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 0xA0629D03
46 enum lfsck_nameentry_check {
47 LFSCK_NAMEENTRY_DEAD = 1, /* The object has been unlinked. */
48 LFSCK_NAMEENTRY_REMOVED = 2, /* The entry has been removed. */
49 LFSCK_NAMEENTRY_RECREATED = 3, /* The entry has been recreated. */
52 static const char lfsck_namespace_name[] = "lfsck_namespace";
54 static struct lfsck_namespace_req *
55 lfsck_namespace_assistant_req_init(struct lfsck_instance *lfsck,
56 struct lu_dirent *ent, __u16 type)
58 struct lfsck_namespace_req *lnr;
61 size = sizeof(*lnr) + (ent->lde_namelen & ~3) + 4;
64 return ERR_PTR(-ENOMEM);
66 INIT_LIST_HEAD(&lnr->lnr_lar.lar_list);
67 lnr->lnr_obj = lfsck_object_get(lfsck->li_obj_dir);
68 lnr->lnr_lmv = lfsck_lmv_get(lfsck->li_lmv);
69 lnr->lnr_fid = ent->lde_fid;
70 lnr->lnr_oit_cookie = lfsck->li_pos_current.lp_oit_cookie;
71 lnr->lnr_dir_cookie = ent->lde_hash;
72 lnr->lnr_attr = ent->lde_attrs;
75 lnr->lnr_namelen = ent->lde_namelen;
76 memcpy(lnr->lnr_name, ent->lde_name, ent->lde_namelen);
81 static void lfsck_namespace_assistant_req_fini(const struct lu_env *env,
82 struct lfsck_assistant_req *lar)
84 struct lfsck_namespace_req *lnr =
85 container_of0(lar, struct lfsck_namespace_req, lnr_lar);
87 if (lnr->lnr_lmv != NULL)
88 lfsck_lmv_put(env, lnr->lnr_lmv);
90 lu_object_put(env, &lnr->lnr_obj->do_lu);
91 OBD_FREE(lnr, lnr->lnr_size);
94 static void lfsck_namespace_le_to_cpu(struct lfsck_namespace *dst,
95 struct lfsck_namespace *src)
97 dst->ln_magic = le32_to_cpu(src->ln_magic);
98 dst->ln_status = le32_to_cpu(src->ln_status);
99 dst->ln_flags = le32_to_cpu(src->ln_flags);
100 dst->ln_success_count = le32_to_cpu(src->ln_success_count);
101 dst->ln_run_time_phase1 = le32_to_cpu(src->ln_run_time_phase1);
102 dst->ln_run_time_phase2 = le32_to_cpu(src->ln_run_time_phase2);
103 dst->ln_time_last_complete = le64_to_cpu(src->ln_time_last_complete);
104 dst->ln_time_latest_start = le64_to_cpu(src->ln_time_latest_start);
105 dst->ln_time_last_checkpoint =
106 le64_to_cpu(src->ln_time_last_checkpoint);
107 lfsck_position_le_to_cpu(&dst->ln_pos_latest_start,
108 &src->ln_pos_latest_start);
109 lfsck_position_le_to_cpu(&dst->ln_pos_last_checkpoint,
110 &src->ln_pos_last_checkpoint);
111 lfsck_position_le_to_cpu(&dst->ln_pos_first_inconsistent,
112 &src->ln_pos_first_inconsistent);
113 dst->ln_items_checked = le64_to_cpu(src->ln_items_checked);
114 dst->ln_items_repaired = le64_to_cpu(src->ln_items_repaired);
115 dst->ln_items_failed = le64_to_cpu(src->ln_items_failed);
116 dst->ln_dirs_checked = le64_to_cpu(src->ln_dirs_checked);
117 dst->ln_objs_checked_phase2 = le64_to_cpu(src->ln_objs_checked_phase2);
118 dst->ln_objs_repaired_phase2 =
119 le64_to_cpu(src->ln_objs_repaired_phase2);
120 dst->ln_objs_failed_phase2 = le64_to_cpu(src->ln_objs_failed_phase2);
121 dst->ln_objs_nlink_repaired = le64_to_cpu(src->ln_objs_nlink_repaired);
122 fid_le_to_cpu(&dst->ln_fid_latest_scanned_phase2,
123 &src->ln_fid_latest_scanned_phase2);
124 dst->ln_dirent_repaired = le64_to_cpu(src->ln_dirent_repaired);
125 dst->ln_linkea_repaired = le64_to_cpu(src->ln_linkea_repaired);
126 dst->ln_mul_linked_checked = le64_to_cpu(src->ln_mul_linked_checked);
127 dst->ln_mul_linked_repaired = le64_to_cpu(src->ln_mul_linked_repaired);
128 dst->ln_unknown_inconsistency =
129 le64_to_cpu(src->ln_unknown_inconsistency);
130 dst->ln_unmatched_pairs_repaired =
131 le64_to_cpu(src->ln_unmatched_pairs_repaired);
132 dst->ln_dangling_repaired = le64_to_cpu(src->ln_dangling_repaired);
133 dst->ln_mul_ref_repaired = le64_to_cpu(src->ln_mul_ref_repaired);
134 dst->ln_bad_type_repaired = le64_to_cpu(src->ln_bad_type_repaired);
135 dst->ln_lost_dirent_repaired =
136 le64_to_cpu(src->ln_lost_dirent_repaired);
137 dst->ln_striped_dirs_scanned =
138 le64_to_cpu(src->ln_striped_dirs_scanned);
139 dst->ln_striped_dirs_repaired =
140 le64_to_cpu(src->ln_striped_dirs_repaired);
141 dst->ln_striped_dirs_failed =
142 le64_to_cpu(src->ln_striped_dirs_failed);
143 dst->ln_striped_dirs_disabled =
144 le64_to_cpu(src->ln_striped_dirs_disabled);
145 dst->ln_striped_dirs_skipped =
146 le64_to_cpu(src->ln_striped_dirs_skipped);
147 dst->ln_striped_shards_scanned =
148 le64_to_cpu(src->ln_striped_shards_scanned);
149 dst->ln_striped_shards_repaired =
150 le64_to_cpu(src->ln_striped_shards_repaired);
151 dst->ln_striped_shards_failed =
152 le64_to_cpu(src->ln_striped_shards_failed);
153 dst->ln_striped_shards_skipped =
154 le64_to_cpu(src->ln_striped_shards_skipped);
155 dst->ln_name_hash_repaired = le64_to_cpu(src->ln_name_hash_repaired);
156 dst->ln_local_lpf_scanned = le64_to_cpu(src->ln_local_lpf_scanned);
157 dst->ln_local_lpf_moved = le64_to_cpu(src->ln_local_lpf_moved);
158 dst->ln_local_lpf_skipped = le64_to_cpu(src->ln_local_lpf_skipped);
159 dst->ln_local_lpf_failed = le64_to_cpu(src->ln_local_lpf_failed);
160 dst->ln_bitmap_size = le32_to_cpu(src->ln_bitmap_size);
163 static void lfsck_namespace_cpu_to_le(struct lfsck_namespace *dst,
164 struct lfsck_namespace *src)
166 dst->ln_magic = cpu_to_le32(src->ln_magic);
167 dst->ln_status = cpu_to_le32(src->ln_status);
168 dst->ln_flags = cpu_to_le32(src->ln_flags);
169 dst->ln_success_count = cpu_to_le32(src->ln_success_count);
170 dst->ln_run_time_phase1 = cpu_to_le32(src->ln_run_time_phase1);
171 dst->ln_run_time_phase2 = cpu_to_le32(src->ln_run_time_phase2);
172 dst->ln_time_last_complete = cpu_to_le64(src->ln_time_last_complete);
173 dst->ln_time_latest_start = cpu_to_le64(src->ln_time_latest_start);
174 dst->ln_time_last_checkpoint =
175 cpu_to_le64(src->ln_time_last_checkpoint);
176 lfsck_position_cpu_to_le(&dst->ln_pos_latest_start,
177 &src->ln_pos_latest_start);
178 lfsck_position_cpu_to_le(&dst->ln_pos_last_checkpoint,
179 &src->ln_pos_last_checkpoint);
180 lfsck_position_cpu_to_le(&dst->ln_pos_first_inconsistent,
181 &src->ln_pos_first_inconsistent);
182 dst->ln_items_checked = cpu_to_le64(src->ln_items_checked);
183 dst->ln_items_repaired = cpu_to_le64(src->ln_items_repaired);
184 dst->ln_items_failed = cpu_to_le64(src->ln_items_failed);
185 dst->ln_dirs_checked = cpu_to_le64(src->ln_dirs_checked);
186 dst->ln_objs_checked_phase2 = cpu_to_le64(src->ln_objs_checked_phase2);
187 dst->ln_objs_repaired_phase2 =
188 cpu_to_le64(src->ln_objs_repaired_phase2);
189 dst->ln_objs_failed_phase2 = cpu_to_le64(src->ln_objs_failed_phase2);
190 dst->ln_objs_nlink_repaired = cpu_to_le64(src->ln_objs_nlink_repaired);
191 fid_cpu_to_le(&dst->ln_fid_latest_scanned_phase2,
192 &src->ln_fid_latest_scanned_phase2);
193 dst->ln_dirent_repaired = cpu_to_le64(src->ln_dirent_repaired);
194 dst->ln_linkea_repaired = cpu_to_le64(src->ln_linkea_repaired);
195 dst->ln_mul_linked_checked = cpu_to_le64(src->ln_mul_linked_checked);
196 dst->ln_mul_linked_repaired = cpu_to_le64(src->ln_mul_linked_repaired);
197 dst->ln_unknown_inconsistency =
198 cpu_to_le64(src->ln_unknown_inconsistency);
199 dst->ln_unmatched_pairs_repaired =
200 cpu_to_le64(src->ln_unmatched_pairs_repaired);
201 dst->ln_dangling_repaired = cpu_to_le64(src->ln_dangling_repaired);
202 dst->ln_mul_ref_repaired = cpu_to_le64(src->ln_mul_ref_repaired);
203 dst->ln_bad_type_repaired = cpu_to_le64(src->ln_bad_type_repaired);
204 dst->ln_lost_dirent_repaired =
205 cpu_to_le64(src->ln_lost_dirent_repaired);
206 dst->ln_striped_dirs_scanned =
207 cpu_to_le64(src->ln_striped_dirs_scanned);
208 dst->ln_striped_dirs_repaired =
209 cpu_to_le64(src->ln_striped_dirs_repaired);
210 dst->ln_striped_dirs_failed =
211 cpu_to_le64(src->ln_striped_dirs_failed);
212 dst->ln_striped_dirs_disabled =
213 cpu_to_le64(src->ln_striped_dirs_disabled);
214 dst->ln_striped_dirs_skipped =
215 cpu_to_le64(src->ln_striped_dirs_skipped);
216 dst->ln_striped_shards_scanned =
217 cpu_to_le64(src->ln_striped_shards_scanned);
218 dst->ln_striped_shards_repaired =
219 cpu_to_le64(src->ln_striped_shards_repaired);
220 dst->ln_striped_shards_failed =
221 cpu_to_le64(src->ln_striped_shards_failed);
222 dst->ln_striped_shards_skipped =
223 cpu_to_le64(src->ln_striped_shards_skipped);
224 dst->ln_name_hash_repaired = cpu_to_le64(src->ln_name_hash_repaired);
225 dst->ln_local_lpf_scanned = cpu_to_le64(src->ln_local_lpf_scanned);
226 dst->ln_local_lpf_moved = cpu_to_le64(src->ln_local_lpf_moved);
227 dst->ln_local_lpf_skipped = cpu_to_le64(src->ln_local_lpf_skipped);
228 dst->ln_local_lpf_failed = cpu_to_le64(src->ln_local_lpf_failed);
229 dst->ln_bitmap_size = cpu_to_le32(src->ln_bitmap_size);
232 static void lfsck_namespace_record_failure(const struct lu_env *env,
233 struct lfsck_instance *lfsck,
234 struct lfsck_namespace *ns)
236 struct lfsck_position pos;
238 ns->ln_items_failed++;
239 lfsck_pos_fill(env, lfsck, &pos, false);
240 if (lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent) ||
241 lfsck_pos_is_eq(&pos, &ns->ln_pos_first_inconsistent) < 0) {
242 ns->ln_pos_first_inconsistent = pos;
244 CDEBUG(D_LFSCK, "%s: namespace LFSCK hit first non-repaired "
245 "inconsistency at the pos ["LPU64", "DFID", "LPX64"]\n",
246 lfsck_lfsck2name(lfsck),
247 ns->ln_pos_first_inconsistent.lp_oit_cookie,
248 PFID(&ns->ln_pos_first_inconsistent.lp_dir_parent),
249 ns->ln_pos_first_inconsistent.lp_dir_cookie);
254 * Load the MDT bitmap from the lfsck_namespace trace file.
256 * \param[in] env pointer to the thread context
257 * \param[in] com pointer to the lfsck component
259 * \retval 0 for success
260 * \retval negative error number on failure or data corruption
262 static int lfsck_namespace_load_bitmap(const struct lu_env *env,
263 struct lfsck_component *com)
265 struct dt_object *obj = com->lc_obj;
266 struct lfsck_assistant_data *lad = com->lc_data;
267 struct lfsck_namespace *ns = com->lc_file_ram;
268 cfs_bitmap_t *bitmap = lad->lad_bitmap;
274 if (com->lc_lfsck->li_mdt_descs.ltd_tgts_bitmap->size >
276 nbits = com->lc_lfsck->li_mdt_descs.ltd_tgts_bitmap->size;
278 nbits = ns->ln_bitmap_size;
280 if (unlikely(nbits < BITS_PER_LONG))
281 nbits = BITS_PER_LONG;
283 if (nbits > bitmap->size) {
284 __u32 new_bits = bitmap->size;
285 cfs_bitmap_t *new_bitmap;
287 while (new_bits < nbits)
290 new_bitmap = CFS_ALLOCATE_BITMAP(new_bits);
291 if (new_bitmap == NULL)
294 lad->lad_bitmap = new_bitmap;
295 CFS_FREE_BITMAP(bitmap);
299 if (ns->ln_bitmap_size == 0) {
300 lad->lad_incomplete = 0;
301 CFS_RESET_BITMAP(bitmap);
306 size = (ns->ln_bitmap_size + 7) >> 3;
307 rc = dt_xattr_get(env, obj,
308 lfsck_buf_get(env, bitmap->data, size),
309 XATTR_NAME_LFSCK_BITMAP, BYPASS_CAPA);
311 RETURN(rc >= 0 ? -EINVAL : rc);
313 if (cfs_bitmap_check_empty(bitmap))
314 lad->lad_incomplete = 0;
316 lad->lad_incomplete = 1;
322 * Load namespace LFSCK statistics information from the trace file.
324 * For old release (Lustre-2.6 or older), the statistics information was
325 * stored as XATTR_NAME_LFSCK_NAMESPACE_OLD EA. But in Lustre-2.7, we need
326 * more statistics information. To avoid confusing old MDT when downgrade,
327 * Lustre-2.7 stores the namespace LFSCK statistics information as new
328 * XATTR_NAME_LFSCK_NAMESPACE EA.
330 * \param[in] env pointer to the thread context
331 * \param[in] com pointer to the lfsck component
333 * \retval 0 for success
334 * \retval negative error number on failure
336 static int lfsck_namespace_load(const struct lu_env *env,
337 struct lfsck_component *com)
339 int len = com->lc_file_size;
342 rc = dt_xattr_get(env, com->lc_obj,
343 lfsck_buf_get(env, com->lc_file_disk, len),
344 XATTR_NAME_LFSCK_NAMESPACE, BYPASS_CAPA);
346 struct lfsck_namespace *ns = com->lc_file_ram;
348 lfsck_namespace_le_to_cpu(ns,
349 (struct lfsck_namespace *)com->lc_file_disk);
350 if (ns->ln_magic != LFSCK_NAMESPACE_MAGIC) {
351 CDEBUG(D_LFSCK, "%s: invalid lfsck_namespace magic "
352 "%#x != %#x\n", lfsck_lfsck2name(com->lc_lfsck),
353 ns->ln_magic, LFSCK_NAMESPACE_MAGIC);
358 } else if (rc != -ENODATA) {
359 CDEBUG(D_LFSCK, "%s: fail to load lfsck_namespace, "
360 "expected = %d: rc = %d\n",
361 lfsck_lfsck2name(com->lc_lfsck), len, rc);
365 /* Check whether it is old trace file or not.
366 * If yes, it should be reset via returning -ESTALE. */
367 rc = dt_xattr_get(env, com->lc_obj,
368 lfsck_buf_get(env, com->lc_file_disk, len),
369 XATTR_NAME_LFSCK_NAMESPACE_OLD, BYPASS_CAPA);
377 static int lfsck_namespace_store(const struct lu_env *env,
378 struct lfsck_component *com, bool init)
380 struct dt_object *obj = com->lc_obj;
381 struct lfsck_instance *lfsck = com->lc_lfsck;
382 struct lfsck_namespace *ns = com->lc_file_ram;
383 struct lfsck_assistant_data *lad = com->lc_data;
384 cfs_bitmap_t *bitmap = NULL;
385 struct thandle *handle;
387 int len = com->lc_file_size;
389 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 8, 53, 0)
390 struct lu_buf tbuf = { &len, sizeof(len) };
395 bitmap = lad->lad_bitmap;
396 nbits = bitmap->size;
399 LASSERTF((nbits & 7) == 0, "Invalid nbits %u\n", nbits);
402 ns->ln_bitmap_size = nbits;
403 lfsck_namespace_cpu_to_le((struct lfsck_namespace *)com->lc_file_disk,
405 handle = dt_trans_create(env, lfsck->li_bottom);
407 GOTO(log, rc = PTR_ERR(handle));
409 rc = dt_declare_xattr_set(env, obj,
410 lfsck_buf_get(env, com->lc_file_disk, len),
411 XATTR_NAME_LFSCK_NAMESPACE, 0, handle);
415 if (bitmap != NULL) {
416 rc = dt_declare_xattr_set(env, obj,
417 lfsck_buf_get(env, bitmap->data, nbits >> 3),
418 XATTR_NAME_LFSCK_BITMAP, 0, handle);
423 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 8, 53, 0)
424 /* To be compatible with old Lustre-2.x MDT (x <= 6), generate dummy
425 * XATTR_NAME_LFSCK_NAMESPACE_OLD EA, then when downgrade to Lustre-2.x,
426 * the old LFSCK will find "invalid" XATTR_NAME_LFSCK_NAMESPACE_OLD EA,
427 * then reset the namespace LFSCK trace file. */
429 rc = dt_declare_xattr_set(env, obj, &tbuf,
430 XATTR_NAME_LFSCK_NAMESPACE_OLD,
431 LU_XATTR_CREATE, handle);
437 rc = dt_trans_start_local(env, lfsck->li_bottom, handle);
441 rc = dt_xattr_set(env, obj,
442 lfsck_buf_get(env, com->lc_file_disk, len),
443 XATTR_NAME_LFSCK_NAMESPACE, 0, handle, BYPASS_CAPA);
444 if (rc == 0 && bitmap != NULL)
445 rc = dt_xattr_set(env, obj,
446 lfsck_buf_get(env, bitmap->data, nbits >> 3),
447 XATTR_NAME_LFSCK_BITMAP, 0, handle,
450 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 8, 53, 0)
452 rc = dt_xattr_set(env, obj, &tbuf,
453 XATTR_NAME_LFSCK_NAMESPACE_OLD,
454 LU_XATTR_CREATE, handle, BYPASS_CAPA);
460 dt_trans_stop(env, lfsck->li_bottom, handle);
464 CDEBUG(D_LFSCK, "%s: fail to store lfsck_namespace: rc = %d\n",
465 lfsck_lfsck2name(lfsck), rc);
469 static int lfsck_namespace_init(const struct lu_env *env,
470 struct lfsck_component *com)
472 struct lfsck_namespace *ns = com->lc_file_ram;
475 memset(ns, 0, sizeof(*ns));
476 ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
477 ns->ln_status = LS_INIT;
478 down_write(&com->lc_sem);
479 rc = lfsck_namespace_store(env, com, true);
480 up_write(&com->lc_sem);
485 * Update the namespace LFSCK trace file for the given @fid
487 * \param[in] env pointer to the thread context
488 * \param[in] com pointer to the lfsck component
489 * \param[in] fid the fid which flags to be updated in the lfsck
491 * \param[in] add true if add new flags, otherwise remove flags
493 * \retval 0 for succeed or nothing to be done
494 * \retval negative error number on failure
496 int lfsck_namespace_trace_update(const struct lu_env *env,
497 struct lfsck_component *com,
498 const struct lu_fid *fid,
499 const __u8 flags, bool add)
501 struct lfsck_instance *lfsck = com->lc_lfsck;
502 struct dt_object *obj = com->lc_obj;
503 struct lu_fid *key = &lfsck_env_info(env)->lti_fid3;
504 struct dt_device *dev = lfsck->li_bottom;
505 struct thandle *th = NULL;
513 down_write(&com->lc_sem);
514 fid_cpu_to_be(key, fid);
515 rc = dt_lookup(env, obj, (struct dt_rec *)&old,
516 (const struct dt_key *)key, BYPASS_CAPA);
519 GOTO(unlock, rc = 0);
523 } else if (rc == 0) {
525 if ((old & flags) == flags)
526 GOTO(unlock, rc = 0);
530 if ((old & flags) == 0)
531 GOTO(unlock, rc = 0);
539 th = dt_trans_create(env, dev);
541 GOTO(log, rc = PTR_ERR(th));
544 rc = dt_declare_delete(env, obj,
545 (const struct dt_key *)key, th);
551 rc = dt_declare_insert(env, obj,
552 (const struct dt_rec *)&new,
553 (const struct dt_key *)key, th);
558 rc = dt_trans_start_local(env, dev, th);
563 rc = dt_delete(env, obj, (const struct dt_key *)key,
570 rc = dt_insert(env, obj, (const struct dt_rec *)&new,
571 (const struct dt_key *)key, th, BYPASS_CAPA, 1);
579 if (th != NULL && !IS_ERR(th))
580 dt_trans_stop(env, dev, th);
582 CDEBUG(D_LFSCK, "%s: namespace LFSCK %s flags for "DFID" in the "
583 "trace file, flags %x, old %x, new %x: rc = %d\n",
584 lfsck_lfsck2name(lfsck), add ? "add" : "del", PFID(fid),
585 (__u32)flags, (__u32)old, (__u32)new, rc);
588 up_write(&com->lc_sem);
593 int lfsck_namespace_check_exist(const struct lu_env *env,
594 struct dt_object *dir,
595 struct dt_object *obj, const char *name)
597 struct lu_fid *fid = &lfsck_env_info(env)->lti_fid;
601 if (unlikely(lfsck_is_dead_obj(obj)))
602 RETURN(LFSCK_NAMEENTRY_DEAD);
604 rc = dt_lookup(env, dir, (struct dt_rec *)fid,
605 (const struct dt_key *)name, BYPASS_CAPA);
607 RETURN(LFSCK_NAMEENTRY_REMOVED);
612 if (!lu_fid_eq(fid, lfsck_dto2fid(obj)))
613 RETURN(LFSCK_NAMEENTRY_RECREATED);
618 static int lfsck_declare_namespace_exec_dir(const struct lu_env *env,
619 struct dt_object *obj,
620 struct thandle *handle)
624 /* For destroying all invalid linkEA entries. */
625 rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, handle);
629 /* For insert new linkEA entry. */
630 rc = dt_declare_xattr_set(env, obj,
631 lfsck_buf_get_const(env, NULL, DEFAULT_LINKEA_SIZE),
632 XATTR_NAME_LINK, 0, handle);
636 int __lfsck_links_read(const struct lu_env *env, struct dt_object *obj,
637 struct linkea_data *ldata)
641 if (ldata->ld_buf->lb_buf == NULL)
644 if (!dt_object_exists(obj))
647 rc = dt_xattr_get(env, obj, ldata->ld_buf, XATTR_NAME_LINK, BYPASS_CAPA);
649 /* Buf was too small, figure out what we need. */
650 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LINK,
655 lu_buf_realloc(ldata->ld_buf, rc);
656 if (ldata->ld_buf->lb_buf == NULL)
659 rc = dt_xattr_get(env, obj, ldata->ld_buf, XATTR_NAME_LINK,
664 rc = linkea_init(ldata);
670 * Remove linkEA for the given object.
