6084 "trace file: rc = %d\n", lfsck_lfsck2name(lfsck), i, rc);
6089 static void lfsck_namespace_assistant_fill_pos(const struct lu_env *env,
6090 struct lfsck_component *com,
6091 struct lfsck_position *pos)
6093 struct lfsck_assistant_data *lad = com->lc_data;
6094 struct lfsck_namespace_req *lnr;
6096 if (list_empty(&lad->lad_req_list))
6099 lnr = list_entry(lad->lad_req_list.next,
6100 struct lfsck_namespace_req,
6102 pos->lp_oit_cookie = lnr->lnr_oit_cookie;
6103 pos->lp_dir_cookie = lnr->lnr_dir_cookie - 1;
6104 pos->lp_dir_parent = lnr->lnr_lar.lar_fid;
6107 static int lfsck_namespace_double_scan_result(const struct lu_env *env,
6108 struct lfsck_component *com,
6111 struct lfsck_instance *lfsck = com->lc_lfsck;
6112 struct lfsck_namespace *ns = com->lc_file_ram;
6114 down_write(&com->lc_sem);
6115 ns->ln_run_time_phase2 += cfs_duration_sec(cfs_time_current() +
6116 HALF_SEC - lfsck->li_time_last_checkpoint);
6117 ns->ln_time_last_checkpoint = cfs_time_current_sec();
6118 ns->ln_objs_checked_phase2 += com->lc_new_checked;
6119 com->lc_new_checked = 0;
6122 if (ns->ln_flags & LF_INCOMPLETE)
6123 ns->ln_status = LS_PARTIAL;
6125 ns->ln_status = LS_COMPLETED;
6126 if (!(lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN))
6127 ns->ln_flags &= ~(LF_SCANNED_ONCE | LF_INCONSISTENT);
6128 ns->ln_time_last_complete = ns->ln_time_last_checkpoint;
6129 ns->ln_success_count++;
6130 } else if (rc == 0) {
6131 if (lfsck->li_status != 0)
6132 ns->ln_status = lfsck->li_status;
6134 ns->ln_status = LS_STOPPED;
6136 ns->ln_status = LS_FAILED;
6139 rc = lfsck_namespace_store(env, com, false);
6140 up_write(&com->lc_sem);
6146 lfsck_namespace_assistant_sync_failures_interpret(const struct lu_env *env,
6147 struct ptlrpc_request *req,
6151 struct lfsck_async_interpret_args *laia = args;
6152 struct lfsck_tgt_desc *ltd = laia->laia_ltd;
6154 ltd->ltd_synced_failures = 1;
6161 * Notify remote LFSCK instances about former failures.
6163 * The local LFSCK instance has recorded which MDTs have ever failed to respond
6164 * some LFSCK verification requests (maybe because of network issues or the MDT
6165 * itself trouble). During the respond gap the MDT may missed some name entries
6166 * verification, then the MDT cannot know whether related MDT-objects have been
6167 * referenced by related name entries or not, then in the second-stage scanning,
6168 * these MDT-objects will be regarded as orphan, if the MDT-object contains bad
6169 * linkEA for back reference, then it will misguide the LFSCK to generate wrong
6170 * name entry for repairing the orphan.
6172 * To avoid above trouble, when layout LFSCK finishes the first-stage scanning,
6173 * it will scan the bitmap for the ever failed MDTs, and notify them that they
6174 * have ever missed some name entries verification and should skip the handling
6175 * for orphan MDT-objects.
6177 * \param[in] env pointer to the thread context
6178 * \param[in] com pointer to the lfsck component
6179 * \param[in] lr pointer to the lfsck request
6181 static void lfsck_namespace_assistant_sync_failures(const struct lu_env *env,
6182 struct lfsck_component *com,
6183 struct lfsck_request *lr)
6185 struct lfsck_async_interpret_args *laia =
6186 &lfsck_env_info(env)->lti_laia2;
6187 struct lfsck_assistant_data *lad = com->lc_data;
6188 struct lfsck_namespace *ns = com->lc_file_ram;
6189 struct lfsck_instance *lfsck = com->lc_lfsck;
6190 struct lfsck_tgt_descs *ltds = &lfsck->li_mdt_descs;
6191 struct lfsck_tgt_desc *ltd;
6192 struct ptlrpc_request_set *set;
6197 if (!lad->lad_incomplete)
6200 set = ptlrpc_prep_set();
6202 GOTO(out, rc = -ENOMEM);
6204 lr->lr_flags2 = ns->ln_flags | LF_INCOMPLETE;
6205 memset(laia, 0, sizeof(*laia));
6206 lad->lad_touch_gen++;
6208 down_read(<ds->ltd_rw_sem);
6209 cfs_foreach_bit(lad->lad_bitmap, idx) {
6210 ltd = LTD_TGT(ltds, idx);
6211 LASSERT(ltd != NULL);
6213 laia->laia_ltd = ltd;
6214 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
6215 lfsck_namespace_assistant_sync_failures_interpret,
6216 laia, LFSCK_NOTIFY);
6218 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail "
6219 "to sync failure with MDT %x: rc = %d\n",
6220 lfsck_lfsck2name(lfsck), ltd->ltd_index, rc);
6222 up_read(<ds->ltd_rw_sem);
6224 rc = ptlrpc_set_wait(set);
6225 ptlrpc_set_destroy(set);
6231 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail "
6232 "to sync failure with MDTs, and related MDTs "
6233 "may handle orphan improperly: rc = %d\n",
6234 lfsck_lfsck2name(lfsck), rc);
6239 struct lfsck_assistant_operations lfsck_namespace_assistant_ops = {
6240 .la_handler_p1 = lfsck_namespace_assistant_handler_p1,
6241 .la_handler_p2 = lfsck_namespace_assistant_handler_p2,
6242 .la_fill_pos = lfsck_namespace_assistant_fill_pos,
6243 .la_double_scan_result = lfsck_namespace_double_scan_result,
6244 .la_req_fini = lfsck_namespace_assistant_req_fini,
6245 .la_sync_failures = lfsck_namespace_assistant_sync_failures,
6249 * Verify the specified linkEA entry for the given directory object.