672 * The caller should take the ldlm lock before the calling.
674 * \param[in] env pointer to the thread context
675 * \param[in] com pointer to the lfsck component
676 * \param[in] obj pointer to the dt_object to be handled
678 * \retval 0 for repaired cases
679 * \retval negative error number on failure
681 static int lfsck_namespace_links_remove(const struct lu_env *env,
682 struct lfsck_component *com,
683 struct dt_object *obj)
685 struct lfsck_instance *lfsck = com->lc_lfsck;
686 struct dt_device *dev = lfsck->li_bottom;
687 struct thandle *th = NULL;
691 LASSERT(dt_object_remote(obj) == 0);
693 th = dt_trans_create(env, dev);
695 GOTO(log, rc = PTR_ERR(th));
697 rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, th);
701 rc = dt_trans_start_local(env, dev, th);
705 dt_write_lock(env, obj, 0);
706 if (unlikely(lfsck_is_dead_obj(obj)))
707 GOTO(unlock, rc = -ENOENT);
709 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
710 GOTO(unlock, rc = 0);
712 rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, th, BYPASS_CAPA);
717 dt_write_unlock(env, obj);
720 dt_trans_stop(env, dev, th);
723 CDEBUG(D_LFSCK, "%s: namespace LFSCK remove invalid linkEA "
724 "for the object "DFID": rc = %d\n",
725 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
728 struct lfsck_namespace *ns = com->lc_file_ram;
730 ns->ln_flags |= LF_INCONSISTENT;
736 static int lfsck_links_write(const struct lu_env *env, struct dt_object *obj,
737 struct linkea_data *ldata, struct thandle *handle)
739 const struct lu_buf *buf = lfsck_buf_get_const(env,
740 ldata->ld_buf->lb_buf,
741 ldata->ld_leh->leh_len);
743 return dt_xattr_set(env, obj, buf, XATTR_NAME_LINK, 0, handle,
747 static void lfsck_namespace_unpack_linkea_entry(struct linkea_data *ldata,
748 struct lu_name *cname,
752 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, cname, pfid);
753 /* To guarantee the 'name' is terminated with '0'. */
754 memcpy(buf, cname->ln_name, cname->ln_namelen);
755 buf[cname->ln_namelen] = 0;
756 cname->ln_name = buf;
759 static int lfsck_namespace_filter_linkea_entry(struct linkea_data *ldata,
760 struct lu_name *cname,
764 struct link_ea_entry *oldlee;
768 oldlee = ldata->ld_lee;
769 oldlen = ldata->ld_reclen;
770 linkea_next_entry(ldata);
771 while (ldata->ld_lee != NULL) {
772 ldata->ld_reclen = (ldata->ld_lee->lee_reclen[0] << 8) |
773 ldata->ld_lee->lee_reclen[1];
774 if (unlikely(ldata->ld_reclen == oldlen &&
775 memcmp(ldata->ld_lee, oldlee, oldlen) == 0)) {
780 linkea_del_buf(ldata, cname);
782 linkea_next_entry(ldata);
785 ldata->ld_lee = oldlee;
786 ldata->ld_reclen = oldlen;
792 * Insert orphan into .lustre/lost+found/MDTxxxx/ locally.
794 * Add the specified orphan MDT-object to the .lustre/lost+found/MDTxxxx/
795 * with the given type to generate the name, the detailed rules for name
796 * have been described as following.
798 * The function also generates the linkEA corresponding to the name entry
799 * under the .lustre/lost+found/MDTxxxx/ for the orphan MDT-object.
801 * \param[in] env pointer to the thread context
802 * \param[in] com pointer to the lfsck component
803 * \param[in] orphan pointer to the orphan MDT-object
804 * \param[in] infix additional information for the orphan name, such as
805 * the FID for original
806 * \param[in] type the type for describing why the orphan MDT-object is
807 * created. The rules are as following:
809 * type "D": The MDT-object is a directory, it may knows its parent
810 * but because there is no valid linkEA, the LFSCK cannot
811 * know where to put it back to the namespace.
812 * type "O": The MDT-object has no linkEA, and there is no name
813 * entry that references the MDT-object.
815 * type "S": The orphan MDT-object is a shard of a striped directory
817 * \see lfsck_layout_recreate_parent() for more types.
819 * The orphan name will be like:
820 * ${FID}-${infix}-${type}-${conflict_version}
822 * \param[out] count if some others inserted some linkEA entries by race,
823 * then return the linkEA entries count.
825 * \retval positive number for repaired cases
826 * \retval 0 if needs to repair nothing
827 * \retval negative error number on failure
829 static int lfsck_namespace_insert_orphan(const struct lu_env *env,
830 struct lfsck_component *com,
831 struct dt_object *orphan,
832 const char *infix, const char *type,
835 struct lfsck_thread_info *info = lfsck_env_info(env);
836 struct lu_name *cname = &info->lti_name;
837 struct dt_insert_rec *rec = &info->lti_dt_rec;
838 struct lu_attr *la = &info->lti_la3;
839 const struct lu_fid *cfid = lfsck_dto2fid(orphan);
840 const struct lu_fid *pfid;
842 struct lfsck_instance *lfsck = com->lc_lfsck;
843 struct dt_device *dev = lfsck->li_bottom;
844 struct dt_object *parent;
845 struct thandle *th = NULL;
846 struct lustre_handle plh = { 0 };
847 struct lustre_handle clh = { 0 };
848 struct linkea_data ldata = { NULL };
849 struct lu_buf linkea_buf;
856 cname->ln_name = NULL;
857 if (unlikely(lfsck->li_lpf_obj == NULL))
858 GOTO(log, rc = -ENXIO);
860 parent = lfsck->li_lpf_obj;
861 pfid = lfsck_dto2fid(parent);
863 /* Hold update lock on the parent to prevent others to access. */
864 rc = lfsck_ibits_lock(env, lfsck, parent, &plh,
865 MDS_INODELOCK_UPDATE, LCK_EX);
870 namelen = snprintf(info->lti_key, NAME_MAX, DFID"%s-%s-%d",
871 PFID(cfid), infix, type, idx++);
872 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
873 (const struct dt_key *)info->lti_key,
875 if (rc != 0 && rc != -ENOENT)
878 if (unlikely(rc == 0 && lu_fid_eq(cfid, &tfid)))
880 } while (rc == 0 && !exist);
882 cname->ln_name = info->lti_key;
883 cname->ln_namelen = namelen;
884 rc = linkea_data_new(&ldata, &info->lti_linkea_buf2);
888 rc = linkea_add_buf(&ldata, cname, pfid);
892 rc = lfsck_ibits_lock(env, lfsck, orphan, &clh,
893 MDS_INODELOCK_UPDATE | MDS_INODELOCK_LOOKUP,
898 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
899 ldata.ld_leh->leh_len);
900 th = dt_trans_create(env, dev);
902 GOTO(log, rc = PTR_ERR(th));
904 if (S_ISDIR(lfsck_object_type(orphan))) {
905 rc = dt_declare_delete(env, orphan,
906 (const struct dt_key *)dotdot, th);
910 rec->rec_type = S_IFDIR;
912 rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
913 (const struct dt_key *)dotdot, th);
918 rc = dt_declare_xattr_set(env, orphan, &linkea_buf,
919 XATTR_NAME_LINK, 0, th);
924 rec->rec_type = lfsck_object_type(orphan) & S_IFMT;
926 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
927 (const struct dt_key *)cname->ln_name,
932 if (S_ISDIR(rec->rec_type)) {
933 rc = dt_declare_ref_add(env, parent, th);
939 memset(la, 0, sizeof(*la));
940 la->la_ctime = cfs_time_current_sec();
941 la->la_valid = LA_CTIME;
942 rc = dt_declare_attr_set(env, orphan, la, th);
946 rc = dt_trans_start_local(env, dev, th);
950 dt_write_lock(env, orphan, 0);
951 rc = lfsck_links_read(env, orphan, &ldata);
952 if (likely((rc == -ENODATA) || (rc == -EINVAL) ||
953 (rc == 0 && ldata.ld_leh->leh_reccount == 0))) {
954 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
955 GOTO(unlock, rc = 1);
957 if (S_ISDIR(lfsck_object_type(orphan))) {
958 rc = dt_delete(env, orphan,
959 (const struct dt_key *)dotdot, th,
964 rec->rec_type = S_IFDIR;
966 rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
967 (const struct dt_key *)dotdot, th,
973 rc = dt_xattr_set(env, orphan, &linkea_buf, XATTR_NAME_LINK, 0,
976 if (rc == 0 && count != NULL)
977 *count = ldata.ld_leh->leh_reccount;
981 dt_write_unlock(env, orphan);
983 if (rc == 0 && !exist) {
984 rec->rec_type = lfsck_object_type(orphan) & S_IFMT;
986 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
987 (const struct dt_key *)cname->ln_name,
989 if (rc == 0 && S_ISDIR(rec->rec_type)) {
990 dt_write_lock(env, parent, 0);
991 rc = dt_ref_add(env, parent, th);
992 dt_write_unlock(env, parent);
997 rc = dt_attr_set(env, orphan, la, th, BYPASS_CAPA);
999 GOTO(stop, rc = (rc == 0 ? 1 : rc));
1002 dt_write_unlock(env, orphan);
1005 dt_trans_stop(env, dev, th);
1008 lfsck_ibits_unlock(&clh, LCK_EX);
1009 lfsck_ibits_unlock(&plh, LCK_EX);
1010 CDEBUG(D_LFSCK, "%s: namespace LFSCK insert orphan for the "
1011 "object "DFID", name = %s: rc = %d\n",
1012 lfsck_lfsck2name(lfsck), PFID(cfid),
1013 cname->ln_name != NULL ? cname->ln_name : "<NULL>", rc);
1016 struct lfsck_namespace *ns = com->lc_file_ram;
1018 ns->ln_flags |= LF_INCONSISTENT;
1025 * Add the specified name entry back to namespace.
1027 * If there is a linkEA entry that back references a name entry under
1028 * some parent directory, but such parent directory does not have the
1029 * claimed name entry. On the other hand, the linkEA entries count is
1030 * not larger than the MDT-object's hard link count. Under such case,
1031 * it is quite possible that the name entry is lost. Then the LFSCK
1032 * should add the name entry back to the namespace.
1034 * \param[in] env pointer to the thread context
1035 * \param[in] com pointer to the lfsck component
1036 * \param[in] parent pointer to the directory under which the name entry
1037 * will be inserted into
1038 * \param[in] child pointer to the object referenced by the name entry
1039 * that to be inserted into the parent
1040 * \param[in] name the name for the child in the parent directory
1042 * \retval positive number for repaired cases
1043 * \retval 0 if nothing to be repaired
1044 * \retval negative error number on failure
1046 static int lfsck_namespace_insert_normal(const struct lu_env *env,
1047 struct lfsck_component *com,
1048 struct dt_object *parent,
1049 struct dt_object *child,
1052 struct lfsck_thread_info *info = lfsck_env_info(env);
1053 struct lu_attr *la = &info->lti_la;
1054 struct dt_insert_rec *rec = &info->lti_dt_rec;
1055 struct lfsck_instance *lfsck = com->lc_lfsck;
1056 struct dt_device *dev = lfsck->li_next;
1057 struct thandle *th = NULL;
1058 struct lustre_handle lh = { 0 };
1062 if (unlikely(!dt_try_as_dir(env, parent)))
1063 GOTO(log, rc = -ENOTDIR);
1065 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1068 /* Hold update lock on the parent to prevent others to access. */
1069 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
1070 MDS_INODELOCK_UPDATE, LCK_EX);
1074 th = dt_trans_create(env, dev);
1076 GOTO(unlock, rc = PTR_ERR(th));
1078 rec->rec_type = lfsck_object_type(child) & S_IFMT;
1079 rec->rec_fid = lfsck_dto2fid(child);
1080 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1081 (const struct dt_key *)name, th);
1085 if (S_ISDIR(rec->rec_type)) {
1086 rc = dt_declare_ref_add(env, parent, th);
1091 memset(la, 0, sizeof(*la));
1092 la->la_ctime = cfs_time_current_sec();
1093 la->la_valid = LA_CTIME;
1094 rc = dt_declare_attr_set(env, parent, la, th);
1098 rc = dt_declare_attr_set(env, child, la, th);
1102 rc = dt_trans_start_local(env, dev, th);
1106 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1107 (const struct dt_key *)name, th, BYPASS_CAPA, 1);
1111 if (S_ISDIR(rec->rec_type)) {
1112 dt_write_lock(env, parent, 0);
1113 rc = dt_ref_add(env, parent, th);
1114 dt_write_unlock(env, parent);
1119 la->la_ctime = cfs_time_current_sec();
1120 rc = dt_attr_set(env, parent, la, th, BYPASS_CAPA);
1124 rc = dt_attr_set(env, child, la, th, BYPASS_CAPA);
1126 GOTO(stop, rc = (rc == 0 ? 1 : rc));
1129 dt_trans_stop(env, dev, th);
1132 lfsck_ibits_unlock(&lh, LCK_EX);
1135 CDEBUG(D_LFSCK, "%s: namespace LFSCK insert object "DFID" with "
1136 "the name %s and type %o to the parent "DFID": rc = %d\n",
1137 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(child)), name,
1138 lfsck_object_type(child) & S_IFMT,
1139 PFID(lfsck_dto2fid(parent)), rc);
1142 struct lfsck_namespace *ns = com->lc_file_ram;
1144 ns->ln_flags |= LF_INCONSISTENT;
1146 ns->ln_lost_dirent_repaired++;
1153 * Create the specified orphan directory.
1155 * For the case that the parent MDT-object stored in some MDT-object's
1156 * linkEA entry is lost, the LFSCK will re-create the parent object as
1157 * an orphan and insert it into .lustre/lost+found/MDTxxxx/ directory
1158 * with the name ${FID}-P-${conflict_version}.
1160 * \param[in] env pointer to the thread context
1161 * \param[in] com pointer to the lfsck component
1162 * \param[in] orphan pointer to the orphan MDT-object to be created
1163 * \param[in] lmv pointer to master LMV EA that will be set to the orphan
1165 * \retval positive number for repaired cases
1166 * \retval negative error number on failure
1168 static int lfsck_namespace_create_orphan_dir(const struct lu_env *env,
1169 struct lfsck_component *com,
1170 struct dt_object *orphan,
1171 struct lmv_mds_md_v1 *lmv)
1173 struct lfsck_thread_info *info = lfsck_env_info(env);
1174 struct lu_attr *la = &info->lti_la;
1175 struct dt_allocation_hint *hint = &info->lti_hint;
1176 struct dt_object_format *dof = &info->lti_dof;
1177 struct lu_name *cname = &info->lti_name2;
1178 struct dt_insert_rec *rec = &info->lti_dt_rec;
1179 struct lmv_mds_md_v1 *lmv2 = &info->lti_lmv2;
1180 const struct lu_fid *cfid = lfsck_dto2fid(orphan);
1182 struct lfsck_instance *lfsck = com->lc_lfsck;
1183 struct lfsck_namespace *ns = com->lc_file_ram;
1184 struct dt_device *dev;
1185 struct dt_object *parent = NULL;
1186 struct dt_object *child = NULL;
1187 struct thandle *th = NULL;
1188 struct lustre_handle lh = { 0 };
1189 struct linkea_data ldata = { NULL };
1190 struct lu_buf linkea_buf;
1191 struct lu_buf lmv_buf;
1198 LASSERT(!dt_object_exists(orphan));
1200 cname->ln_name = NULL;
1201 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1204 if (dt_object_remote(orphan)) {
1205 LASSERT(lfsck->li_lpf_root_obj != NULL);
1207 idx = lfsck_find_mdt_idx_by_fid(env, lfsck, cfid);
1209 GOTO(log, rc = idx);
1211 snprintf(name, 8, "MDT%04x", idx);
1212 rc = dt_lookup(env, lfsck->li_lpf_root_obj,
1213 (struct dt_rec *)&tfid,
1214 (const struct dt_key *)name, BYPASS_CAPA);
1216 GOTO(log, rc = (rc == -ENOENT ? -ENXIO : rc));
1218 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
1220 GOTO(log, rc = PTR_ERR(parent));
1222 if (unlikely(!dt_try_as_dir(env, parent)))
1223 GOTO(log, rc = -ENOTDIR);
1225 if (unlikely(lfsck->li_lpf_obj == NULL))
1226 GOTO(log, rc = -ENXIO);
1228 parent = lfsck->li_lpf_obj;
1231 dev = lfsck_find_dev_by_fid(env, lfsck, cfid);
1233 GOTO(log, rc = PTR_ERR(dev));
1235 child = lfsck_object_find_by_dev(env, dev, cfid);
1237 GOTO(log, rc = PTR_ERR(child));
1239 /* Hold update lock on the parent to prevent others to access. */
1240 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
1241 MDS_INODELOCK_UPDATE, LCK_EX);
1247 namelen = snprintf(name, 31, DFID"-P-%d",
1249 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
1250 (const struct dt_key *)name, BYPASS_CAPA);
1251 if (rc != 0 && rc != -ENOENT)
1255 cname->ln_name = name;
1256 cname->ln_namelen = namelen;
1258 memset(la, 0, sizeof(*la));
1259 la->la_mode = S_IFDIR | 0700;
1260 la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
1261 LA_ATIME | LA_MTIME | LA_CTIME;
1263 child->do_ops->do_ah_init(env, hint, parent, child,
1264 la->la_mode & S_IFMT);
1266 memset(dof, 0, sizeof(*dof));
1267 dof->dof_type = dt_mode_to_dft(S_IFDIR);
1269 rc = linkea_data_new(&ldata, &info->lti_linkea_buf2);
1273 rc = linkea_add_buf(&ldata, cname, lfsck_dto2fid(parent));
1277 th = dt_trans_create(env, dev);
1279 GOTO(unlock1, rc = PTR_ERR(th));
1281 /* Sync the remote transaction to guarantee that the subsequent
1282 * lock against the @orphan can find the @orphan in time. */
1283 if (dt_object_remote(orphan))
1286 rc = dt_declare_create(env, child, la, hint, dof, th);
1290 if (unlikely(!dt_try_as_dir(env, child)))
1291 GOTO(stop, rc = -ENOTDIR);
1293 rec->rec_type = S_IFDIR;
1294 rec->rec_fid = cfid;
1295 rc = dt_declare_insert(env, child, (const struct dt_rec *)rec,
1296 (const struct dt_key *)dot, th);
1300 rec->rec_fid = lfsck_dto2fid(parent);
1301 rc = dt_declare_insert(env, child, (const struct dt_rec *)rec,
1302 (const struct dt_key *)dotdot, th);
1304 rc = dt_declare_ref_add(env, child, th);
1309 rc = dt_declare_ref_add(env, child, th);
1314 lmv->lmv_magic = LMV_MAGIC;
1315 lmv->lmv_master_mdt_index = lfsck_dev_idx(dev);
1316 lfsck_lmv_header_cpu_to_le(lmv2, lmv);
1317 lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
1318 rc = dt_declare_xattr_set(env, child, &lmv_buf,
1319 XATTR_NAME_LMV, 0, th);
1324 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
1325 ldata.ld_leh->leh_len);
1326 rc = dt_declare_xattr_set(env, child, &linkea_buf,
1327 XATTR_NAME_LINK, 0, th);
1331 rec->rec_fid = cfid;
1332 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1333 (const struct dt_key *)name, th);
1335 rc = dt_declare_ref_add(env, parent, th);
1340 rc = dt_trans_start_local(env, dev, th);
1344 dt_write_lock(env, child, 0);
1345 rc = dt_create(env, child, la, hint, dof, th);
1349 rec->rec_fid = cfid;
1350 rc = dt_insert(env, child, (const struct dt_rec *)rec,
1351 (const struct dt_key *)dot, th, BYPASS_CAPA, 1);
1355 rec->rec_fid = lfsck_dto2fid(parent);
1356 rc = dt_insert(env, child, (const struct dt_rec *)rec,
1357 (const struct dt_key *)dotdot, th,
1362 rc = dt_ref_add(env, child, th);
1367 rc = dt_xattr_set(env, child, &lmv_buf, XATTR_NAME_LMV, 0,
1373 rc = dt_xattr_set(env, child, &linkea_buf,
1374 XATTR_NAME_LINK, 0, th, BYPASS_CAPA);
1375 dt_write_unlock(env, child);
1379 rec->rec_fid = cfid;
1380 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1381 (const struct dt_key *)name, th, BYPASS_CAPA, 1);
1383 dt_write_lock(env, parent, 0);
1384 rc = dt_ref_add(env, parent, th);
1385 dt_write_unlock(env, parent);
1388 GOTO(stop, rc = (rc == 0 ? 1 : rc));
1391 dt_write_unlock(env, child);
1394 dt_trans_stop(env, dev, th);
1397 lfsck_ibits_unlock(&lh, LCK_EX);
1400 CDEBUG(D_LFSCK, "%s: namespace LFSCK create orphan dir for "
1401 "the object "DFID", name = %s: rc = %d\n",
1402 lfsck_lfsck2name(lfsck), PFID(cfid),
1403 cname->ln_name != NULL ? cname->ln_name : "<NULL>", rc);
1405 if (child != NULL && !IS_ERR(child))
1406 lfsck_object_put(env, child);
1408 if (parent != NULL && !IS_ERR(parent) && parent != lfsck->li_lpf_obj)
1409 lfsck_object_put(env, parent);
1412 ns->ln_flags |= LF_INCONSISTENT;
1418 * Remove the specified entry from the linkEA.
1420 * Locate the linkEA entry with the given @cname and @pfid, then
1421 * remove this entry or the other entries those are repeated with
1424 * \param[in] env pointer to the thread context
1425 * \param[in] com pointer to the lfsck component
1426 * \param[in] obj pointer to the dt_object to be handled
1427 * \param[in,out]ldata pointer to the buffer that holds the linkEA
1428 * \param[in] cname the name for the child in the parent directory
1429 * \param[in] pfid the parent directory's FID for the linkEA
1430 * \param[in] next if true, then remove the first found linkEA
1431 * entry, and move the ldata->ld_lee to next entry
1433 * \retval positive number for repaired cases
1434 * \retval 0 if nothing to be repaired
1435 * \retval negative error number on failure
1437 static int lfsck_namespace_shrink_linkea(const struct lu_env *env,
1438 struct lfsck_component *com,
1439 struct dt_object *obj,
1440 struct linkea_data *ldata,
1441 struct lu_name *cname,
1442 struct lu_fid *pfid,
1445 struct lfsck_instance *lfsck = com->lc_lfsck;
1446 struct dt_device *dev = lfsck->li_bottom;
1447 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
1448 struct thandle *th = NULL;
1449 struct lustre_handle lh = { 0 };
1450 struct linkea_data ldata_new = { NULL };
1451 struct lu_buf linkea_buf;
1455 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
1456 MDS_INODELOCK_UPDATE |
1457 MDS_INODELOCK_XATTR, LCK_EX);
1462 linkea_del_buf(ldata, cname);
1464 lfsck_namespace_filter_linkea_entry(ldata, cname, pfid,
1466 lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
1467 ldata->ld_leh->leh_len);
1470 th = dt_trans_create(env, dev);
1472 GOTO(unlock1, rc = PTR_ERR(th));
1474 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
1475 XATTR_NAME_LINK, 0, th);
1479 rc = dt_trans_start_local(env, dev, th);
1483 dt_write_lock(env, obj, 0);
1484 if (unlikely(lfsck_is_dead_obj(obj)))
1485 GOTO(unlock2, rc = -ENOENT);
1487 rc = lfsck_links_read2(env, obj, &ldata_new);
1491 /* The specified linkEA entry has been removed by race. */
1492 rc = linkea_links_find(&ldata_new, cname, pfid);
1494 GOTO(unlock2, rc = 0);
1496 if (bk->lb_param & LPF_DRYRUN)
1497 GOTO(unlock2, rc = 1);
1500 linkea_del_buf(&ldata_new, cname);
1502 lfsck_namespace_filter_linkea_entry(&ldata_new, cname, pfid,
1505 if (linkea_buf.lb_len < ldata_new.ld_leh->leh_len) {
1506 dt_write_unlock(env, obj);
1507 dt_trans_stop(env, dev, th);
1508 lfsck_buf_init(&linkea_buf, ldata_new.ld_buf->lb_buf,
1509 ldata_new.ld_leh->leh_len);
1513 lfsck_buf_init(&linkea_buf, ldata_new.ld_buf->lb_buf,
1514 ldata_new.ld_leh->leh_len);
1515 rc = dt_xattr_set(env, obj, &linkea_buf,
1516 XATTR_NAME_LINK, 0, th, BYPASS_CAPA);
1518 GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
1521 dt_write_unlock(env, obj);
1524 dt_trans_stop(env, dev, th);
1527 lfsck_ibits_unlock(&lh, LCK_EX);
1530 CDEBUG(D_LFSCK, "%s: namespace LFSCK remove %s linkEA entry "
1531 "for the object: "DFID", parent "DFID", name %.*s\n",
1532 lfsck_lfsck2name(lfsck), next ? "invalid" : "redundant",
1533 PFID(lfsck_dto2fid(obj)), PFID(pfid), cname->ln_namelen,
1537 struct lfsck_namespace *ns = com->lc_file_ram;
1539 ns->ln_flags |= LF_INCONSISTENT;
1546 * Conditionally remove the specified entry from the linkEA.
1548 * Take the parent lock firstly, then check whether the specified
1549 * name entry exists or not: if yes, do nothing; otherwise, call
1550 * lfsck_namespace_shrink_linkea() to remove the linkea entry.
1552 * \param[in] env pointer to the thread context
1553 * \param[in] com pointer to the lfsck component
1554 * \param[in] parent pointer to the parent directory
1555 * \param[in] child pointer to the child object that holds the linkEA
1556 * \param[in,out]ldata pointer to the buffer that holds the linkEA
1557 * \param[in] cname the name for the child in the parent directory
1558 * \param[in] pfid the parent directory's FID for the linkEA
1560 * \retval positive number for repaired cases
1561 * \retval 0 if nothing to be repaired
1562 * \retval negative error number on failure
1564 static int lfsck_namespace_shrink_linkea_cond(const struct lu_env *env,
1565 struct lfsck_component *com,
1566 struct dt_object *parent,
1567 struct dt_object *child,
1568 struct linkea_data *ldata,
1569 struct lu_name *cname,
1570 struct lu_fid *pfid)
1572 struct lu_fid *cfid = &lfsck_env_info(env)->lti_fid3;
1573 struct lustre_handle lh = { 0 };
1577 rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
1578 MDS_INODELOCK_UPDATE, LCK_EX);
1582 dt_read_lock(env, parent, 0);
1583 if (unlikely(lfsck_is_dead_obj(parent))) {
1584 dt_read_unlock(env, parent);
1585 lfsck_ibits_unlock(&lh, LCK_EX);
1586 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
1592 rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
1593 (const struct dt_key *)cname->ln_name,
1595 dt_read_unlock(env, parent);
1597 /* It is safe to release the ldlm lock, because when the logic come
1598 * here, we have got all the needed information above whether the
1599 * linkEA entry is valid or not. It is not important that others
1600 * may add new linkEA entry after the ldlm lock released. If other
1601 * has removed the specified linkEA entry by race, then it is OK,
1602 * because the subsequent lfsck_namespace_shrink_linkea() can handle
1604 lfsck_ibits_unlock(&lh, LCK_EX);
1605 if (rc == -ENOENT) {
1606 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
1615 /* The LFSCK just found some internal status of cross-MDTs
1616 * create operation. That is normal. */
1617 if (lu_fid_eq(cfid, lfsck_dto2fid(child))) {
1618 linkea_next_entry(ldata);
1623 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata, cname,
1630 * Conditionally replace name entry in the parent.
1632 * As required, the LFSCK may re-create the lost MDT-object for dangling
1633 * name entry, but such repairing may be wrong because of bad FID in the
1634 * name entry. As the LFSCK processing, the real MDT-object may be found,
1635 * then the LFSCK should check whether the former re-created MDT-object
1636 * has been modified or not, if not, then destroy it and update the name
1637 * entry in the parent to reference the real MDT-object.
1639 * \param[in] env pointer to the thread context
1640 * \param[in] com pointer to the lfsck component
1641 * \param[in] parent pointer to the parent directory
1642 * \param[in] child pointer to the MDT-object that may be the real
1643 * MDT-object corresponding to the name entry in parent
1644 * \param[in] cfid the current FID in the name entry
1645 * \param[in] cname contains the name of the child in the parent directory
1647 * \retval positive number for repaired cases
1648 * \retval 0 if nothing to be repaired
1649 * \retval negative error number on failure
1651 static int lfsck_namespace_replace_cond(const struct lu_env *env,
1652 struct lfsck_component *com,
1653 struct dt_object *parent,
1654 struct dt_object *child,
1655 const struct lu_fid *cfid,
1656 const struct lu_name *cname)
1658 struct lfsck_thread_info *info = lfsck_env_info(env);
1659 struct lu_attr *la = &info->lti_la;
1660 struct dt_insert_rec *rec = &info->lti_dt_rec;
1662 struct lfsck_instance *lfsck = com->lc_lfsck;
1663 struct dt_device *dev = lfsck->li_next;
1664 const char *name = cname->ln_name;
1665 struct dt_object *obj = NULL;
1666 struct lustre_handle plh = { 0 };
1667 struct lustre_handle clh = { 0 };
1668 struct linkea_data ldata = { NULL };
1669 struct thandle *th = NULL;
1674 rc = lfsck_ibits_lock(env, lfsck, parent, &plh,
1675 MDS_INODELOCK_UPDATE, LCK_EX);
1679 if (!fid_is_sane(cfid)) {
1684 obj = lfsck_object_find(env, lfsck, cfid);
1687 if (rc == -ENOENT) {
1695 if (!dt_object_exists(obj)) {
1700 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
1701 (const struct dt_key *)name, BYPASS_CAPA);
1702 if (rc == -ENOENT) {
1710 /* Someone changed the name entry, cannot replace it. */
1711 if (!lu_fid_eq(cfid, &tfid))
1714 /* lock the object to be destroyed. */
1715 rc = lfsck_ibits_lock(env, lfsck, obj, &clh,
1716 MDS_INODELOCK_UPDATE |
1717 MDS_INODELOCK_XATTR, LCK_EX);
1721 if (unlikely(lfsck_is_dead_obj(obj))) {
1726 rc = dt_attr_get(env, obj, la, BYPASS_CAPA);
1730 /* The object has been modified by other(s), or it is not created by
1731 * LFSCK, the two cases are indistinguishable. So cannot replace it. */
1732 if (la->la_ctime != 0)
1735 if (S_ISREG(la->la_mode)) {
1736 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LOV,
1738 /* If someone has created related OST-object(s),
1740 if ((rc > 0) || (rc < 0 && rc != -ENODATA))
1741 GOTO(log, rc = (rc > 0 ? 0 : rc));
1745 dt_read_lock(env, child, 0);
1746 rc = lfsck_links_read2(env, child, &ldata);
1747 dt_read_unlock(env, child);
1749 /* Someone changed the child, no need to replace. */
1756 rc = linkea_links_find(&ldata, cname, lfsck_dto2fid(parent));
1757 /* Someone moved the child, no need to replace. */
1761 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1764 th = dt_trans_create(env, dev);
1766 GOTO(log, rc = PTR_ERR(th));
1769 rc = dt_declare_destroy(env, obj, th);
1774 rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
1778 rec->rec_type = S_IFDIR;
1779 rec->rec_fid = lfsck_dto2fid(child);
1780 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1781 (const struct dt_key *)name, th);
1785 rc = dt_trans_start(env, dev, th);
1790 rc = dt_destroy(env, obj, th);
1795 /* The old name entry maybe not exist. */
1796 dt_delete(env, parent, (const struct dt_key *)name, th,
1799 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1800 (const struct dt_key *)name, th, BYPASS_CAPA, 1);
1802 GOTO(stop, rc = (rc == 0 ? 1 : rc));
1805 dt_trans_stop(env, dev, th);
1808 lfsck_ibits_unlock(&clh, LCK_EX);
1809 lfsck_ibits_unlock(&plh, LCK_EX);
1810 if (obj != NULL && !IS_ERR(obj))
1811 lfsck_object_put(env, obj);
1813 CDEBUG(D_LFSCK, "%s: namespace LFSCK conditionally destroy the "
1814 "object "DFID" because of conflict with the object "DFID
1815 " under the parent "DFID" with name %s: rc = %d\n",
1816 lfsck_lfsck2name(lfsck), PFID(cfid),
1817 PFID(lfsck_dto2fid(child)), PFID(lfsck_dto2fid(parent)),
1824 * Overwrite the linkEA for the object with the given ldata.
1826 * The caller should take the ldlm lock before the calling.
1828 * \param[in] env pointer to the thread context
1829 * \param[in] com pointer to the lfsck component
1830 * \param[in] obj pointer to the dt_object to be handled
1831 * \param[in] ldata pointer to the new linkEA data
1833 * \retval positive number for repaired cases
1834 * \retval 0 if nothing to be repaired
1835 * \retval negative error number on failure
1837 int lfsck_namespace_rebuild_linkea(const struct lu_env *env,
1838 struct lfsck_component *com,
1839 struct dt_object *obj,
1840 struct linkea_data *ldata)
1842 struct lfsck_instance *lfsck = com->lc_lfsck;
1843 struct dt_device *dev = lfsck->li_bottom;
1844 struct thandle *th = NULL;
1845 struct lu_buf linkea_buf;
1849 th = dt_trans_create(env, dev);
1851 GOTO(log, rc = PTR_ERR(th));
1853 lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
1854 ldata->ld_leh->leh_len);
1855 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
1856 XATTR_NAME_LINK, 0, th);
1860 rc = dt_trans_start_local(env, dev, th);
1864 dt_write_lock(env, obj, 0);
1865 if (unlikely(lfsck_is_dead_obj(obj)))
1866 GOTO(unlock, rc = 0);
1868 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1869 GOTO(unlock, rc = 1);
1871 rc = dt_xattr_set(env, obj, &linkea_buf,
1872 XATTR_NAME_LINK, 0, th, BYPASS_CAPA);
1874 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
1877 dt_write_unlock(env, obj);
1880 dt_trans_stop(env, dev, th);
1883 CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild linkEA for the "
1884 "object "DFID": rc = %d\n",
1885 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
1888 struct lfsck_namespace *ns = com->lc_file_ram;
1890 ns->ln_flags |= LF_INCONSISTENT;
1897 * Repair invalid name entry.
1899 * If the name entry contains invalid information, such as bad file type
1900 * or (and) corrupted object FID, then either remove the name entry or
1901 * udpate the name entry with the given (right) information.
1903 * \param[in] env pointer to the thread context
1904 * \param[in] com pointer to the lfsck component
1905 * \param[in] parent pointer to the parent directory
1906 * \param[in] child pointer to the object referenced by the name entry
1907 * \param[in] name the old name of the child under the parent directory
1908 * \param[in] name2 the new name of the child under the parent directory
1909 * \param[in] type the type claimed by the name entry
1910 * \param[in] update update the name entry if true; otherwise, remove it
1911 * \param[in] dec decrease the parent nlink count if true
1913 * \retval positive number for repaired successfully
1914 * \retval 0 if nothing to be repaired
1915 * \retval negative error number on failure
1917 int lfsck_namespace_repair_dirent(const struct lu_env *env,
1918 struct lfsck_component *com,
1919 struct dt_object *parent,
1920 struct dt_object *child,
1921 const char *name, const char *name2,
1922 __u16 type, bool update, bool dec)
1924 struct dt_insert_rec *rec = &lfsck_env_info(env)->lti_dt_rec;
1925 const struct lu_fid *cfid = lfsck_dto2fid(child);
1927 struct lfsck_instance *lfsck = com->lc_lfsck;
1928 struct dt_device *dev = lfsck->li_next;
1929 struct thandle *th = NULL;
1930 struct lustre_handle lh = { 0 };
1934 if (unlikely(!dt_try_as_dir(env, parent)))
1935 GOTO(log, rc = -ENOTDIR);
1937 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
1938 MDS_INODELOCK_UPDATE, LCK_EX);
1942 th = dt_trans_create(env, dev);
1944 GOTO(unlock1, rc = PTR_ERR(th));
1946 rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
1951 rec->rec_type = lfsck_object_type(child) & S_IFMT;
1952 rec->rec_fid = cfid;
1953 rc = dt_declare_insert(env, parent,
1954 (const struct dt_rec *)rec,
1955 (const struct dt_key *)name2, th);
1961 rc = dt_declare_ref_del(env, parent, th);
1966 rc = dt_trans_start(env, dev, th);
1970 dt_write_lock(env, parent, 0);
1971 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
1972 (const struct dt_key *)name, BYPASS_CAPA);
1973 /* Someone has removed the bad name entry by race. */
1975 GOTO(unlock2, rc = 0);
1980 /* Someone has removed the bad name entry and reused it for other
1981 * object by race. */
1982 if (!lu_fid_eq(&tfid, cfid))
1983 GOTO(unlock2, rc = 0);
1985 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1986 GOTO(unlock2, rc = 1);
1988 rc = dt_delete(env, parent, (const struct dt_key *)name, th,
1994 rc = dt_insert(env, parent,
1995 (const struct dt_rec *)rec,
1996 (const struct dt_key *)name2, th,
2003 rc = dt_ref_del(env, parent, th);
2008 GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
2011 dt_write_unlock(env, parent);
2014 dt_trans_stop(env, dev, th);
2016 /* We are not sure whether the child will become orphan or not.
2017 * Record it in the LFSCK trace file for further checking in
2018 * the second-stage scanning. */
2019 if (!update && !dec && rc == 0)
2020 lfsck_namespace_trace_update(env, com, cfid,
2021 LNTF_CHECK_LINKEA, true);
2024 lfsck_ibits_unlock(&lh, LCK_EX);
2027 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found bad name "
2028 "entry for: parent "DFID", child "DFID", name %s, type "
2029 "in name entry %o, type claimed by child %o. repair it "
2030 "by %s with new name2 %s: rc = %d\n", lfsck_lfsck2name(lfsck),
2031 PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)),
2032 name, type, update ? lfsck_object_type(child) : 0,
2033 update ? "updating" : "removing", name2, rc);
2036 struct lfsck_namespace *ns = com->lc_file_ram;
2038 ns->ln_flags |= LF_INCONSISTENT;
2045 * Update the ".." name entry for the given object.
2047 * The object's ".." is corrupted, this function will update the ".." name
2048 * entry with the given pfid, and the linkEA with the given ldata.
2050 * The caller should take the ldlm lock before the calling.
2052 * \param[in] env pointer to the thread context
2053 * \param[in] com pointer to the lfsck component
2054 * \param[in] obj pointer to the dt_object to be handled
2055 * \param[in] pfid the new fid for the object's ".." name entry
2056 * \param[in] cname the name for the @obj in the parent directory
2058 * \retval positive number for repaired cases
2059 * \retval 0 if nothing to be repaired
2060 * \retval negative error number on failure
2062 static int lfsck_namespace_repair_unmatched_pairs(const struct lu_env *env,
2063 struct lfsck_component *com,
2064 struct dt_object *obj,
2065 const struct lu_fid *pfid,
2066 struct lu_name *cname)
2068 struct lfsck_thread_info *info = lfsck_env_info(env);
2069 struct dt_insert_rec *rec = &info->lti_dt_rec;
2070 struct lfsck_instance *lfsck = com->lc_lfsck;
2071 struct dt_device *dev = lfsck->li_bottom;
2072 struct thandle *th = NULL;
2073 struct linkea_data ldata = { NULL };
2074 struct lu_buf linkea_buf;
2078 LASSERT(!dt_object_remote(obj));
2079 LASSERT(S_ISDIR(lfsck_object_type(obj)));
2081 rc = linkea_data_new(&ldata, &info->lti_big_buf);
2085 rc = linkea_add_buf(&ldata, cname, pfid);
2089 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
2090 ldata.ld_leh->leh_len);
2092 th = dt_trans_create(env, dev);
2094 GOTO(log, rc = PTR_ERR(th));
2096 rc = dt_declare_delete(env, obj, (const struct dt_key *)dotdot, th);
2100 rec->rec_type = S_IFDIR;
2101 rec->rec_fid = pfid;
2102 rc = dt_declare_insert(env, obj, (const struct dt_rec *)rec,
2103 (const struct dt_key *)dotdot, th);
2107 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
2108 XATTR_NAME_LINK, 0, th);
2112 rc = dt_trans_start_local(env, dev, th);
2116 dt_write_lock(env, obj, 0);
2117 if (unlikely(lfsck_is_dead_obj(obj)))
2118 GOTO(unlock, rc = 0);
2120 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2121 GOTO(unlock, rc = 1);
2123 /* The old ".." name entry maybe not exist. */
2124 dt_delete(env, obj, (const struct dt_key *)dotdot, th,
2127 rc = dt_insert(env, obj, (const struct dt_rec *)rec,
2128 (const struct dt_key *)dotdot, th, BYPASS_CAPA, 1);
2132 rc = dt_xattr_set(env, obj, &linkea_buf,
2133 XATTR_NAME_LINK, 0, th, BYPASS_CAPA);
2135 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2138 dt_write_unlock(env, obj);
2141 dt_trans_stop(env, dev, th);
2144 CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild dotdot name entry for "
2145 "the object "DFID", new parent "DFID": rc = %d\n",
2146 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)),
2150 struct lfsck_namespace *ns = com->lc_file_ram;
2152 ns->ln_flags |= LF_INCONSISTENT;
2159 * Handle orphan @obj during Double Scan Directory.