6250 * If the object has no such linkEA entry or it has more other linkEA
6251 * entries, then re-generate the linkEA with the given information.
6253 * \param[in] env pointer to the thread context
6254 * \param[in] obj pointer to the dt_object to be handled
6255 * \param[in] cname the name for the child in the parent directory
6256 * \param[in] pfid the parent directory's FID for the linkEA
6258 * \retval 0 for success
6259 * \retval negative error number on failure
6261 int lfsck_verify_linkea(const struct lu_env *env, struct dt_object *obj,
6262 const struct lu_name *cname, const struct lu_fid *pfid)
6264 struct dt_device *dev = lfsck_obj2dev(obj);
6265 struct linkea_data ldata = { NULL };
6266 struct lu_buf linkea_buf;
6269 int fl = LU_XATTR_CREATE;
6273 LASSERT(S_ISDIR(lfsck_object_type(obj)));
6275 rc = lfsck_links_read(env, obj, &ldata);
6276 if (rc == -ENODATA) {
6278 } else if (rc == 0) {
6279 fl = LU_XATTR_REPLACE;
6280 if (ldata.ld_leh->leh_reccount != 1) {
6283 rc = linkea_links_find(&ldata, cname, pfid);
6292 rc = linkea_data_new(&ldata, &lfsck_env_info(env)->lti_linkea_buf);
6296 rc = linkea_add_buf(&ldata, cname, pfid);
6300 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
6301 ldata.ld_leh->leh_len);
6302 th = dt_trans_create(env, dev);
6304 RETURN(PTR_ERR(th));
6306 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
6307 XATTR_NAME_LINK, fl, th);
6311 rc = dt_trans_start_local(env, dev, th);
6315 dt_write_lock(env, obj, 0);
6316 rc = dt_xattr_set(env, obj, &linkea_buf,
6317 XATTR_NAME_LINK, fl, th, BYPASS_CAPA);
6318 dt_write_unlock(env, obj);
6323 dt_trans_stop(env, dev, th);
6328 * Get the name and parent directory's FID from the first linkEA entry.
6330 * \param[in] env pointer to the thread context
6331 * \param[in] obj pointer to the object which get linkEA from
6332 * \param[out] name pointer to the buffer to hold the name
6333 * in the first linkEA entry
6334 * \param[out] pfid pointer to the buffer to hold the parent
6335 * directory's FID in the first linkEA entry
6337 * \retval 0 for success
6338 * \retval negative error number on failure
6340 int lfsck_links_get_first(const struct lu_env *env, struct dt_object *obj,
6341 char *name, struct lu_fid *pfid)
6343 struct lu_name *cname = &lfsck_env_info(env)->lti_name;
6344 struct linkea_data ldata = { NULL };
6347 rc = lfsck_links_read(env, obj, &ldata);
6351 linkea_first_entry(&ldata);
6352 if (ldata.ld_lee == NULL)
6355 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, cname, pfid);
6356 /* To guarantee the 'name' is terminated with '0'. */
6357 memcpy(name, cname->ln_name, cname->ln_namelen);
6358 name[cname->ln_namelen] = 0;
6364 * Update the object's name entry with the given FID.