2161 * Remove the @obj's current (invalid) linkEA entries, and insert
2162 * it in the directory .lustre/lost+found/MDTxxxx/ with the name:
2163 * ${FID}-${PFID}-D-${conflict_version}
2165 * The caller should take the ldlm lock before the calling.
2167 * \param[in] env pointer to the thread context
2168 * \param[in] com pointer to the lfsck component
2169 * \param[in] obj pointer to the orphan object to be handled
2170 * \param[in] pfid the new fid for the object's ".." name entry
2171 * \param[in,out] lh ldlm lock handler for the given @obj
2172 * \param[out] type to tell the caller what the inconsistency is
2174 * \retval positive number for repaired cases
2175 * \retval 0 if nothing to be repaired
2176 * \retval negative error number on failure
2179 lfsck_namespace_dsd_orphan(const struct lu_env *env,
2180 struct lfsck_component *com,
2181 struct dt_object *obj,
2182 const struct lu_fid *pfid,
2183 struct lustre_handle *lh,
2184 enum lfsck_namespace_inconsistency_type *type)
2186 struct lfsck_thread_info *info = lfsck_env_info(env);
2187 struct lfsck_namespace *ns = com->lc_file_ram;
2191 /* Remove the unrecognized linkEA. */
2192 rc = lfsck_namespace_links_remove(env, com, obj);
2193 lfsck_ibits_unlock(lh, LCK_EX);
2194 if (rc < 0 && rc != -ENODATA)
2197 *type = LNIT_MUL_REF;
2199 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
2200 * ever tried to verify some remote MDT-object that resides on this
2201 * MDT, but this MDT failed to respond such request. So means there
2202 * may be some remote name entry on other MDT that references this
2203 * object with another name, so we cannot know whether this linkEA
2204 * is valid or not. So keep it there and maybe resolved when next
2206 if (ns->ln_flags & LF_INCOMPLETE)
2209 /* The unique linkEA is invalid, even if the ".." name entry may be
2210 * valid, we still cannot know via which name entry this directory
2211 * will be referenced. Then handle it as pure orphan. */
2212 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2213 "-"DFID, PFID(pfid));
2214 rc = lfsck_namespace_insert_orphan(env, com, obj,
2215 info->lti_tmpbuf, "D", NULL);
2221 * Double Scan Directory object for single linkEA entry case.
2223 * The given @child has unique linkEA entry. If the linkEA entry is valid,
2224 * then check whether the name is in the namespace or not, if not, add the
2225 * missing name entry back to namespace. If the linkEA entry is invalid,
2226 * then remove it and insert the @child in the .lustre/lost+found/MDTxxxx/
2229 * \param[in] env pointer to the thread context
2230 * \param[in] com pointer to the lfsck component
2231 * \param[in] child pointer to the directory to be double scanned
2232 * \param[in] pfid the FID corresponding to the ".." entry
2233 * \param[in] ldata pointer to the linkEA data for the given @child
2234 * \param[in,out] lh ldlm lock handler for the given @child
2235 * \param[out] type to tell the caller what the inconsistency is
2236 * \param[in] retry if found inconsistency, but the caller does not hold
2237 * ldlm lock on the @child, then set @retry as true
2239 * \retval positive number for repaired cases
2240 * \retval 0 if nothing to be repaired
2241 * \retval negative error number on failure
2244 lfsck_namespace_dsd_single(const struct lu_env *env,
2245 struct lfsck_component *com,
2246 struct dt_object *child,
2247 const struct lu_fid *pfid,
2248 struct linkea_data *ldata,
2249 struct lustre_handle *lh,
2250 enum lfsck_namespace_inconsistency_type *type,
2253 struct lfsck_thread_info *info = lfsck_env_info(env);
2254 struct lu_name *cname = &info->lti_name;
2255 const struct lu_fid *cfid = lfsck_dto2fid(child);
2257 struct lfsck_namespace *ns = com->lc_file_ram;
2258 struct lfsck_instance *lfsck = com->lc_lfsck;
2259 struct dt_object *parent = NULL;
2260 struct lmv_mds_md_v1 *lmv;
2264 lfsck_namespace_unpack_linkea_entry(ldata, cname, &tfid, info->lti_key);
2265 /* The unique linkEA entry with bad parent will be handled as orphan. */
2266 if (!fid_is_sane(&tfid)) {
2267 if (!lustre_handle_is_used(lh) && retry != NULL)
2270 rc = lfsck_namespace_dsd_orphan(env, com, child,
2276 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
2278 GOTO(out, rc = PTR_ERR(parent));
2280 /* We trust the unique linkEA entry in spite of whether it matches the
2281 * ".." name entry or not. Because even if the linkEA entry is wrong
2282 * and the ".." name entry is right, we still cannot know via which
2283 * name entry the child will be referenced, since all known entries
2284 * have been verified during the first-stage scanning. */
2285 if (!dt_object_exists(parent)) {
2286 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2287 * has ever tried to verify some remote MDT-object that resides
2288 * on this MDT, but this MDT failed to respond such request. So
2289 * means there may be some remote name entry on other MDT that
2290 * references this object with another name, so we cannot know
2291 * whether this linkEA is valid or not. So keep it there and
2292 * maybe resolved when next LFSCK run. */
2293 if (ns->ln_flags & LF_INCOMPLETE)
2296 if (!lustre_handle_is_used(lh) && retry != NULL) {
2302 lfsck_ibits_unlock(lh, LCK_EX);
2305 lmv = &info->lti_lmv;
2306 rc = lfsck_read_stripe_lmv(env, child, lmv);
2307 if (rc != 0 && rc != -ENODATA)
2310 if (rc == -ENODATA || lmv->lmv_magic != LMV_MAGIC_STRIPE) {
2312 } else if (lfsck_shard_name_to_index(env,
2313 cname->ln_name, cname->ln_namelen,
2314 S_IFDIR, cfid) < 0) {
2315 /* It is an invalid name entry, we
2316 * cannot trust the parent also. */
2317 rc = lfsck_namespace_shrink_linkea(env, com, child,
2318 ldata, cname, &tfid, true);
2322 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2323 "-"DFID, PFID(pfid));
2324 rc = lfsck_namespace_insert_orphan(env, com, child,
2325 info->lti_tmpbuf, "S", NULL);
2330 /* Create the lost parent as an orphan. */
2331 rc = lfsck_namespace_create_orphan_dir(env, com, parent, lmv);
2333 /* Add the missing name entry to the parent. */
2334 rc = lfsck_namespace_insert_normal(env, com, parent,
2335 child, cname->ln_name);
2336 if (unlikely(rc == -EEXIST)) {
2337 /* Unfortunately, someone reused the name
2338 * under the parent by race. So we have
2339 * to remove the linkEA entry from
2340 * current child object. It means that the
2341 * LFSCK cannot recover the system
2342 * totally back to its original status,
2343 * but it is necessary to make the
2344 * current system to be consistent. */
2345 rc = lfsck_namespace_shrink_linkea(env,
2347 cname, &tfid, true);
2349 snprintf(info->lti_tmpbuf,
2350 sizeof(info->lti_tmpbuf),
2351 "-"DFID, PFID(pfid));
2352 rc = lfsck_namespace_insert_orphan(env,
2353 com, child, info->lti_tmpbuf,
2362 /* The unique linkEA entry with bad parent will be handled as orphan. */
2363 if (unlikely(!dt_try_as_dir(env, parent))) {
2364 if (!lustre_handle_is_used(lh) && retry != NULL)
2367 rc = lfsck_namespace_dsd_orphan(env, com, child,
2373 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
2374 (const struct dt_key *)cname->ln_name, BYPASS_CAPA);
2375 if (rc == -ENOENT) {
2376 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2377 * has ever tried to verify some remote MDT-object that resides
2378 * on this MDT, but this MDT failed to respond such request. So
2379 * means there may be some remote name entry on other MDT that
2380 * references this object with another name, so we cannot know
2381 * whether this linkEA is valid or not. So keep it there and
2382 * maybe resolved when next LFSCK run. */
2383 if (ns->ln_flags & LF_INCOMPLETE)
2386 if (!lustre_handle_is_used(lh) && retry != NULL) {
2392 lfsck_ibits_unlock(lh, LCK_EX);
2393 rc = lfsck_namespace_check_name(env, parent, child, cname);
2400 /* It is an invalid name entry, drop it. */
2401 if (unlikely(rc > 0)) {
2402 rc = lfsck_namespace_shrink_linkea(env, com, child,
2403 ldata, cname, &tfid, true);
2405 snprintf(info->lti_tmpbuf,
2406 sizeof(info->lti_tmpbuf),
2407 "-"DFID, PFID(pfid));
2408 rc = lfsck_namespace_insert_orphan(env, com,
2409 child, info->lti_tmpbuf, "D", NULL);
2415 /* Add the missing name entry back to the namespace. */
2416 rc = lfsck_namespace_insert_normal(env, com, parent, child,
2418 if (unlikely(rc == -ESTALE))
2419 /* It may happen when the remote object has been
2420 * removed, but the local MDT is not aware of that. */
2423 if (unlikely(rc == -EEXIST)) {
2424 /* Unfortunately, someone reused the name under the
2425 * parent by race. So we have to remove the linkEA
2426 * entry from current child object. It means that the
2427 * LFSCK cannot recover the system totally back to
2428 * its original status, but it is necessary to make
2429 * the current system to be consistent.
2431 * It also may be because of the LFSCK found some
2432 * internal status of create operation. Under such
2433 * case, nothing to be done. */
2434 rc = lfsck_namespace_shrink_linkea_cond(env, com,
2435 parent, child, ldata, cname, &tfid);
2437 snprintf(info->lti_tmpbuf,
2438 sizeof(info->lti_tmpbuf),
2439 "-"DFID, PFID(pfid));
2440 rc = lfsck_namespace_insert_orphan(env, com,
2441 child, info->lti_tmpbuf, "D", NULL);
2451 if (!lu_fid_eq(&tfid, cfid)) {
2452 if (!lustre_handle_is_used(lh) && retry != NULL) {
2458 lfsck_ibits_unlock(lh, LCK_EX);
2459 /* The name entry references another MDT-object that
2460 * may be created by the LFSCK for repairing dangling
2461 * name entry. Try to replace it. */
2462 rc = lfsck_namespace_replace_cond(env, com, parent, child,
2465 rc = lfsck_namespace_dsd_orphan(env, com, child,
2471 if (fid_is_zero(pfid))
2474 /* The ".." name entry is wrong, update it. */
2475 if (!lu_fid_eq(pfid, lfsck_dto2fid(parent))) {
2476 if (!lustre_handle_is_used(lh) && retry != NULL) {
2482 *type = LNIT_UNMATCHED_PAIRS;
2483 rc = lfsck_namespace_repair_unmatched_pairs(env, com, child,
2484 lfsck_dto2fid(parent), cname);
2490 if (parent != NULL && !IS_ERR(parent))
2491 lfsck_object_put(env, parent);
2497 * Double Scan Directory object for multiple linkEA entries case.
2499 * The given @child has multiple linkEA entries. There is at most one linkEA
2500 * entry will be valid, all the others will be removed. Firstly, the function
2501 * will try to find out the linkEA entry for which the name entry exists under
2502 * the given parent (@pfid). If there is no linkEA entry that matches the given
2503 * ".." name entry, then tries to find out the first linkEA entry that both the
2504 * parent and the name entry exist to rebuild a new ".." name entry.
2506 * \param[in] env pointer to the thread context
2507 * \param[in] com pointer to the lfsck component
2508 * \param[in] child pointer to the directory to be double scanned
2509 * \param[in] pfid the FID corresponding to the ".." entry
2510 * \param[in] ldata pointer to the linkEA data for the given @child
2511 * \param[in,out] lh ldlm lock handler for the given @child
2512 * \param[out] type to tell the caller what the inconsistency is
2513 * \param[in] lpf true if the ".." entry is under lost+found/MDTxxxx/
2515 * \retval positive number for repaired cases
2516 * \retval 0 if nothing to be repaired
2517 * \retval negative error number on failure
2520 lfsck_namespace_dsd_multiple(const struct lu_env *env,
2521 struct lfsck_component *com,
2522 struct dt_object *child,
2523 const struct lu_fid *pfid,
2524 struct linkea_data *ldata,
2525 struct lustre_handle *lh,
2526 enum lfsck_namespace_inconsistency_type *type,
2529 struct lfsck_thread_info *info = lfsck_env_info(env);
2530 struct lu_name *cname = &info->lti_name;
2531 const struct lu_fid *cfid = lfsck_dto2fid(child);
2532 struct lu_fid *pfid2 = &info->lti_fid3;
2534 struct lfsck_namespace *ns = com->lc_file_ram;
2535 struct lfsck_instance *lfsck = com->lc_lfsck;
2536 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
2537 struct dt_object *parent = NULL;
2538 struct linkea_data ldata_new = { NULL };
2539 int dirent_count = 0;
2540 int linkea_count = 0;
2546 while (ldata->ld_lee != NULL) {
2547 lfsck_namespace_unpack_linkea_entry(ldata, cname, &tfid,
2549 /* Drop repeated linkEA entries. */
2550 lfsck_namespace_filter_linkea_entry(ldata, cname, &tfid, true);
2551 /* Drop invalid linkEA entry. */
2552 if (!fid_is_sane(&tfid)) {
2553 linkea_del_buf(ldata, cname);
2558 /* If current dotdot is the .lustre/lost+found/MDTxxxx/,
2559 * then it is possible that: the directry object has ever
2560 * been lost, but its name entry was there. In the former
2561 * LFSCK run, during the first-stage scanning, the LFSCK
2562 * found the dangling name entry, but it did not recreate
2563 * the lost object, and when moved to the second-stage
2564 * scanning, some children objects of the lost directory
2565 * object were found, then the LFSCK recreated such lost
2566 * directory object as an orphan.
2568 * When the LFSCK runs again, if the dangling name is still
2569 * there, the LFSCK should move the orphan directory object
2570 * back to the normal namespace. */
2571 if (!lpf && !lu_fid_eq(pfid, &tfid) && once) {
2572 linkea_next_entry(ldata);
2576 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
2578 RETURN(PTR_ERR(parent));
2580 if (!dt_object_exists(parent)) {
2581 lfsck_object_put(env, parent);
2582 if (ldata->ld_leh->leh_reccount > 1) {
2583 /* If it is NOT the last linkEA entry, then
2584 * there is still other chance to make the
2585 * child to be visible via other parent, then
2586 * remove this linkEA entry. */
2587 linkea_del_buf(ldata, cname);
2595 /* The linkEA entry with bad parent will be removed. */
2596 if (unlikely(!dt_try_as_dir(env, parent))) {
2597 lfsck_object_put(env, parent);
2598 linkea_del_buf(ldata, cname);
2603 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
2604 (const struct dt_key *)cname->ln_name,
2606 *pfid2 = *lfsck_dto2fid(parent);
2607 if (rc == -ENOENT) {
2608 lfsck_object_put(env, parent);
2609 linkea_next_entry(ldata);
2614 lfsck_object_put(env, parent);
2619 if (lu_fid_eq(&tfid, cfid)) {
2620 lfsck_object_put(env, parent);
2621 if (!lu_fid_eq(pfid, pfid2)) {
2622 *type = LNIT_UNMATCHED_PAIRS;
2623 rc = lfsck_namespace_repair_unmatched_pairs(env,
2624 com, child, pfid2, cname);
2630 /* It is the most common case that we find the
2631 * name entry corresponding to the linkEA entry
2632 * that matches the ".." name entry. */
2633 rc = linkea_data_new(&ldata_new, &info->lti_big_buf);
2637 rc = linkea_add_buf(&ldata_new, cname, pfid2);
2641 rc = lfsck_namespace_rebuild_linkea(env, com, child,
2646 linkea_del_buf(ldata, cname);
2648 linkea_first_entry(ldata);
2649 /* There may be some invalid dangling name entries under
2650 * other parent directories, remove all of them. */
2651 while (ldata->ld_lee != NULL) {
2652 lfsck_namespace_unpack_linkea_entry(ldata,
2653 cname, &tfid, info->lti_key);
2654 if (!fid_is_sane(&tfid))
2657 parent = lfsck_object_find_bottom(env, lfsck,
2659 if (IS_ERR(parent)) {
2660 rc = PTR_ERR(parent);
2661 if (rc != -ENOENT &&
2662 bk->lb_param & LPF_FAILOUT)
2668 if (!dt_object_exists(parent)) {
2669 lfsck_object_put(env, parent);
2673 rc = lfsck_namespace_repair_dirent(env, com,
2674 parent, child, cname->ln_name,
2675 cname->ln_name, S_IFDIR, false, true);
2676 lfsck_object_put(env, parent);
2678 if (bk->lb_param & LPF_FAILOUT)
2687 linkea_del_buf(ldata, cname);
2690 ns->ln_dirent_repaired += dirent_count;
2695 lfsck_ibits_unlock(lh, LCK_EX);
2696 /* The name entry references another MDT-object that may be
2697 * created by the LFSCK for repairing dangling name entry.
2698 * Try to replace it. */
2699 rc = lfsck_namespace_replace_cond(env, com, parent, child,
2701 lfsck_object_put(env, parent);
2708 linkea_del_buf(ldata, cname);
2711 linkea_first_entry(ldata);
2712 if (ldata->ld_leh->leh_reccount == 1) {
2713 rc = lfsck_namespace_dsd_single(env, com, child, pfid, ldata,
2716 if (rc == 0 && fid_is_zero(pfid) && linkea_count > 0)
2717 rc = lfsck_namespace_rebuild_linkea(env, com, child,
2723 /* All linkEA entries are invalid and removed, then handle the @child
2725 if (ldata->ld_leh->leh_reccount == 0) {
2726 rc = lfsck_namespace_dsd_orphan(env, com, child, pfid, lh,
2732 /* If the dangling name entry for the orphan directory object has
2733 * been remvoed, then just check whether the directory object is
2734 * still under the .lustre/lost+found/MDTxxxx/ or not. */
2740 /* There is no linkEA entry that matches the ".." name entry. Find
2741 * the first linkEA entry that both parent and name entry exist to
2742 * rebuild a new ".." name entry. */
2752 * Repair the object's nlink attribute.
2754 * If all the known name entries have been verified, then the object's hard
2755 * link attribute should match the object's linkEA entries count unless the
2756 * object's has too much hard link to be recorded in the linkEA. Such cases
2757 * should have been marked in the LFSCK trace file. Otherwise, trust the
2758 * linkEA to update the object's nlink attribute.
2760 * \param[in] env pointer to the thread context
2761 * \param[in] com pointer to the lfsck component
2762 * \param[in] obj pointer to the dt_object to be handled
2763 * \param[in,out] la pointer to buffer to object's attribute before
2764 * and after the repairing
2766 * \retval positive number for repaired cases
2767 * \retval 0 if nothing to be repaired
2768 * \retval negative error number on failure
2770 static int lfsck_namespace_repair_nlink(const struct lu_env *env,
2771 struct lfsck_component *com,
2772 struct dt_object *obj,
2775 struct lfsck_thread_info *info = lfsck_env_info(env);
2776 struct lu_fid *tfid = &info->lti_fid3;
2777 struct lfsck_namespace *ns = com->lc_file_ram;
2778 struct lfsck_instance *lfsck = com->lc_lfsck;
2779 struct dt_device *dev = lfsck->li_bottom;
2780 const struct lu_fid *cfid = lfsck_dto2fid(obj);
2781 struct dt_object *child = NULL;
2782 struct thandle *th = NULL;
2783 struct linkea_data ldata = { NULL };
2784 struct lustre_handle lh = { 0 };
2785 __u32 old = la->la_nlink;
2790 LASSERT(!dt_object_remote(obj));
2791 LASSERT(S_ISREG(lfsck_object_type(obj)));
2793 child = lfsck_object_find_by_dev(env, dev, cfid);
2795 GOTO(log, rc = PTR_ERR(child));
2797 rc = lfsck_ibits_lock(env, lfsck, child, &lh,
2798 MDS_INODELOCK_UPDATE |
2799 MDS_INODELOCK_XATTR, LCK_EX);
2803 th = dt_trans_create(env, dev);
2805 GOTO(log, rc = PTR_ERR(th));
2807 la->la_valid = LA_NLINK;
2808 rc = dt_declare_attr_set(env, child, la, th);
2812 rc = dt_trans_start_local(env, dev, th);
2816 dt_write_lock(env, child, 0);
2817 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
2818 * ever tried to verify some remote MDT-object that resides on this
2819 * MDT, but this MDT failed to respond such request. So means there
2820 * may be some remote name entry on other MDT that references this
2821 * object with another name, so we cannot know whether this linkEA
2822 * is valid or not. So keep it there and maybe resolved when next
2824 if (ns->ln_flags & LF_INCOMPLETE)
2825 GOTO(unlock, rc = 0);
2827 fid_cpu_to_be(tfid, cfid);
2828 rc = dt_lookup(env, com->lc_obj, (struct dt_rec *)&flags,
2829 (const struct dt_key *)tfid, BYPASS_CAPA);
2833 if (flags & LNTF_SKIP_NLINK)
2834 GOTO(unlock, rc = 0);
2836 rc = dt_attr_get(env, child, la, BYPASS_CAPA);
2838 GOTO(unlock, rc = (rc == -ENOENT ? 0 : rc));
2840 rc = lfsck_links_read2(env, child, &ldata);
2842 GOTO(unlock, rc = (rc == -ENODATA ? 0 : rc));
2844 if (la->la_nlink == ldata.ld_leh->leh_reccount ||
2845 unlikely(la->la_nlink == 0))
2846 GOTO(unlock, rc = 0);
2848 la->la_nlink = ldata.ld_leh->leh_reccount;
2849 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2850 GOTO(unlock, rc = 1);
2852 rc = dt_attr_set(env, child, la, th, BYPASS_CAPA);
2854 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2857 dt_write_unlock(env, child);
2860 dt_trans_stop(env, dev, th);
2863 lfsck_ibits_unlock(&lh, LCK_EX);
2864 if (child != NULL && !IS_ERR(child))
2865 lfsck_object_put(env, child);
2867 CDEBUG(D_LFSCK, "%s: namespace LFSCK repaired the object "DFID"'s "
2868 "nlink count from %u to %u: rc = %d\n",
2869 lfsck_lfsck2name(lfsck), PFID(cfid), old, la->la_nlink, rc);
2872 ns->ln_flags |= LF_INCONSISTENT;
2878 * Double scan the directory object for namespace LFSCK.
2880 * This function will verify the <parent, child> pairs in the namespace tree:
2881 * the parent references the child via some name entry that should be in the
2882 * child's linkEA entry, the child should back references the parent via its
2885 * The LFSCK will scan every linkEA entry in turn until find out the first
2886 * matched pairs. If found, then all other linkEA entries will be dropped.
2887 * If all the linkEA entries cannot match the ".." name entry, then there
2888 * are serveral possible cases:
2890 * 1) If there is only one linkEA entry, then trust it as long as the PFID
2891 * in the linkEA entry is valid.
2893 * 2) If there are multiple linkEA entries, then try to find the linkEA
2894 * that matches the ".." name entry. If found, then all other entries
2895 * are invalid; otherwise, it is quite possible that the ".." name entry
2896 * is corrupted. Under such case, the LFSCK will rebuild the ".." name
2897 * entry according to the first valid linkEA entry (both the parent and
2898 * the name entry should exist).
2900 * 3) If the directory object has no (valid) linkEA entry, then the
2901 * directory object will be handled as pure orphan and inserted
2902 * in the .lustre/lost+found/MDTxxxx/ with the name:
2903 * ${self_FID}-${PFID}-D-${conflict_version}
2905 * \param[in] env pointer to the thread context
2906 * \param[in] com pointer to the lfsck component
2907 * \param[in] child pointer to the directory object to be handled
2908 * \param[in] flags to indicate the specical checking on the @child
2910 * \retval positive number for repaired cases
2911 * \retval 0 if nothing to be repaired
2912 * \retval negative error number on failure
2914 static int lfsck_namespace_double_scan_dir(const struct lu_env *env,
2915 struct lfsck_component *com,
2916 struct dt_object *child, __u8 flags)
2918 struct lfsck_thread_info *info = lfsck_env_info(env);
2919 const struct lu_fid *cfid = lfsck_dto2fid(child);
2920 struct lu_fid *pfid = &info->lti_fid2;
2921 struct lfsck_namespace *ns = com->lc_file_ram;
2922 struct lfsck_instance *lfsck = com->lc_lfsck;
2923 struct lustre_handle lh = { 0 };
2924 struct linkea_data ldata = { NULL };
2925 bool unknown = false;
2928 enum lfsck_namespace_inconsistency_type type = LNIT_BAD_LINKEA;
2932 LASSERT(!dt_object_remote(child));
2934 if (flags & LNTF_UNCERTAIN_LMV) {
2935 if (flags & LNTF_RECHECK_NAME_HASH) {
2936 rc = lfsck_namespace_scan_shard(env, com, child);
2940 ns->ln_striped_shards_scanned++;
2942 ns->ln_striped_shards_skipped++;
2946 flags &= ~(LNTF_RECHECK_NAME_HASH | LNTF_UNCERTAIN_LMV);
2950 if (flags & (LNTF_CHECK_LINKEA | LNTF_CHECK_PARENT) &&
2951 !(lfsck->li_bookmark_ram.lb_param & LPF_ALL_TGT)) {
2952 CDEBUG(D_LFSCK, "%s: some MDT(s) maybe NOT take part in the"
2953 "the namespace LFSCK, then the LFSCK cannot guarantee"
2954 "all the name entries have been verified in first-stage"
2955 "scanning. So have to skip orphan related handling for"
2956 "the directory object "DFID" with remote name entry\n",
2957 lfsck_lfsck2name(lfsck), PFID(cfid));
2962 if (unlikely(!dt_try_as_dir(env, child)))
2963 GOTO(out, rc = -ENOTDIR);
2965 /* We only take ldlm lock on the @child when required. When the
2966 * logic comes here for the first time, it is always false. */
2970 rc = lfsck_ibits_lock(env, lfsck, child, &lh,
2971 MDS_INODELOCK_UPDATE |
2972 MDS_INODELOCK_XATTR, LCK_EX);
2977 dt_read_lock(env, child, 0);
2978 if (unlikely(lfsck_is_dead_obj(child))) {
2979 dt_read_unlock(env, child);
2984 rc = dt_lookup(env, child, (struct dt_rec *)pfid,
2985 (const struct dt_key *)dotdot, BYPASS_CAPA);
2987 if (rc != -ENOENT && rc != -ENODATA && rc != -EINVAL) {
2988 dt_read_unlock(env, child);
2993 if (!lustre_handle_is_used(&lh)) {
2994 dt_read_unlock(env, child);
2999 } else if (lfsck->li_lpf_obj != NULL &&
3000 lu_fid_eq(pfid, lfsck_dto2fid(lfsck->li_lpf_obj))) {
3002 } else if (unlikely(!fid_is_sane(pfid))) {
3006 rc = lfsck_links_read(env, child, &ldata);
3007 dt_read_unlock(env, child);
3009 if (rc != -ENODATA && rc != -EINVAL)
3012 if (!lustre_handle_is_used(&lh))
3015 if (rc == -EINVAL && !fid_is_zero(pfid)) {
3016 /* Remove the corrupted linkEA. */
3017 rc = lfsck_namespace_links_remove(env, com, child);
3019 /* Here, because of the crashed linkEA, we
3020 * cannot know whether there is some parent
3021 * that references the child directory via
3022 * some name entry or not. So keep it there,
3023 * when the LFSCK run next time, if there is
3024 * some parent that references this object,
3025 * then the LFSCK can rebuild the linkEA;
3026 * otherwise, this object will be handled
3027 * as orphan as above. */
3030 /* 1. If we have neither ".." nor linkEA,
3031 * then it is an orphan.
3033 * 2. If we only have the ".." name entry,
3034 * but no parent references this child
3035 * directory, then handle it as orphan. */
3036 lfsck_ibits_unlock(&lh, LCK_EX);
3037 type = LNIT_MUL_REF;
3039 /* If the LFSCK is marked as LF_INCOMPLETE,
3040 * then means some MDT has ever tried to
3041 * verify some remote MDT-object that resides
3042 * on this MDT, but this MDT failed to respond
3043 * such request. So means there may be some
3044 * remote name entry on other MDT that
3045 * references this object with another name,
3046 * so we cannot know whether this linkEA is
3047 * valid or not. So keep it there and maybe
3048 * resolved when next LFSCK run. */
3049 if (ns->ln_flags & LF_INCOMPLETE)
3052 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
3053 "-"DFID, PFID(pfid));
3054 rc = lfsck_namespace_insert_orphan(env, com, child,
3055 info->lti_tmpbuf, "D", NULL);
3061 linkea_first_entry(&ldata);
3062 /* This is the most common case: the object has unique linkEA entry. */
3063 if (ldata.ld_leh->leh_reccount == 1) {
3064 rc = lfsck_namespace_dsd_single(env, com, child, pfid, &ldata,
3065 &lh, &type, &retry);
3067 LASSERT(!lustre_handle_is_used(&lh));
3076 if (!lustre_handle_is_used(&lh))
3079 if (unlikely(ldata.ld_leh->leh_reccount == 0)) {
3080 rc = lfsck_namespace_dsd_orphan(env, com, child, pfid, &lh,
3086 /* When we come here, the cases usually like that:
3087 * 1) The directory object has a corrupted linkEA entry. During the
3088 * first-stage scanning, the LFSCK cannot know such corruption,
3089 * then it appends the right linkEA entry according to the found
3090 * name entry after the bad one.
3092 * 2) The directory object has a right linkEA entry. During the
3093 * first-stage scanning, the LFSCK finds some bad name entry,
3094 * but the LFSCK cannot aware that at that time, then it adds
3095 * the bad linkEA entry for further processing. */
3096 rc = lfsck_namespace_dsd_multiple(env, com, child, pfid, &ldata,
3102 lfsck_ibits_unlock(&lh, LCK_EX);
3105 case LNIT_BAD_LINKEA:
3106 ns->ln_linkea_repaired++;
3108 case LNIT_UNMATCHED_PAIRS:
3109 ns->ln_unmatched_pairs_repaired++;
3112 ns->ln_mul_ref_repaired++;
3120 ns->ln_unknown_inconsistency++;
3126 * Double scan the MDT-object for namespace LFSCK.
3128 * If the MDT-object contains invalid or repeated linkEA entries, then drop
3129 * those entries from the linkEA; if the linkEA becomes empty or the object
3130 * has no linkEA, then it is an orphan and will be added into the directory
3131 * .lustre/lost+found/MDTxxxx/; if the remote parent is lost, then recreate
3132 * the remote parent; if the name entry corresponding to some linkEA entry
3133 * is lost, then add the name entry back to the namespace.
3135 * \param[in] env pointer to the thread context
3136 * \param[in] com pointer to the lfsck component
3137 * \param[in] child pointer to the dt_object to be handled
3138 * \param[in] flags some hints to indicate how the @child should be handled
3140 * \retval positive number for repaired cases
3141 * \retval 0 if nothing to be repaired
3142 * \retval negative error number on failure
3144 static int lfsck_namespace_double_scan_one(const struct lu_env *env,
3145 struct lfsck_component *com,
3146 struct dt_object *child, __u8 flags)
3148 struct lfsck_thread_info *info = lfsck_env_info(env);
3149 struct lu_attr *la = &info->lti_la;
3150 struct lu_name *cname = &info->lti_name;
3151 struct lu_fid *pfid = &info->lti_fid;
3152 struct lu_fid *cfid = &info->lti_fid2;
3153 struct lfsck_instance *lfsck = com->lc_lfsck;
3154 struct lfsck_namespace *ns = com->lc_file_ram;
3155 struct dt_object *parent = NULL;
3156 struct linkea_data ldata = { NULL };
3157 bool repaired = false;
3162 dt_read_lock(env, child, 0);
3163 if (unlikely(lfsck_is_dead_obj(child))) {
3164 dt_read_unlock(env, child);
3169 if (S_ISDIR(lfsck_object_type(child))) {
3170 dt_read_unlock(env, child);
3171 rc = lfsck_namespace_double_scan_dir(env, com, child, flags);
3176 rc = lfsck_links_read(env, child, &ldata);
3177 dt_read_unlock(env, child);
3181 linkea_first_entry(&ldata);
3182 while (ldata.ld_lee != NULL) {
3183 lfsck_namespace_unpack_linkea_entry(&ldata, cname, pfid,
3185 rc = lfsck_namespace_filter_linkea_entry(&ldata, cname, pfid,
3187 /* Found repeated linkEA entries */
3189 rc = lfsck_namespace_shrink_linkea(env, com, child,
3190 &ldata, cname, pfid, false);
3202 /* Invalid PFID in the linkEA entry. */
3203 if (!fid_is_sane(pfid)) {
3204 rc = lfsck_namespace_shrink_linkea(env, com, child,
3205 &ldata, cname, pfid, true);
3215 parent = lfsck_object_find_bottom(env, lfsck, pfid);
3217 GOTO(out, rc = PTR_ERR(parent));
3219 if (!dt_object_exists(parent)) {
3222 if (ldata.ld_leh->leh_reccount > 1) {
3223 /* If it is NOT the last linkEA entry, then
3224 * there is still other chance to make the
3225 * child to be visible via other parent, then
3226 * remove this linkEA entry. */
3227 rc = lfsck_namespace_shrink_linkea(env, com,
3228 child, &ldata, cname, pfid, true);
3230 /* If the LFSCK is marked as LF_INCOMPLETE,
3231 * then means some MDT has ever tried to
3232 * verify some remote MDT-object that resides
3233 * on this MDT, but this MDT failed to respond
3234 * such request. So means there may be some
3235 * remote name entry on other MDT that
3236 * references this object with another name,
3237 * so we cannot know whether this linkEA is
3238 * valid or not. So keep it there and maybe
3239 * resolved when next LFSCK run. */
3240 if (ns->ln_flags & LF_INCOMPLETE) {
3241 lfsck_object_put(env, parent);
3246 /* Create the lost parent as an orphan. */
3247 rc = lfsck_namespace_create_orphan_dir(env, com,
3250 lfsck_object_put(env, parent);
3258 /* Add the missing name entry to the parent. */
3259 rc = lfsck_namespace_insert_normal(env, com,
3260 parent, child, cname->ln_name);
3261 if (unlikely(rc == -EEXIST))
3262 /* Unfortunately, someone reused the
3263 * name under the parent by race. So we
3264 * have to remove the linkEA entry from
3265 * current child object. It means that
3266 * the LFSCK cannot recover the system
3267 * totally back to its original status,
3268 * but it is necessary to make the
3269 * current system to be consistent. */
3270 rc = lfsck_namespace_shrink_linkea(env,
3274 linkea_next_entry(&ldata);
3277 lfsck_object_put(env, parent);
3287 /* The linkEA entry with bad parent will be removed. */
3288 if (unlikely(!dt_try_as_dir(env, parent))) {
3289 lfsck_object_put(env, parent);
3290 rc = lfsck_namespace_shrink_linkea(env, com, child,
3291 &ldata, cname, pfid, true);
3301 rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
3302 (const struct dt_key *)cname->ln_name,
3304 if (rc != 0 && rc != -ENOENT) {
3305 lfsck_object_put(env, parent);
3311 if (lu_fid_eq(cfid, lfsck_dto2fid(child))) {
3312 /* It is the most common case that we
3313 * find the name entry corresponding
3314 * to the linkEA entry. */
3315 lfsck_object_put(env, parent);
3316 linkea_next_entry(&ldata);
3318 /* The name entry references another
3319 * MDT-object that may be created by
3320 * the LFSCK for repairing dangling
3321 * name entry. Try to replace it. */
3322 rc = lfsck_namespace_replace_cond(env, com,
3323 parent, child, cfid, cname);
3324 lfsck_object_put(env, parent);
3330 linkea_next_entry(&ldata);
3332 rc = lfsck_namespace_shrink_linkea(env,
3346 rc = dt_attr_get(env, child, la, BYPASS_CAPA);
3350 /* If there is no name entry in the parent dir and the object
3351 * link count is less than the linkea entries count, then the
3352 * linkea entry should be removed. */
3353 if (ldata.ld_leh->leh_reccount > la->la_nlink) {
3354 rc = lfsck_namespace_shrink_linkea_cond(env, com,
3355 parent, child, &ldata, cname, pfid);
3356 lfsck_object_put(env, parent);
3366 /* If the LFSCK is marked as LF_INCOMPLETE, then means some
3367 * MDT has ever tried to verify some remote MDT-object that
3368 * resides on this MDT, but this MDT failed to respond such
3369 * request. So means there may be some remote name entry on
3370 * other MDT that references this object with another name,
3371 * so we cannot know whether this linkEA is valid or not.
3372 * So keep it there and maybe resolved when next LFSCK run. */
3373 if (ns->ln_flags & LF_INCOMPLETE) {
3374 lfsck_object_put(env, parent);
3379 rc = lfsck_namespace_check_name(env, parent, child, cname);
3384 lfsck_object_put(env, parent);
3389 /* It is an invalid name entry, drop it. */
3390 if (unlikely(rc > 0)) {
3391 lfsck_object_put(env, parent);
3392 rc = lfsck_namespace_shrink_linkea(env, com, child,
3393 &ldata, cname, pfid, true);
3403 /* Add the missing name entry back to the namespace. */
3404 rc = lfsck_namespace_insert_normal(env, com, parent, child,
3406 if (unlikely(rc == -ESTALE))
3407 /* It may happen when the remote object has been
3408 * removed, but the local MDT is not aware of that. */
3411 if (unlikely(rc == -EEXIST))
3412 /* Unfortunately, someone reused the name under the
3413 * parent by race. So we have to remove the linkEA
3414 * entry from current child object. It means that the
3415 * LFSCK cannot recover the system totally back to
3416 * its original status, but it is necessary to make
3417 * the current system to be consistent.
3419 * It also may be because of the LFSCK found some
3420 * internal status of create operation. Under such
3421 * case, nothing to be done. */
3422 rc = lfsck_namespace_shrink_linkea_cond(env, com,
3423 parent, child, &ldata, cname, pfid);
3425 linkea_next_entry(&ldata);
3427 lfsck_object_put(env, parent);
3438 if (rc < 0 && rc != -ENODATA)
3442 LASSERT(ldata.ld_leh != NULL);
3444 count = ldata.ld_leh->leh_reccount;
3448 /* If the LFSCK is marked as LF_INCOMPLETE, then means some
3449 * MDT has ever tried to verify some remote MDT-object that
3450 * resides on this MDT, but this MDT failed to respond such
3451 * request. So means there may be some remote name entry on
3452 * other MDT that references this object with another name,
3453 * so we cannot know whether this linkEA is valid or not.