6366 * \param[in] env pointer to the thread context
6367 * \param[in] lfsck pointer to the lfsck instance
6368 * \param[in] dir pointer to the directory that holds
6370 * \param[in] name the name for the entry to be updated
6371 * \param[in] fid the new FID for the name entry referenced
6372 * \param[in] type the type for the name entry to be updated
6374 * \retval 0 for success
6375 * \retval negative error number on failure
6377 int lfsck_update_name_entry(const struct lu_env *env,
6378 struct lfsck_instance *lfsck,
6379 struct dt_object *dir, const char *name,
6380 const struct lu_fid *fid, __u32 type)
6382 struct lfsck_thread_info *info = lfsck_env_info(env);
6383 struct dt_insert_rec *rec = &info->lti_dt_rec;
6384 struct lfsck_lock_handle *llh = &info->lti_llh;
6385 struct dt_device *dev = lfsck_obj2dev(dir);
6391 rc = lfsck_lock(env, lfsck, dir, name, llh,
6392 MDS_INODELOCK_UPDATE, LCK_PW);
6396 th = dt_trans_create(env, dev);
6398 GOTO(unlock, rc = PTR_ERR(th));
6400 rc = dt_declare_delete(env, dir, (const struct dt_key *)name, th);
6404 rec->rec_type = type;
6406 rc = dt_declare_insert(env, dir, (const struct dt_rec *)rec,
6407 (const struct dt_key *)name, th);
6411 rc = dt_declare_ref_add(env, dir, th);
6415 rc = dt_trans_start_local(env, dev, th);
6419 rc = dt_delete(env, dir, (const struct dt_key *)name, th,
6421 if (rc == -ENOENT) {
6429 rc = dt_insert(env, dir, (const struct dt_rec *)rec,
6430 (const struct dt_key *)name, th, BYPASS_CAPA, 1);
6431 if (rc == 0 && S_ISDIR(type) && !exists) {
6432 dt_write_lock(env, dir, 0);
6433 rc = dt_ref_add(env, dir, th);
6434 dt_write_unlock(env, dir);
6440 dt_trans_stop(env, dev, th);
6444 CDEBUG(D_LFSCK, "%s: update name entry "DFID"/%s with the FID "DFID
6445 " and the type %o: rc = %d\n", lfsck_lfsck2name(lfsck),
6446 PFID(lfsck_dto2fid(dir)), name, PFID(fid), type, rc);
6451 int lfsck_namespace_setup(const struct lu_env *env,
6452 struct lfsck_instance *lfsck)
6454 struct lfsck_component *com;
6455 struct lfsck_namespace *ns;
6456 struct dt_object *root = NULL;
6457 struct dt_object *obj;
6462 LASSERT(lfsck->li_master);
6468 INIT_LIST_HEAD(&com->lc_link);
6469 INIT_LIST_HEAD(&com->lc_link_dir);
6470 init_rwsem(&com->lc_sem);
6471 atomic_set(&com->lc_ref, 1);
6472 com->lc_lfsck = lfsck;
6473 com->lc_type = LFSCK_TYPE_NAMESPACE;
6474 com->lc_ops = &lfsck_namespace_ops;
6475 com->lc_data = lfsck_assistant_data_init(
6476 &lfsck_namespace_assistant_ops,
6478 if (com->lc_data == NULL)
6479 GOTO(out, rc = -ENOMEM);
6481 com->lc_file_size = sizeof(struct lfsck_namespace);
6482 OBD_ALLOC(com->lc_file_ram, com->lc_file_size);
6483 if (com->lc_file_ram == NULL)
6484 GOTO(out, rc = -ENOMEM);
6486 OBD_ALLOC(com->lc_file_disk, com->lc_file_size);
6487 if (com->lc_file_disk == NULL)
6488 GOTO(out, rc = -ENOMEM);
6490 for (i = 0; i < LFSCK_STF_COUNT; i++)
6491 mutex_init(&com->lc_sub_trace_objs[i].lsto_mutex);
6493 root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
6495 GOTO(out, rc = PTR_ERR(root));
6497 if (unlikely(!dt_try_as_dir(env, root)))
6498 GOTO(out, rc = -ENOTDIR);
6500 obj = local_index_find_or_create(env, lfsck->li_los, root,
6502 S_IFREG | S_IRUGO | S_IWUSR,
6503 &dt_lfsck_features);
6505 GOTO(out, rc = PTR_ERR(obj));
6508 rc = lfsck_namespace_load(env, com);
6510 rc = lfsck_namespace_init(env, com);
6512 rc = lfsck_namespace_reset(env, com, true);
6514 rc = lfsck_namespace_load_sub_trace_files(env, com, false);
6518 ns = com->lc_file_ram;
6519 switch (ns->ln_status) {
6524 spin_lock(&lfsck->li_lock);
6525 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
6526 spin_unlock(&lfsck->li_lock);
6529 CERROR("%s: unknown lfsck_namespace status %d\n",
6530 lfsck_lfsck2name(lfsck), ns->ln_status);
6532 case LS_SCANNING_PHASE1:
6533 case LS_SCANNING_PHASE2:
6534 /* No need to store the status to disk right now.
6535 * If the system crashed before the status stored,
6536 * it will be loaded back when next time. */
6537 ns->ln_status = LS_CRASHED;
6541 spin_lock(&lfsck->li_lock);
6542 list_add_tail(&com->lc_link, &lfsck->li_list_scan);
6543 list_add_tail(&com->lc_link_dir, &lfsck->li_list_dir);
6544 spin_unlock(&lfsck->li_lock);
6551 if (root != NULL && !IS_ERR(root))
6552 lfsck_object_put(env, root);
6554 lfsck_component_cleanup(env, com);
6555 CERROR("%s: fail to init namespace LFSCK component: rc = %d\n",
6556 lfsck_lfsck2name(lfsck), rc);