3454 * So keep it there and maybe resolved when next LFSCK run. */
3455 if (!(ns->ln_flags & LF_INCOMPLETE)) {
3456 /* If the child becomes orphan, then insert it into
3457 * the global .lustre/lost+found/MDTxxxx directory. */
3458 rc = lfsck_namespace_insert_orphan(env, com, child,
3464 ns->ln_mul_ref_repaired++;
3469 rc = dt_attr_get(env, child, la, BYPASS_CAPA);
3473 if (la->la_nlink != 0 && la->la_nlink != count) {
3474 rc = lfsck_namespace_repair_nlink(env, com, child, la);
3476 ns->ln_objs_nlink_repaired++;
3483 if (la->la_nlink > 1)
3484 ns->ln_mul_linked_repaired++;
3493 static void lfsck_namespace_dump_statistics(struct seq_file *m,
3494 struct lfsck_namespace *ns,
3495 __u64 checked_phase1,
3496 __u64 checked_phase2,
3500 seq_printf(m, "checked_phase1: "LPU64"\n"
3501 "checked_phase2: "LPU64"\n"
3502 "updated_phase1: "LPU64"\n"
3503 "updated_phase2: "LPU64"\n"
3504 "failed_phase1: "LPU64"\n"
3505 "failed_phase2: "LPU64"\n"
3506 "directories: "LPU64"\n"
3507 "dirent_repaired: "LPU64"\n"
3508 "linkea_repaired: "LPU64"\n"
3509 "nlinks_repaired: "LPU64"\n"
3510 "multiple_linked_checked: "LPU64"\n"
3511 "multiple_linked_repaired: "LPU64"\n"
3512 "unknown_inconsistency: "LPU64"\n"
3513 "unmatched_pairs_repaired: "LPU64"\n"
3514 "dangling_repaired: "LPU64"\n"
3515 "multiple_referenced_repaired: "LPU64"\n"
3516 "bad_file_type_repaired: "LPU64"\n"
3517 "lost_dirent_repaired: "LPU64"\n"
3518 "local_lost_found_scanned: "LPU64"\n"
3519 "local_lost_found_moved: "LPU64"\n"
3520 "local_lost_found_skipped: "LPU64"\n"
3521 "local_lost_found_failed: "LPU64"\n"
3522 "striped_dirs_scanned: "LPU64"\n"
3523 "striped_dirs_repaired: "LPU64"\n"
3524 "striped_dirs_failed: "LPU64"\n"
3525 "striped_dirs_disabled: "LPU64"\n"
3526 "striped_dirs_skipped: "LPU64"\n"
3527 "striped_shards_scanned: "LPU64"\n"
3528 "striped_shards_repaired: "LPU64"\n"
3529 "striped_shards_failed: "LPU64"\n"
3530 "striped_shards_skipped: "LPU64"\n"
3531 "name_hash_repaired: "LPU64"\n"
3532 "success_count: %u\n"
3533 "run_time_phase1: %u seconds\n"
3534 "run_time_phase2: %u seconds\n",
3537 ns->ln_items_repaired,
3538 ns->ln_objs_repaired_phase2,
3539 ns->ln_items_failed,
3540 ns->ln_objs_failed_phase2,
3541 ns->ln_dirs_checked,
3542 ns->ln_dirent_repaired,
3543 ns->ln_linkea_repaired,
3544 ns->ln_objs_nlink_repaired,
3545 ns->ln_mul_linked_checked,
3546 ns->ln_mul_linked_repaired,
3547 ns->ln_unknown_inconsistency,
3548 ns->ln_unmatched_pairs_repaired,
3549 ns->ln_dangling_repaired,
3550 ns->ln_mul_ref_repaired,
3551 ns->ln_bad_type_repaired,
3552 ns->ln_lost_dirent_repaired,
3553 ns->ln_local_lpf_scanned,
3554 ns->ln_local_lpf_moved,
3555 ns->ln_local_lpf_skipped,
3556 ns->ln_local_lpf_failed,
3557 ns->ln_striped_dirs_scanned,
3558 ns->ln_striped_dirs_repaired,
3559 ns->ln_striped_dirs_failed,
3560 ns->ln_striped_dirs_disabled,
3561 ns->ln_striped_dirs_skipped,
3562 ns->ln_striped_shards_scanned,
3563 ns->ln_striped_shards_repaired,
3564 ns->ln_striped_shards_failed,
3565 ns->ln_striped_shards_skipped,
3566 ns->ln_name_hash_repaired,
3567 ns->ln_success_count,
3572 static void lfsck_namespace_release_lmv(const struct lu_env *env,
3573 struct lfsck_component *com)
3575 struct lfsck_instance *lfsck = com->lc_lfsck;
3576 struct lfsck_namespace *ns = com->lc_file_ram;
3578 while (!list_empty(&lfsck->li_list_lmv)) {
3579 struct lfsck_lmv_unit *llu;
3580 struct lfsck_lmv *llmv;
3582 llu = list_entry(lfsck->li_list_lmv.next,
3583 struct lfsck_lmv_unit, llu_link);
3584 llmv = &llu->llu_lmv;
3586 LASSERTF(atomic_read(&llmv->ll_ref) == 1,
3587 "still in using: %u\n",
3588 atomic_read(&llmv->ll_ref));
3590 ns->ln_striped_dirs_skipped++;
3591 lfsck_lmv_put(env, llmv);
3595 /* namespace APIs */
3597 static int lfsck_namespace_reset(const struct lu_env *env,
3598 struct lfsck_component *com, bool init)
3600 struct lfsck_instance *lfsck = com->lc_lfsck;
3601 struct lfsck_namespace *ns = com->lc_file_ram;
3602 struct lfsck_assistant_data *lad = com->lc_data;
3603 struct dt_object *root;
3604 struct dt_object *dto;
3608 root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
3610 GOTO(log, rc = PTR_ERR(root));
3612 if (unlikely(!dt_try_as_dir(env, root)))
3613 GOTO(put, rc = -ENOTDIR);
3615 down_write(&com->lc_sem);
3617 memset(ns, 0, sizeof(*ns));
3619 __u32 count = ns->ln_success_count;
3620 __u64 last_time = ns->ln_time_last_complete;
3622 memset(ns, 0, sizeof(*ns));
3623 ns->ln_success_count = count;
3624 ns->ln_time_last_complete = last_time;
3626 ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
3627 ns->ln_status = LS_INIT;
3629 rc = local_object_unlink(env, lfsck->li_bottom, root,
3630 lfsck_namespace_name);
3634 lfsck_object_put(env, com->lc_obj);
3636 dto = local_index_find_or_create(env, lfsck->li_los, root,
3637 lfsck_namespace_name,
3638 S_IFREG | S_IRUGO | S_IWUSR,
3639 &dt_lfsck_features);
3641 GOTO(out, rc = PTR_ERR(dto));
3644 rc = dto->do_ops->do_index_try(env, dto, &dt_lfsck_features);
3648 lad->lad_incomplete = 0;
3649 CFS_RESET_BITMAP(lad->lad_bitmap);
3651 rc = lfsck_namespace_store(env, com, true);
3656 up_write(&com->lc_sem);
3659 lu_object_put(env, &root->do_lu);
3661 CDEBUG(D_LFSCK, "%s: namespace LFSCK reset: rc = %d\n",
3662 lfsck_lfsck2name(lfsck), rc);
3667 lfsck_namespace_fail(const struct lu_env *env, struct lfsck_component *com,
3670 struct lfsck_namespace *ns = com->lc_file_ram;
3672 down_write(&com->lc_sem);
3674 com->lc_new_checked++;
3675 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
3676 up_write(&com->lc_sem);
3679 static void lfsck_namespace_close_dir(const struct lu_env *env,
3680 struct lfsck_component *com)
3682 struct lfsck_namespace *ns = com->lc_file_ram;
3683 struct lfsck_assistant_data *lad = com->lc_data;
3684 struct lfsck_instance *lfsck = com->lc_lfsck;
3685 struct lfsck_lmv *llmv = lfsck->li_lmv;
3686 struct lfsck_namespace_req *lnr;
3688 sizeof(*lnr) + LFSCK_TMPBUF_LEN;
3689 bool wakeup = false;
3695 OBD_ALLOC(lnr, size);
3697 ns->ln_striped_dirs_skipped++;
3702 /* Generate a dummy request to indicate that all shards' name entry
3703 * in this striped directory has been scanned for the first time. */
3704 INIT_LIST_HEAD(&lnr->lnr_lar.lar_list);
3705 lnr->lnr_obj = lfsck_object_get(lfsck->li_obj_dir);
3706 lnr->lnr_lmv = lfsck_lmv_get(llmv);
3707 lnr->lnr_fid = *lfsck_dto2fid(lfsck->li_obj_dir);
3708 lnr->lnr_oit_cookie = lfsck->li_pos_current.lp_oit_cookie;
3709 lnr->lnr_dir_cookie = MDS_DIR_END_OFF;
3710 lnr->lnr_size = size;
3712 spin_lock(&lad->lad_lock);
3713 if (lad->lad_assistant_status < 0) {
3714 spin_unlock(&lad->lad_lock);
3715 lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar);
3716 ns->ln_striped_dirs_skipped++;
3721 list_add_tail(&lnr->lnr_lar.lar_list, &lad->lad_req_list);
3722 if (lad->lad_prefetched == 0)
3725 lad->lad_prefetched++;
3726 spin_unlock(&lad->lad_lock);
3728 wake_up_all(&lad->lad_thread.t_ctl_waitq);
3733 static int lfsck_namespace_open_dir(const struct lu_env *env,
3734 struct lfsck_component *com)
3736 struct lfsck_instance *lfsck = com->lc_lfsck;
3737 struct lfsck_namespace *ns = com->lc_file_ram;
3738 struct lfsck_lmv *llmv = lfsck->li_lmv;
3745 if (llmv->ll_lmv_master) {
3746 struct lmv_mds_md_v1 *lmv = &llmv->ll_lmv;
3748 if (lmv->lmv_master_mdt_index !=
3749 lfsck_dev_idx(lfsck->li_bottom)) {
3750 lmv->lmv_master_mdt_index =
3751 lfsck_dev_idx(lfsck->li_bottom);
3752 ns->ln_flags |= LF_INCONSISTENT;
3753 llmv->ll_lmv_updated = 1;
3756 rc = lfsck_namespace_verify_stripe_slave(env, com,
3757 lfsck->li_obj_dir, llmv);
3760 RETURN(rc > 0 ? 0 : rc);
3763 static int lfsck_namespace_checkpoint(const struct lu_env *env,
3764 struct lfsck_component *com, bool init)
3766 struct lfsck_instance *lfsck = com->lc_lfsck;
3767 struct lfsck_namespace *ns = com->lc_file_ram;
3771 rc = lfsck_checkpoint_generic(env, com);
3776 down_write(&com->lc_sem);
3778 ns->ln_pos_latest_start = lfsck->li_pos_checkpoint;
3780 ns->ln_pos_last_checkpoint = lfsck->li_pos_checkpoint;
3781 ns->ln_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
3782 HALF_SEC - lfsck->li_time_last_checkpoint);
3783 ns->ln_time_last_checkpoint = cfs_time_current_sec();
3784 ns->ln_items_checked += com->lc_new_checked;
3785 com->lc_new_checked = 0;
3788 rc = lfsck_namespace_store(env, com, false);
3789 up_write(&com->lc_sem);
3792 CDEBUG(D_LFSCK, "%s: namespace LFSCK checkpoint at the pos ["LPU64
3793 ", "DFID", "LPX64"]: rc = %d\n", lfsck_lfsck2name(lfsck),
3794 lfsck->li_pos_current.lp_oit_cookie,
3795 PFID(&lfsck->li_pos_current.lp_dir_parent),
3796 lfsck->li_pos_current.lp_dir_cookie, rc);
3798 return rc > 0 ? 0 : rc;
3801 static int lfsck_namespace_prep(const struct lu_env *env,
3802 struct lfsck_component *com,
3803 struct lfsck_start_param *lsp)
3805 struct lfsck_instance *lfsck = com->lc_lfsck;
3806 struct lfsck_namespace *ns = com->lc_file_ram;
3807 struct lfsck_position *pos = &com->lc_pos_start;
3810 rc = lfsck_namespace_load_bitmap(env, com);
3811 if (rc != 0 || ns->ln_status == LS_COMPLETED) {
3812 rc = lfsck_namespace_reset(env, com, false);
3814 rc = lfsck_set_param(env, lfsck, lsp->lsp_start, true);
3817 CDEBUG(D_LFSCK, "%s: namespace LFSCK prep failed: "
3818 "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
3824 down_write(&com->lc_sem);
3825 ns->ln_time_latest_start = cfs_time_current_sec();
3826 spin_lock(&lfsck->li_lock);
3828 if (ns->ln_flags & LF_SCANNED_ONCE) {
3829 if (!lfsck->li_drop_dryrun ||
3830 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent)) {
3831 ns->ln_status = LS_SCANNING_PHASE2;
3832 list_move_tail(&com->lc_link,
3833 &lfsck->li_list_double_scan);
3834 if (!list_empty(&com->lc_link_dir))
3835 list_del_init(&com->lc_link_dir);
3836 lfsck_pos_set_zero(pos);
3838 ns->ln_status = LS_SCANNING_PHASE1;
3839 ns->ln_run_time_phase1 = 0;
3840 ns->ln_run_time_phase2 = 0;
3841 ns->ln_items_checked = 0;
3842 ns->ln_items_repaired = 0;
3843 ns->ln_items_failed = 0;
3844 ns->ln_dirs_checked = 0;
3845 ns->ln_objs_checked_phase2 = 0;
3846 ns->ln_objs_repaired_phase2 = 0;
3847 ns->ln_objs_failed_phase2 = 0;
3848 ns->ln_objs_nlink_repaired = 0;
3849 ns->ln_dirent_repaired = 0;
3850 ns->ln_linkea_repaired = 0;
3851 ns->ln_mul_linked_checked = 0;
3852 ns->ln_mul_linked_repaired = 0;
3853 ns->ln_unknown_inconsistency = 0;
3854 ns->ln_unmatched_pairs_repaired = 0;
3855 ns->ln_dangling_repaired = 0;
3856 ns->ln_mul_ref_repaired = 0;
3857 ns->ln_bad_type_repaired = 0;
3858 ns->ln_lost_dirent_repaired = 0;
3859 ns->ln_striped_dirs_scanned = 0;
3860 ns->ln_striped_dirs_repaired = 0;
3861 ns->ln_striped_dirs_failed = 0;
3862 ns->ln_striped_dirs_disabled = 0;
3863 ns->ln_striped_dirs_skipped = 0;
3864 ns->ln_striped_shards_scanned = 0;
3865 ns->ln_striped_shards_repaired = 0;
3866 ns->ln_striped_shards_failed = 0;
3867 ns->ln_striped_shards_skipped = 0;
3868 ns->ln_name_hash_repaired = 0;
3869 fid_zero(&ns->ln_fid_latest_scanned_phase2);
3870 if (list_empty(&com->lc_link_dir))
3871 list_add_tail(&com->lc_link_dir,
3872 &lfsck->li_list_dir);
3873 *pos = ns->ln_pos_first_inconsistent;
3876 ns->ln_status = LS_SCANNING_PHASE1;
3877 if (list_empty(&com->lc_link_dir))
3878 list_add_tail(&com->lc_link_dir,
3879 &lfsck->li_list_dir);
3880 if (!lfsck->li_drop_dryrun ||
3881 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent)) {
3882 *pos = ns->ln_pos_last_checkpoint;
3883 pos->lp_oit_cookie++;
3885 *pos = ns->ln_pos_first_inconsistent;
3889 spin_unlock(&lfsck->li_lock);
3890 up_write(&com->lc_sem);
3892 rc = lfsck_start_assistant(env, com, lsp);
3894 CDEBUG(D_LFSCK, "%s: namespace LFSCK prep done, start pos ["LPU64", "
3895 DFID", "LPX64"]: rc = %d\n",
3896 lfsck_lfsck2name(lfsck), pos->lp_oit_cookie,
3897 PFID(&pos->lp_dir_parent), pos->lp_dir_cookie, rc);
3902 static int lfsck_namespace_exec_oit(const struct lu_env *env,
3903 struct lfsck_component *com,
3904 struct dt_object *obj)
3906 struct lfsck_thread_info *info = lfsck_env_info(env);
3907 struct lfsck_namespace *ns = com->lc_file_ram;
3908 struct lfsck_instance *lfsck = com->lc_lfsck;
3909 const struct lu_fid *fid = lfsck_dto2fid(obj);
3910 struct lu_attr *la = &info->lti_la;
3911 struct lu_fid *pfid = &info->lti_fid2;
3912 struct lu_name *cname = &info->lti_name;
3913 struct lu_seq_range *range = &info->lti_range;
3914 struct dt_device *dev = lfsck->li_bottom;
3915 struct seq_server_site *ss =
3916 lu_site2seq(dev->dd_lu_dev.ld_site);
3917 struct linkea_data ldata = { NULL };
3918 __u32 idx = lfsck_dev_idx(dev);
3922 rc = lfsck_links_read(env, obj, &ldata);
3926 /* -EINVAL means crashed linkEA, should be verified. */
3927 if (rc == -EINVAL) {
3928 rc = lfsck_namespace_trace_update(env, com, fid,
3929 LNTF_CHECK_LINKEA, true);
3931 struct lustre_handle lh = { 0 };
3933 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
3934 MDS_INODELOCK_UPDATE |
3935 MDS_INODELOCK_XATTR, LCK_EX);
3937 rc = lfsck_namespace_links_remove(env, com,
3939 lfsck_ibits_unlock(&lh, LCK_EX);
3943 GOTO(out, rc = (rc == -ENOENT ? 0 : rc));
3946 /* zero-linkEA object may be orphan, but it also maybe because
3947 * of upgrading. Currently, we cannot record it for double scan.
3948 * Because it may cause the LFSCK trace file to be too large. */
3949 if (rc == -ENODATA) {
3950 if (S_ISDIR(lfsck_object_type(obj)))
3953 rc = dt_attr_get(env, obj, la, BYPASS_CAPA);
3957 /* "la_ctime" == 1 means that it has ever been removed from
3958 * backend /lost+found directory but not been added back to
3959 * the normal namespace yet. */
3960 if (la->la_nlink > 1 || unlikely(la->la_ctime == 1))
3961 rc = lfsck_namespace_trace_update(env, com, fid,
3962 LNTF_CHECK_LINKEA, true);
3970 /* Record multiple-linked object. */
3971 if (ldata.ld_leh->leh_reccount > 1) {
3972 rc = lfsck_namespace_trace_update(env, com, fid,
3973 LNTF_CHECK_LINKEA, true);
3978 linkea_first_entry(&ldata);
3979 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, cname, pfid);
3980 if (!fid_is_sane(pfid)) {
3981 rc = lfsck_namespace_trace_update(env, com, fid,
3982 LNTF_CHECK_PARENT, true);
3984 fld_range_set_mdt(range);
3985 rc = fld_local_lookup(env, ss->ss_server_fld,
3986 fid_seq(pfid), range);
3987 if ((rc == -ENOENT) ||
3988 (rc == 0 && range->lsr_index != idx)) {
3989 rc = lfsck_namespace_trace_update(env, com, fid,
3990 LNTF_CHECK_LINKEA, true);
3992 if (S_ISDIR(lfsck_object_type(obj)))
3995 rc = dt_attr_get(env, obj, la, BYPASS_CAPA);
3999 /* "la_ctime" == 1 means that it has ever been
4000 * removed from backend /lost+found directory but
4001 * not been added back to the normal namespace yet. */
4002 if (la->la_nlink > 1 || unlikely(la->la_ctime == 1))
4003 rc = lfsck_namespace_trace_update(env, com,
4004 fid, LNTF_CHECK_LINKEA, true);
4011 down_write(&com->lc_sem);
4012 com->lc_new_checked++;
4013 if (S_ISDIR(lfsck_object_type(obj)))
4014 ns->ln_dirs_checked++;
4016 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
4017 up_write(&com->lc_sem);
4022 static int lfsck_namespace_exec_dir(const struct lu_env *env,
4023 struct lfsck_component *com,
4024 struct lu_dirent *ent, __u16 type)
4026 struct lfsck_assistant_data *lad = com->lc_data;
4027 struct lfsck_instance *lfsck = com->lc_lfsck;
4028 struct lfsck_namespace_req *lnr;
4029 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4030 struct ptlrpc_thread *mthread = &lfsck->li_thread;
4031 struct ptlrpc_thread *athread = &lad->lad_thread;
4032 struct l_wait_info lwi = { 0 };
4033 bool wakeup = false;
4035 l_wait_event(mthread->t_ctl_waitq,
4036 bk->lb_async_windows == 0 ||
4037 lad->lad_prefetched < bk->lb_async_windows ||
4038 !thread_is_running(mthread) ||
4039 thread_is_stopped(athread),
4042 if (unlikely(!thread_is_running(mthread)) ||
4043 thread_is_stopped(athread))
4046 if (unlikely(lfsck_is_dead_obj(lfsck->li_obj_dir)))
4049 lnr = lfsck_namespace_assistant_req_init(com->lc_lfsck, ent, type);
4051 struct lfsck_namespace *ns = com->lc_file_ram;
4053 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
4054 return PTR_ERR(lnr);
4057 spin_lock(&lad->lad_lock);
4058 if (lad->lad_assistant_status < 0) {
4059 spin_unlock(&lad->lad_lock);
4060 lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar);
4061 return lad->lad_assistant_status;
4064 list_add_tail(&lnr->lnr_lar.lar_list, &lad->lad_req_list);
4065 if (lad->lad_prefetched == 0)
4068 lad->lad_prefetched++;
4069 spin_unlock(&lad->lad_lock);
4071 wake_up_all(&lad->lad_thread.t_ctl_waitq);
4073 down_write(&com->lc_sem);
4074 com->lc_new_checked++;
4075 up_write(&com->lc_sem);
4080 static int lfsck_namespace_post(const struct lu_env *env,
4081 struct lfsck_component *com,
4082 int result, bool init)
4084 struct lfsck_instance *lfsck = com->lc_lfsck;
4085 struct lfsck_namespace *ns = com->lc_file_ram;
4089 lfsck_post_generic(env, com, &result);
4091 down_write(&com->lc_sem);
4092 lfsck_namespace_release_lmv(env, com);
4094 spin_lock(&lfsck->li_lock);
4096 ns->ln_pos_last_checkpoint = lfsck->li_pos_checkpoint;
4098 ns->ln_status = LS_SCANNING_PHASE2;
4099 ns->ln_flags |= LF_SCANNED_ONCE;
4100 ns->ln_flags &= ~LF_UPGRADE;
4101 list_del_init(&com->lc_link_dir);
4102 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
4103 } else if (result == 0) {
4104 if (lfsck->li_status != 0)
4105 ns->ln_status = lfsck->li_status;
4107 ns->ln_status = LS_STOPPED;
4108 if (ns->ln_status != LS_PAUSED) {
4109 list_del_init(&com->lc_link_dir);
4110 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4113 ns->ln_status = LS_FAILED;
4114 list_del_init(&com->lc_link_dir);
4115 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4117 spin_unlock(&lfsck->li_lock);
4120 ns->ln_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
4121 HALF_SEC - lfsck->li_time_last_checkpoint);
4122 ns->ln_time_last_checkpoint = cfs_time_current_sec();
4123 ns->ln_items_checked += com->lc_new_checked;
4124 com->lc_new_checked = 0;
4127 rc = lfsck_namespace_store(env, com, false);
4128 up_write(&com->lc_sem);
4130 CDEBUG(D_LFSCK, "%s: namespace LFSCK post done: rc = %d\n",
4131 lfsck_lfsck2name(lfsck), rc);
4137 lfsck_namespace_dump(const struct lu_env *env, struct lfsck_component *com,
4140 struct lfsck_instance *lfsck = com->lc_lfsck;
4141 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4142 struct lfsck_namespace *ns = com->lc_file_ram;
4145 down_read(&com->lc_sem);
4146 seq_printf(m, "name: lfsck_namespace\n"
4152 lfsck_status2names(ns->ln_status));
4154 rc = lfsck_bits_dump(m, ns->ln_flags, lfsck_flags_names, "flags");
4158 rc = lfsck_bits_dump(m, bk->lb_param, lfsck_param_names, "param");
4162 rc = lfsck_time_dump(m, ns->ln_time_last_complete,
4163 "time_since_last_completed");
4167 rc = lfsck_time_dump(m, ns->ln_time_latest_start,
4168 "time_since_latest_start");
4172 rc = lfsck_time_dump(m, ns->ln_time_last_checkpoint,
4173 "time_since_last_checkpoint");
4177 rc = lfsck_pos_dump(m, &ns->ln_pos_latest_start,
4178 "latest_start_position");
4182 rc = lfsck_pos_dump(m, &ns->ln_pos_last_checkpoint,
4183 "last_checkpoint_position");
4187 rc = lfsck_pos_dump(m, &ns->ln_pos_first_inconsistent,
4188 "first_failure_position");
4192 if (ns->ln_status == LS_SCANNING_PHASE1) {
4193 struct lfsck_position pos;
4194 const struct dt_it_ops *iops;
4195 cfs_duration_t duration = cfs_time_current() -
4196 lfsck->li_time_last_checkpoint;
4197 __u64 checked = ns->ln_items_checked + com->lc_new_checked;
4198 __u64 speed = checked;
4199 __u64 new_checked = com->lc_new_checked *
4200 msecs_to_jiffies(MSEC_PER_SEC);
4201 __u32 rtime = ns->ln_run_time_phase1 +
4202 cfs_duration_sec(duration + HALF_SEC);
4205 do_div(new_checked, duration);
4207 do_div(speed, rtime);
4208 lfsck_namespace_dump_statistics(m, ns, checked,
4209 ns->ln_objs_checked_phase2,
4210 rtime, ns->ln_run_time_phase2);
4212 seq_printf(m, "average_speed_phase1: "LPU64" items/sec\n"
4213 "average_speed_phase2: N/A\n"
4214 "real_time_speed_phase1: "LPU64" items/sec\n"
4215 "real_time_speed_phase2: N/A\n",
4219 LASSERT(lfsck->li_di_oit != NULL);
4221 iops = &lfsck->li_obj_oit->do_index_ops->dio_it;
4223 /* The low layer otable-based iteration position may NOT
4224 * exactly match the namespace-based directory traversal
4225 * cookie. Generally, it is not a serious issue. But the
4226 * caller should NOT make assumption on that. */
4227 pos.lp_oit_cookie = iops->store(env, lfsck->li_di_oit);
4228 if (!lfsck->li_current_oit_processed)
4229 pos.lp_oit_cookie--;
4231 spin_lock(&lfsck->li_lock);
4232 if (lfsck->li_di_dir != NULL) {
4233 pos.lp_dir_cookie = lfsck->li_cookie_dir;
4234 if (pos.lp_dir_cookie >= MDS_DIR_END_OFF) {
4235 fid_zero(&pos.lp_dir_parent);
4236 pos.lp_dir_cookie = 0;
4239 *lfsck_dto2fid(lfsck->li_obj_dir);
4242 fid_zero(&pos.lp_dir_parent);
4243 pos.lp_dir_cookie = 0;
4245 spin_unlock(&lfsck->li_lock);
4246 lfsck_pos_dump(m, &pos, "current_position");
4247 } else if (ns->ln_status == LS_SCANNING_PHASE2) {
4248 cfs_duration_t duration = cfs_time_current() -
4249 lfsck->li_time_last_checkpoint;
4250 __u64 checked = ns->ln_objs_checked_phase2 +
4251 com->lc_new_checked;
4252 __u64 speed1 = ns->ln_items_checked;
4253 __u64 speed2 = checked;
4254 __u64 new_checked = com->lc_new_checked *
4255 msecs_to_jiffies(MSEC_PER_SEC);
4256 __u32 rtime = ns->ln_run_time_phase2 +
4257 cfs_duration_sec(duration + HALF_SEC);
4260 do_div(new_checked, duration);
4261 if (ns->ln_run_time_phase1 != 0)
4262 do_div(speed1, ns->ln_run_time_phase1);
4264 do_div(speed2, rtime);
4265 lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked,
4267 ns->ln_run_time_phase1, rtime);
4269 seq_printf(m, "average_speed_phase1: "LPU64" items/sec\n"
4270 "average_speed_phase2: "LPU64" objs/sec\n"
4271 "real_time_speed_phase1: N/A\n"
4272 "real_time_speed_phase2: "LPU64" objs/sec\n"
4273 "current_position: "DFID"\n",
4277 PFID(&ns->ln_fid_latest_scanned_phase2));
4279 __u64 speed1 = ns->ln_items_checked;
4280 __u64 speed2 = ns->ln_objs_checked_phase2;
4282 if (ns->ln_run_time_phase1 != 0)
4283 do_div(speed1, ns->ln_run_time_phase1);
4284 if (ns->ln_run_time_phase2 != 0)
4285 do_div(speed2, ns->ln_run_time_phase2);
4286 lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked,
4287 ns->ln_objs_checked_phase2,
4288 ns->ln_run_time_phase1,
4289 ns->ln_run_time_phase2);
4291 seq_printf(m, "average_speed_phase1: "LPU64" items/sec\n"
4292 "average_speed_phase2: "LPU64" objs/sec\n"
4293 "real_time_speed_phase1: N/A\n"
4294 "real_time_speed_phase2: N/A\n"
4295 "current_position: N/A\n",
4300 up_read(&com->lc_sem);
4304 static int lfsck_namespace_double_scan(const struct lu_env *env,
4305 struct lfsck_component *com)
4307 struct lfsck_namespace *ns = com->lc_file_ram;
4308 struct lfsck_assistant_data *lad = com->lc_data;
4309 struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
4310 struct lfsck_tgt_desc *ltd;
4311 struct lfsck_tgt_desc *next;
4314 rc = lfsck_double_scan_generic(env, com, ns->ln_status);
4315 if (thread_is_stopped(&lad->lad_thread)) {
4316 LASSERT(list_empty(&lad->lad_req_list));
4317 LASSERT(list_empty(&lad->lad_mdt_phase1_list));
4319 spin_lock(<ds->ltd_lock);
4320 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4321 ltd_namespace_phase_list) {
4322 list_del_init(<d->ltd_namespace_phase_list);
4324 spin_unlock(<ds->ltd_lock);
4330 static void lfsck_namespace_data_release(const struct lu_env *env,
4331 struct lfsck_component *com)
4333 struct lfsck_assistant_data *lad = com->lc_data;
4334 struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
4335 struct lfsck_tgt_desc *ltd;
4336 struct lfsck_tgt_desc *next;
4338 LASSERT(lad != NULL);
4339 LASSERT(thread_is_init(&lad->lad_thread) ||
4340 thread_is_stopped(&lad->lad_thread));
4341 LASSERT(list_empty(&lad->lad_req_list));
4343 com->lc_data = NULL;
4344 lfsck_namespace_release_lmv(env, com);
4346 spin_lock(<ds->ltd_lock);
4347 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
4348 ltd_namespace_phase_list) {
4349 list_del_init(<d->ltd_namespace_phase_list);
4351 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4352 ltd_namespace_phase_list) {
4353 list_del_init(<d->ltd_namespace_phase_list);
4355 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_list,
4356 ltd_namespace_list) {
4357 list_del_init(<d->ltd_namespace_list);
4359 spin_unlock(<ds->ltd_lock);
4361 if (likely(lad->lad_bitmap != NULL))
4362 CFS_FREE_BITMAP(lad->lad_bitmap);
4367 static void lfsck_namespace_quit(const struct lu_env *env,
4368 struct lfsck_component *com)
4370 struct lfsck_assistant_data *lad = com->lc_data;
4371 struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
4372 struct lfsck_tgt_desc *ltd;
4373 struct lfsck_tgt_desc *next;
4375 LASSERT(lad != NULL);
4377 lfsck_quit_generic(env, com);
4379 LASSERT(thread_is_init(&lad->lad_thread) ||
4380 thread_is_stopped(&lad->lad_thread));
4381 LASSERT(list_empty(&lad->lad_req_list));
4383 lfsck_namespace_release_lmv(env, com);
4385 spin_lock(<ds->ltd_lock);
4386 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
4387 ltd_namespace_phase_list) {
4388 list_del_init(<d->ltd_namespace_phase_list);
4390 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4391 ltd_namespace_phase_list) {
4392 list_del_init(<d->ltd_namespace_phase_list);
4394 spin_unlock(<ds->ltd_lock);
4397 static int lfsck_namespace_in_notify(const struct lu_env *env,
4398 struct lfsck_component *com,
4399 struct lfsck_request *lr,
4402 struct lfsck_instance *lfsck = com->lc_lfsck;
4403 struct lfsck_namespace *ns = com->lc_file_ram;
4404 struct lfsck_assistant_data *lad = com->lc_data;
4405 struct lfsck_tgt_descs *ltds = &lfsck->li_mdt_descs;
4406 struct lfsck_tgt_desc *ltd;
4411 switch (lr->lr_event) {
4412 case LE_SKIP_NLINK_DECLARE: {
4413 struct dt_object *obj = com->lc_obj;
4414 struct lu_fid *key = &lfsck_env_info(env)->lti_fid3;
4417 LASSERT(th != NULL);
4419 rc = dt_declare_delete(env, obj,
4420 (const struct dt_key *)key, th);
4422 rc = dt_declare_insert(env, obj,
4423 (const struct dt_rec *)&flags,
4424 (const struct dt_key *)key, th);
4428 case LE_SKIP_NLINK: {
4429 struct dt_object *obj = com->lc_obj;
4430 struct lu_fid *key = &lfsck_env_info(env)->lti_fid3;
4435 LASSERT(th != NULL);
4437 fid_cpu_to_be(key, &lr->lr_fid);
4438 rc = dt_lookup(env, obj, (struct dt_rec *)&flags,
4439 (const struct dt_key *)key, BYPASS_CAPA);
4441 if (flags & LNTF_SKIP_NLINK)
4445 } else if (rc != -ENOENT) {
4449 flags |= LNTF_SKIP_NLINK;
4451 rc = dt_delete(env, obj, (const struct dt_key *)key,
4457 rc = dt_insert(env, obj, (const struct dt_rec *)&flags,
4458 (const struct dt_key *)key, th, BYPASS_CAPA, 1);
4463 CDEBUG(D_LFSCK, "%s: RPC service thread mark the "DFID
4464 " to be skipped for namespace double scan: rc = %d\n",
4465 lfsck_lfsck2name(com->lc_lfsck), PFID(&lr->lr_fid), rc);
4468 /* If we cannot record this object in the LFSCK tracing,
4469 * we have to mark the LFSC as LF_INCOMPLETE, then the
4470 * LFSCK will skip nlink attribute verification for
4472 ns->ln_flags |= LF_INCOMPLETE;
4476 case LE_SET_LMV_MASTER: {
4477 struct dt_object *obj;
4479 obj = lfsck_object_find_by_dev(env, lfsck->li_bottom,
4482 RETURN(PTR_ERR(obj));
4484 rc = lfsck_namespace_notify_lmv_master_local(env, com, obj);
4485 lfsck_object_put(env, obj);
4487 RETURN(rc > 0 ? 0 : rc);
4489 case LE_SET_LMV_SLAVE: {
4490 if (!(lr->lr_flags & LEF_RECHECK_NAME_HASH))
4491 ns->ln_striped_shards_repaired++;
4493 rc = lfsck_namespace_trace_update(env, com, &lr->lr_fid,
4494 LNTF_RECHECK_NAME_HASH, true);
4496 RETURN(rc > 0 ? 0 : rc);
4498 case LE_PHASE1_DONE:
4499 case LE_PHASE2_DONE:
4506 CDEBUG(D_LFSCK, "%s: namespace LFSCK handles notify %u from MDT %x, "
4507 "status %d, flags %x\n", lfsck_lfsck2name(lfsck), lr->lr_event,
4508 lr->lr_index, lr->lr_status, lr->lr_flags2);
4510 spin_lock(<ds->ltd_lock);
4511 ltd = LTD_TGT(ltds, lr->lr_index);
4513 spin_unlock(<ds->ltd_lock);
4518 list_del_init(<d->ltd_namespace_phase_list);
4519 switch (lr->lr_event) {
4520 case LE_PHASE1_DONE:
4521 if (lr->lr_status <= 0) {
4522 ltd->ltd_namespace_done = 1;
4523 list_del_init(<d->ltd_namespace_list);
4524 CDEBUG(D_LFSCK, "%s: MDT %x failed/stopped at "
4525 "phase1 for namespace LFSCK: rc = %d.\n",
4526 lfsck_lfsck2name(lfsck),
4527 ltd->ltd_index, lr->lr_status);
4528 ns->ln_flags |= LF_INCOMPLETE;
4533 if (lr->lr_flags2 & LF_INCOMPLETE)
4534 ns->ln_flags |= LF_INCOMPLETE;
4536 if (list_empty(<d->ltd_namespace_list))
4537 list_add_tail(<d->ltd_namespace_list,
4538 &lad->lad_mdt_list);
4539 list_add_tail(<d->ltd_namespace_phase_list,
4540 &lad->lad_mdt_phase2_list);
4542 case LE_PHASE2_DONE:
4543 ltd->ltd_namespace_done = 1;
4544 list_del_init(<d->ltd_namespace_list);
4548 ltd->ltd_namespace_done = 1;
4549 list_del_init(<d->ltd_namespace_list);
4550 if (!(lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT)) {
4552 "%s: the peer MDT %x exit namespace LFSCK\n",
4553 lfsck_lfsck2name(lfsck), ltd->ltd_index);
4554 ns->ln_flags |= LF_INCOMPLETE;
4560 spin_unlock(<ds->ltd_lock);
4562 if (fail && lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) {
4563 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
4565 memset(stop, 0, sizeof(*stop));
4566 stop->ls_status = lr->lr_status;
4567 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
4568 lfsck_stop(env, lfsck->li_bottom, stop);
4569 } else if (lfsck_phase2_next_ready(lad)) {
4570 wake_up_all(&lad->lad_thread.t_ctl_waitq);
4576 static int lfsck_namespace_query(const struct lu_env *env,
4577 struct lfsck_component *com)
4579 struct lfsck_namespace *ns = com->lc_file_ram;
4581 return ns->ln_status;
4584 static struct lfsck_operations lfsck_namespace_ops = {
4585 .lfsck_reset = lfsck_namespace_reset,
4586 .lfsck_fail = lfsck_namespace_fail,
4587 .lfsck_close_dir = lfsck_namespace_close_dir,
4588 .lfsck_open_dir = lfsck_namespace_open_dir,
4589 .lfsck_checkpoint = lfsck_namespace_checkpoint,
4590 .lfsck_prep = lfsck_namespace_prep,
4591 .lfsck_exec_oit = lfsck_namespace_exec_oit,
4592 .lfsck_exec_dir = lfsck_namespace_exec_dir,
4593 .lfsck_post = lfsck_namespace_post,
4594 .lfsck_dump = lfsck_namespace_dump,
4595 .lfsck_double_scan = lfsck_namespace_double_scan,
4596 .lfsck_data_release = lfsck_namespace_data_release,
4597 .lfsck_quit = lfsck_namespace_quit,
4598 .lfsck_in_notify = lfsck_namespace_in_notify,
4599 .lfsck_query = lfsck_namespace_query,
4603 * Repair dangling name entry.
4605 * For the name entry with dangling reference, we need to repare the
4606 * inconsistency according to the LFSCK sponsor's requirement:
4608 * 1) Keep the inconsistency there and report the inconsistency case,
4609 * then give the chance to the application to find related issues,
4610 * and the users can make the decision about how to handle it with
4611 * more human knownledge. (by default)
4613 * 2) Re-create the missing MDT-object with the FID information.
4615 * \param[in] env pointer to the thread context
4616 * \param[in] com pointer to the lfsck component
4617 * \param[in] child pointer to the object corresponding to the dangling
4619 * \param[in] lnr pointer to the namespace request that contains the
4620 * name's name, parent object, parent's LMV, and ect.
4622 * \retval positive number if no need to repair
4623 * \retval zero for repaired successfully
4624 * \retval negative error number on failure
4626 int lfsck_namespace_repair_dangling(const struct lu_env *env,
4627 struct lfsck_component *com,
4628 struct dt_object *child,
4629 struct lfsck_namespace_req *lnr)
4631 struct lfsck_thread_info *info = lfsck_env_info(env);
4632 struct lu_attr *la = &info->lti_la;
4633 struct dt_allocation_hint *hint = &info->lti_hint;
4634 struct dt_object_format *dof = &info->lti_dof;
4635 struct dt_insert_rec *rec = &info->lti_dt_rec;
4636 struct lmv_mds_md_v1 *lmv2 = &info->lti_lmv2;
4637 struct dt_object *parent = lnr->lnr_obj;
4638 const struct lu_name *cname;
4639 struct linkea_data ldata = { NULL };
4640 struct lustre_handle lh = { 0 };
4641 struct lu_buf linkea_buf;
4642 struct lu_buf lmv_buf;
4643 struct lfsck_instance *lfsck = com->lc_lfsck;
4644 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4645 struct dt_device *dev = lfsck_obj2dt_dev(child);
4646 struct thandle *th = NULL;
4648 __u16 type = lnr->lnr_type;
4652 cname = lfsck_name_get_const(env, lnr->lnr_name, lnr->lnr_namelen);
4653 if (bk->lb_param & LPF_CREATE_MDTOBJ)
4658 if (!create || bk->lb_param & LPF_DRYRUN)
4661 rc = linkea_data_new(&ldata, &info->lti_linkea_buf2);
4665 rc = linkea_add_buf(&ldata, cname, lfsck_dto2fid(parent));
4669 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
4670 MDS_INODELOCK_UPDATE, LCK_EX);
4674 rc = lfsck_namespace_check_exist(env, parent, child, lnr->lnr_name);
4678 th = dt_trans_create(env, dev);
4680 GOTO(log, rc = PTR_ERR(th));
4682 /* Set the ctime as zero, then others can know it is created for
4683 * repairing dangling name entry by LFSCK. And if the LFSCK made
4684 * wrong decision and the real MDT-object has been found later,
4685 * then the LFSCK has chance to fix the incosistency properly. */
4686 memset(la, 0, sizeof(*la));
4687 la->la_mode = (type & S_IFMT) | 0600;
4688 la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
4689 LA_ATIME | LA_MTIME | LA_CTIME;
4691 child->do_ops->do_ah_init(env, hint, parent, child,
4692 la->la_mode & S_IFMT);
4694 memset(dof, 0, sizeof(*dof));
4695 dof->dof_type = dt_mode_to_dft(type);
4696 /* If the target is a regular file, then the LFSCK will only create
4697 * the MDT-object without stripes (dof->dof_reg.striped = 0). related
4698 * OST-objects will be created when write open. */
4700 /* 1a. create child. */
4701 rc = dt_declare_create(env, child, la, hint, dof, th);
4705 if (S_ISDIR(type)) {
4706 if (unlikely(!dt_try_as_dir(env, child)))
4707 GOTO(stop, rc = -ENOTDIR);
4709 /* 2a. insert dot into child dir */
4710 rec->rec_type = S_IFDIR;
4711 rec->rec_fid = lfsck_dto2fid(child);
4712 rc = dt_declare_insert(env, child,
4713 (const struct dt_rec *)rec,
4714 (const struct dt_key *)dot, th);
4718 /* 3a. insert dotdot into child dir */
4719 rec->rec_fid = lfsck_dto2fid(parent);
4720 rc = dt_declare_insert(env, child,
4721 (const struct dt_rec *)rec,
4722 (const struct dt_key *)dotdot, th);
4726 /* 4a. increase child nlink */
4727 rc = dt_declare_ref_add(env, child, th);
4731 /* 5a. generate slave LMV EA. */
4732 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
4735 idx = lfsck_shard_name_to_index(env,
4736 lnr->lnr_name, lnr->lnr_namelen,
4737 type, lfsck_dto2fid(child));
4738 if (unlikely(idx < 0))
4739 GOTO(stop, rc = idx);
4741 *lmv2 = lnr->lnr_lmv->ll_lmv;
4742 lmv2->lmv_magic = LMV_MAGIC_STRIPE;
4743 lmv2->lmv_master_mdt_index = idx;
4745 lfsck_lmv_header_cpu_to_le(lmv2, lmv2);
4746 lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
4747 rc = dt_declare_xattr_set(env, child, &lmv_buf,
4748 XATTR_NAME_LMV, 0, th);
4754 /* 6a. insert linkEA for child */
4755 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
4756 ldata.ld_leh->leh_len);
4757 rc = dt_declare_xattr_set(env, child, &linkea_buf,
4758 XATTR_NAME_LINK, 0, th);
4762 rc = dt_trans_start(env, dev, th);
4764 GOTO(stop, rc = (rc == -EEXIST ? 1 : rc));
4766 dt_write_lock(env, child, 0);
4767 /* 1b. create child */
4768 rc = dt_create(env, child, la, hint, dof, th);
4770 GOTO(unlock, rc = (rc == -EEXIST ? 1 : rc));
4772 if (S_ISDIR(type)) {
4773 if (unlikely(!dt_try_as_dir(env, child)))
4774 GOTO(unlock, rc = -ENOTDIR);
4776 /* 2b. insert dot into child dir */
4777 rec->rec_type = S_IFDIR;
4778 rec->rec_fid = lfsck_dto2fid(child);
4779 rc = dt_insert(env, child, (const struct dt_rec *)rec,
4780 (const struct dt_key *)dot, th, BYPASS_CAPA, 1);
4784 /* 3b. insert dotdot into child dir */
4785 rec->rec_fid = lfsck_dto2fid(parent);
4786 rc = dt_insert(env, child, (const struct dt_rec *)rec,
4787 (const struct dt_key *)dotdot, th,
4792 /* 4b. increase child nlink */
4793 rc = dt_ref_add(env, child, th);
4797 /* 5b. generate slave LMV EA. */
4798 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
4799 rc = dt_xattr_set(env, child, &lmv_buf, XATTR_NAME_LMV,
4800 0, th, BYPASS_CAPA);
4806 /* 6b. insert linkEA for child. */
4807 rc = dt_xattr_set(env, child, &linkea_buf,
4808 XATTR_NAME_LINK, 0, th, BYPASS_CAPA);
4813 dt_write_unlock(env, child);
4816 dt_trans_stop(env, dev, th);
4819 lfsck_ibits_unlock(&lh, LCK_EX);
4820 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found dangling "
4821 "reference for: parent "DFID", child "DFID", type %u, "
4822 "name %s. %s: rc = %d\n", lfsck_lfsck2name(lfsck),
4823 PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)),
4824 type, cname->ln_name,
4825 create ? "Create the lost OST-object as required" :
4826 "Keep the MDT-object there by default", rc);
4829 struct lfsck_namespace *ns = com->lc_file_ram;
4831 ns->ln_flags |= LF_INCONSISTENT;
4837 static int lfsck_namespace_assistant_handler_p1(const struct lu_env *env,
4838 struct lfsck_component *com,
4839 struct lfsck_assistant_req *lar)
4841 struct lfsck_thread_info *info = lfsck_env_info(env);
4842 struct lu_attr *la = &info->lti_la;
4843 struct lfsck_instance *lfsck = com->lc_lfsck;
4844 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4845 struct lfsck_namespace *ns = com->lc_file_ram;
4846 struct linkea_data ldata = { NULL };
4847 const struct lu_name *cname;
4848 struct thandle *handle = NULL;
4849 struct lfsck_namespace_req *lnr =
4850 container_of0(lar, struct lfsck_namespace_req, lnr_lar);
4851 struct dt_object *dir = lnr->lnr_obj;
4852 struct dt_object *obj = NULL;
4853 const struct lu_fid *pfid = lfsck_dto2fid(dir);
4854 struct dt_device *dev = NULL;
4855 struct lustre_handle lh = { 0 };
4856 bool repaired = false;
4857 bool dtlocked = false;
4861 bool bad_hash = false;
4865 enum lfsck_namespace_inconsistency_type type = LNIT_NONE;
4868 if (lnr->lnr_attr & LUDA_UPGRADE) {
4869 ns->ln_flags |= LF_UPGRADE;
4870 ns->ln_dirent_repaired++;
4872 } else if (lnr->lnr_attr & LUDA_REPAIR) {
4873 ns->ln_flags |= LF_INCONSISTENT;
4874 ns->ln_dirent_repaired++;
4878 if (unlikely(fid_is_zero(&lnr->lnr_fid))) {
4879 if (strcmp(lnr->lnr_name, dotdot) != 0)
4882 rc = lfsck_namespace_trace_update(env, com, pfid,
4883 LNTF_CHECK_PARENT, true);
4888 if (unlikely(lnr->lnr_dir_cookie == MDS_DIR_END_OFF)) {
4889 rc = lfsck_namespace_striped_dir_rescan(env, com, lnr);
4894 if (lnr->lnr_name[0] == '.' &&
4895 (lnr->lnr_namelen == 1 || fid_seq_is_dot(fid_seq(&lnr->lnr_fid))))
4898 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
4899 rc = lfsck_namespace_handle_striped_master(env, com, lnr);
4904 idx = lfsck_find_mdt_idx_by_fid(env, lfsck, &lnr->lnr_fid);
4906 GOTO(out, rc = idx);
4908 if (idx == lfsck_dev_idx(lfsck->li_bottom)) {
4909 if (unlikely(strcmp(lnr->lnr_name, dotdot) == 0))
4912 dev = lfsck->li_next;
4914 struct lfsck_tgt_desc *ltd;
4916 /* Usually, some local filesystem consistency verification
4917 * tools can guarantee the local namespace tree consistenct.
4918 * So the LFSCK will only verify the remote directory. */
4919 if (unlikely(strcmp(lnr->lnr_name, dotdot) == 0)) {
4920 rc = lfsck_namespace_trace_update(env, com, pfid,
4921 LNTF_CHECK_PARENT, true);
4926 ltd = LTD_TGT(&lfsck->li_mdt_descs, idx);
4927 if (unlikely(ltd == NULL)) {
4928 CDEBUG(D_LFSCK, "%s: cannot talk with MDT %x which "
4929 "did not join the namespace LFSCK\n",
4930 lfsck_lfsck2name(lfsck), idx);
4931 lfsck_lad_set_bitmap(env, com, idx);
4933 GOTO(out, rc = -ENODEV);
4939 obj = lfsck_object_find_by_dev(env, dev, &lnr->lnr_fid);
4941 GOTO(out, rc = PTR_ERR(obj));
4943 cname = lfsck_name_get_const(env, lnr->lnr_name, lnr->lnr_namelen);
4944 if (dt_object_exists(obj) == 0) {
4947 rc = lfsck_namespace_check_exist(env, dir, obj, lnr->lnr_name);
4949 if (!lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
4950 lnr->lnr_name, lnr->lnr_namelen)) {
4951 type = LNIT_BAD_DIRENT;
4956 type = LNIT_DANGLING;
4957 rc = lfsck_namespace_repair_dangling(env, com,
4966 if (!(bk->lb_param & LPF_DRYRUN) && repaired) {
4969 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
4970 MDS_INODELOCK_UPDATE |
4971 MDS_INODELOCK_XATTR, LCK_EX);
4975 handle = dt_trans_create(env, dev);
4977 GOTO(out, rc = PTR_ERR(handle));
4979 rc = lfsck_declare_namespace_exec_dir(env, obj, handle);
4983 rc = dt_trans_start(env, dev, handle);
4987 dt_write_lock(env, obj, 0);
4991 rc = lfsck_namespace_check_exist(env, dir, obj, lnr->lnr_name);
4995 rc = lfsck_links_read(env, obj, &ldata);
4996 if (unlikely(rc == -ENOENT)) {
4997 if (handle != NULL) {
4998 dt_write_unlock(env, obj);
5001 dt_trans_stop(env, dev, handle);
5004 lfsck_ibits_unlock(&lh, LCK_EX);
5007 /* It may happen when the remote object has been removed,
5008 * but the local MDT is not aware of that. */
5010 } else if (rc == 0) {
5011 count = ldata.ld_leh->leh_reccount;
5012 rc = linkea_links_find(&ldata, cname, pfid);
5014 (count == 1 || !S_ISDIR(lfsck_object_type(obj)))) {
5015 if ((lfsck_object_type(obj) & S_IFMT) !=
5017 ns->ln_flags |= LF_INCONSISTENT;
5018 type = LNIT_BAD_TYPE;
5024 ns->ln_flags |= LF_INCONSISTENT;
5026 /* If the name entry hash does not match the slave striped
5027 * directory, and the name entry does not match also, then
5028 * it is quite possible that name entry is corrupted. */
5029 if (!lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5030 lnr->lnr_name, lnr->lnr_namelen)) {
5031 type = LNIT_BAD_DIRENT;
5036 /* If the file type stored in the name entry does not match
5037 * the file type claimed by the object, and the object does
5038 * not recognize the name entry, then it is quite possible
5039 * that the name entry is corrupted. */
5040 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type) {
5041 type = LNIT_BAD_DIRENT;
5046 /* For sub-dir object, we cannot make sure whether the sub-dir
5047 * back references the parent via ".." name entry correctly or
5048 * not in the LFSCK first-stage scanning. It may be that the
5049 * (remote) sub-dir ".." name entry has no parent FID after
5050 * file-level backup/restore and its linkEA may be wrong.
5051 * So under such case, we should replace the linkEA according
5052 * to current name entry. But this needs to be done during the
5053 * LFSCK second-stage scanning. The LFSCK will record the name
5054 * entry for further possible using. */
5058 } else if (unlikely(rc == -EINVAL)) {
5059 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type)
5060 type = LNIT_BAD_TYPE;
5063 ns->ln_flags |= LF_INCONSISTENT;
5064 /* The magic crashed, we are not sure whether there are more
5065 * corrupt data in the linkea, so remove all linkea entries. */
5069 } else if (rc == -ENODATA) {
5070 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type)
5071 type = LNIT_BAD_TYPE;
5074 ns->ln_flags |= LF_UPGRADE;
5079 if (bk->lb_param & LPF_DRYRUN) {
5080 ns->ln_linkea_repaired++;
5086 if (!lustre_handle_is_used(&lh))
5092 rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, handle,
5099 rc = linkea_data_new(&ldata,
5100 &lfsck_env_info(env)->lti_linkea_buf);
5105 rc = linkea_add_buf(&ldata, cname, pfid);
5109 rc = lfsck_links_write(env, obj, &ldata, handle);
5110 if (unlikely(rc == -ENOSPC) &&
5111 S_ISREG(lfsck_object_type(obj)) && !dt_object_remote(obj)) {
5112 if (handle != NULL) {
5113 LASSERT(dt_write_locked(env, obj));
5115 dt_write_unlock(env, obj);
5118 dt_trans_stop(env, dev, handle);
5121 lfsck_ibits_unlock(&lh, LCK_EX);
5124 rc = lfsck_namespace_trace_update(env, com,
5125 &lnr->lnr_fid, LNTF_SKIP_NLINK, true);
5127 /* If we cannot record this object in the
5128 * LFSCK tracing, we have to mark the LFSCK
5129 * as LF_INCOMPLETE, then the LFSCK will
5130 * skip nlink attribute verification for
5132 ns->ln_flags |= LF_INCOMPLETE;
5140 count = ldata.ld_leh->leh_reccount;
5141 if (!S_ISDIR(lfsck_object_type(obj)) ||
5142 !dt_object_remote(obj)) {
5143 ns->ln_linkea_repaired++;
5154 rc = dt_attr_get(env, obj, la, BYPASS_CAPA);
5158 if ((count == 1 && la->la_nlink == 1) ||
5159 S_ISDIR(lfsck_object_type(obj)))
5160 /* Usually, it is for single linked object or dir, do nothing.*/
5163 /* Following modification will be in another transaction. */
5164 if (handle != NULL) {
5165 dt_write_unlock(env, obj);
5168 dt_trans_stop(env, dev, handle);
5171 lfsck_ibits_unlock(&lh, LCK_EX);
5174 ns->ln_mul_linked_checked++;
5175 rc = lfsck_namespace_trace_update(env, com, &lnr->lnr_fid,
5176 LNTF_CHECK_LINKEA, true);
5182 dt_write_unlock(env, obj);
5184 if (handle != NULL && !IS_ERR(handle))
5185 dt_trans_stop(env, dev, handle);
5188 lfsck_ibits_unlock(&lh, LCK_EX);
5190 if (!name_is_dot_or_dotdot(lnr->lnr_name, lnr->lnr_namelen) &&
5191 !lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5192 lnr->lnr_name, lnr->lnr_namelen) &&
5193 type != LNIT_BAD_DIRENT) {
5194 ns->ln_flags |= LF_INCONSISTENT;
5197 rc = lfsck_namespace_repair_bad_name_hash(env, com, dir,
5198 lnr->lnr_lmv, lnr->lnr_name);
5207 rc = lfsck_namespace_repair_dirent(env, com, dir,
5208 obj, lnr->lnr_name, lnr->lnr_name,
5209 lnr->lnr_type, true, false);
5213 case LNIT_BAD_DIRENT:
5215 /* XXX: This is a bad dirent, we do not know whether
5216 * the original name entry reference a regular
5217 * file or a directory, then keep the parent's
5218 * nlink count unchanged here. */
5219 rc = lfsck_namespace_repair_dirent(env, com, dir,
5220 obj, lnr->lnr_name, lnr->lnr_name,
5221 lnr->lnr_type, false, false);
5230 down_write(&com->lc_sem);
5232 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail to handle "
5233 "the entry: "DFID", parent "DFID", name %.*s: rc = %d\n",
5234 lfsck_lfsck2name(lfsck), PFID(&lnr->lnr_fid),
5235 PFID(lfsck_dto2fid(lnr->lnr_obj)),
5236 lnr->lnr_namelen, lnr->lnr_name, rc);
5238 lfsck_namespace_record_failure(env, lfsck, ns);
5239 if ((rc == -ENOTCONN || rc == -ESHUTDOWN || rc == -EREMCHG ||
5240 rc == -ETIMEDOUT || rc == -EHOSTDOWN ||
5241 rc == -EHOSTUNREACH || rc == -EINPROGRESS) &&
5242 dev != NULL && dev != lfsck->li_next)
5243 lfsck_lad_set_bitmap(env, com, idx);
5245 if (!(bk->lb_param & LPF_FAILOUT))
5249 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant "
5250 "repaired the entry: "DFID", parent "DFID
5251 ", name %.*s\n", lfsck_lfsck2name(lfsck),
5252 PFID(&lnr->lnr_fid),
5253 PFID(lfsck_dto2fid(lnr->lnr_obj)),
5254 lnr->lnr_namelen, lnr->lnr_name);
5257 ns->ln_items_repaired++;
5261 ns->ln_dangling_repaired++;
5264 ns->ln_bad_type_repaired++;
5266 case LNIT_BAD_DIRENT:
5267 ns->ln_dirent_repaired++;
5273 if (bk->lb_param & LPF_DRYRUN &&
5274 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent))
5275 lfsck_pos_fill(env, lfsck,
5276 &ns->ln_pos_first_inconsistent,
5281 ns->ln_name_hash_repaired++;
5283 /* Not count repeatedly. */
5285 ns->ln_items_repaired++;
5287 if (bk->lb_param & LPF_DRYRUN &&
5288 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent))
5289 lfsck_pos_fill(env, lfsck,
5290 &ns->ln_pos_first_inconsistent,
5297 up_write(&com->lc_sem);
5299 if (obj != NULL && !IS_ERR(obj))
5300 lfsck_object_put(env, obj);
5306 * Handle one orphan under the backend /lost+found directory
5308 * Insert the orphan FID into the namespace LFSCK trace file for further
5309 * processing (via the subsequent namespace LFSCK second-stage scanning).
5310 * At the same time, remove the orphan name entry from backend /lost+found
5311 * directory. There is an interval between the orphan name entry removed
5312 * from the backend /lost+found directory and the orphan FID in the LFSCK
5313 * trace file handled. In such interval, the LFSCK can be reset, then
5314 * all the FIDs recorded in the namespace LFSCK trace file will be dropped.
5315 * To guarantee that the orphans can be found when LFSCK run next time
5316 * without e2fsck again, when remove the orphan name entry, the LFSCK
5317 * will set the orphan's ctime attribute as 1. Since normal applications
5318 * cannot change the object's ctime attribute as 1. Then when LFSCK run
5319 * next time, it can record the object (that ctime is 1) in the namespace
5320 * LFSCK trace file during the first-stage scanning.
5322 * \param[in] env pointer to the thread context
5323 * \param[in] com pointer to the lfsck component
5324 * \param[in] parent pointer to the object for the backend /lost+found
5325 * \param[in] ent pointer to the name entry for the target under the
5326 * backend /lost+found
5328 * \retval positive for repaired
5329 * \retval 0 if needs to repair nothing
5330 * \retval negative error number on failure
5332 static int lfsck_namespace_scan_local_lpf_one(const struct lu_env *env,
5333 struct lfsck_component *com,
5334 struct dt_object *parent,
5335 struct lu_dirent *ent)
5337 struct lfsck_thread_info *info = lfsck_env_info(env);
5338 struct lu_fid *key = &info->lti_fid;
5339 struct lu_attr *la = &info->lti_la;
5340 struct lfsck_instance *lfsck = com->lc_lfsck;
5341 struct dt_object *obj = com->lc_obj;
5342 struct dt_device *dev = lfsck->li_bottom;
5343 struct dt_object *child = NULL;
5344 struct thandle *th = NULL;
5350 child = lfsck_object_find_by_dev(env, dev, &ent->lde_fid);
5352 RETURN(PTR_ERR(child));
5354 LASSERT(dt_object_exists(child));
5355 LASSERT(!dt_object_remote(child));
5357 fid_cpu_to_be(key, &ent->lde_fid);
5358 rc = dt_lookup(env, obj, (struct dt_rec *)&flags,
5359 (const struct dt_key *)key, BYPASS_CAPA);
5362 flags |= LNTF_CHECK_ORPHAN;
5363 } else if (rc == -ENOENT) {
5364 flags = LNTF_CHECK_ORPHAN;
5369 th = dt_trans_create(env, dev);
5371 GOTO(out, rc = PTR_ERR(th));
5373 /* a1. remove name entry from backend /lost+found */
5374 rc = dt_declare_delete(env, parent,
5375 (const struct dt_key *)ent->lde_name, th);
5379 if (S_ISDIR(lfsck_object_type(child))) {
5380 /* a2. decrease parent's nlink */
5381 rc = dt_declare_ref_del(env, parent, th);
5387 /* a3. remove child's FID from the LFSCK trace file. */
5388 rc = dt_declare_delete(env, obj,
5389 (const struct dt_key *)key, th);
5393 /* a4. set child's ctime as 1 */
5394 memset(la, 0, sizeof(*la));
5396 la->la_valid = LA_CTIME;
5397 rc = dt_declare_attr_set(env, child, la, th);
5402 /* a5. insert child's FID into the LFSCK trace file. */
5403 rc = dt_declare_insert(env, obj, (const struct dt_rec *)&flags,
5404 (const struct dt_key *)key, th);
5408 rc = dt_trans_start_local(env, dev, th);
5412 /* b1. remove name entry from backend /lost+found */
5413 rc = dt_delete(env, parent, (const struct dt_key *)ent->lde_name, th,
5418 if (S_ISDIR(lfsck_object_type(child))) {
5419 /* b2. decrease parent's nlink */
5420 dt_write_lock(env, parent, 0);
5421 rc = dt_ref_del(env, parent, th);
5422 dt_write_unlock(env, parent);
5428 /* a3. remove child's FID from the LFSCK trace file. */
5429 rc = dt_delete(env, obj, (const struct dt_key *)key, th,
5434 /* b4. set child's ctime as 1 */
5435 rc = dt_attr_set(env, child, la, th, BYPASS_CAPA);
5440 /* b5. insert child's FID into the LFSCK trace file. */
5441 rc = dt_insert(env, obj, (const struct dt_rec *)&flags,
5442 (const struct dt_key *)key, th, BYPASS_CAPA, 1);
5444 GOTO(stop, rc = (rc == 0 ? 1 : rc));
5447 dt_trans_stop(env, dev, th);
5450 lu_object_put(env, &child->do_lu);
5456 * Handle orphans under the backend /lost+found directory
5458 * Some backend checker, such as e2fsck for ldiskfs may find some orphans
5459 * and put them under the backend /lost+found directory that is invisible
5460 * to client. The LFSCK will scan such directory, for the original client
5461 * visible orphans, add their fids into the namespace LFSCK trace file,
5462 * then the subsenquent namespace LFSCK second-stage scanning can handle
5463 * them as other objects to be double scanned: either move back to normal
5464 * namespace, or to the global visible orphan directory:
5465 * /ROOT/.lustre/lost+found/MDTxxxx/
5467 * \param[in] env pointer to the thread context
5468 * \param[in] com pointer to the lfsck component
5470 static void lfsck_namespace_scan_local_lpf(const struct lu_env *env,
5471 struct lfsck_component *com)
5473 struct lfsck_thread_info *info = lfsck_env_info(env);
5474 struct lu_dirent *ent =
5475 (struct lu_dirent *)info->lti_key;
5476 struct lu_seq_range *range = &info->lti_range;
5477 struct lfsck_instance *lfsck = com->lc_lfsck;
5478 struct ptlrpc_thread *thread = &lfsck->li_thread;
5479 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5480 struct dt_device *dev = lfsck->li_bottom;
5481 struct lfsck_namespace *ns = com->lc_file_ram;
5482 struct dt_object *parent;
5483 const struct dt_it_ops *iops;
5485 struct seq_server_site *ss =
5486 lu_site2seq(dev->dd_lu_dev.ld_site);
5492 parent = lfsck_object_find_by_dev(env, dev, &LU_BACKEND_LPF_FID);
5493 if (IS_ERR(parent)) {
5494 CERROR("%s: fail to find backend /lost+found: rc = %ld\n",
5495 lfsck_lfsck2name(lfsck), PTR_ERR(parent));
5499 /* It is normal that the /lost+found does not exist for ZFS backend. */
5500 if (!dt_object_exists(parent))
5503 if (unlikely(!dt_try_as_dir(env, parent)))
5504 GOTO(out, rc = -ENOTDIR);
5506 CDEBUG(D_LFSCK, "%s: start to scan backend /lost+found\n",
5507 lfsck_lfsck2name(lfsck));
5509 com->lc_new_scanned = 0;
5510 iops = &parent->do_index_ops->dio_it;
5511 di = iops->init(env, parent, LUDA_64BITHASH | LUDA_TYPE, BYPASS_CAPA);
5513 GOTO(out, rc = PTR_ERR(di));
5515 rc = iops->load(env, di, 0);
5517 rc = iops->next(env, di);
5522 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY3) &&
5524 struct l_wait_info lwi;
5526 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val),
5528 l_wait_event(thread->t_ctl_waitq,
5529 !thread_is_running(thread),
5532 if (unlikely(!thread_is_running(thread)))
5536 rc = iops->rec(env, di, (struct dt_rec *)ent,
5537 LUDA_64BITHASH | LUDA_TYPE);
5539 rc = lfsck_unpack_ent(ent, &cookie, &type);
5541 if (unlikely(rc != 0)) {
5542 CDEBUG(D_LFSCK, "%s: fail to iterate backend "
5543 "/lost+found: rc = %d\n",
5544 lfsck_lfsck2name(lfsck), rc);
5549 /* skip dot and dotdot entries */
5550 if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen))
5553 if (!fid_seq_in_fldb(fid_seq(&ent->lde_fid)))
5556 if (fid_is_norm(&ent->lde_fid)) {
5557 fld_range_set_mdt(range);
5558 rc = fld_local_lookup(env, ss->ss_server_fld,
5559 fid_seq(&ent->lde_fid), range);
5562 } else if (lfsck_dev_idx(dev) != 0) {
5563 /* If the returned FID is IGIF, then there are three
5566 * 1) The object is upgraded from old Lustre-1.8 with
5567 * IGIF assigned to such object.
5568 * 2) The object is a backend local object and is
5569 * invisible to client.
5570 * 3) The object lost its LMV EA, and since there is
5571 * no FID-in-dirent for the orphan in the backend
5572 * /lost+found directory, then the low layer will
5573 * return IGIF for such object.
5575 * For MDTx (x != 0), it is either case 2) or case 3),
5576 * but from the LFSCK view, they are indistinguishable.
5577 * To be safe, the LFSCK will keep it there and report
5578 * some message, then the adminstrator can handle that
5581 * For MDT0, it is more possible the case 1). The LFSCK
5582 * will handle the orphan as an upgraded object. */
5583 CDEBUG(D_LFSCK, "%s: the orphan %.*s with IGIF "DFID
5584 "in the backend /lost+found on the MDT %04x, "
5585 "to be safe, skip it.\n",
5586 lfsck_lfsck2name(lfsck), ent->lde_namelen,
5587 ent->lde_name, PFID(&ent->lde_fid),
5588 lfsck_dev_idx(dev));
5592 rc = lfsck_namespace_scan_local_lpf_one(env, com, parent, ent);
5595 down_write(&com->lc_sem);
5596 com->lc_new_scanned++;
5597 ns->ln_local_lpf_scanned++;
5599 ns->ln_local_lpf_moved++;
5601 ns->ln_local_lpf_skipped++;
5603 ns->ln_local_lpf_failed++;
5604 up_write(&com->lc_sem);
5606 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
5610 lfsck_control_speed_by_self(com);
5611 if (unlikely(!thread_is_running(thread))) {
5616 rc = iops->next(env, di);
5620 iops->fini(env, di);
5625 CDEBUG(D_LFSCK, "%s: stop to scan backend /lost+found: rc = %d\n",
5626 lfsck_lfsck2name(lfsck), rc);
5628 lu_object_put(env, &parent->do_lu);
5632 * Rescan the striped directory after the master LMV EA reset.
5634 * Sometimes, the master LMV EA of the striped directory maybe lost, so when
5635 * the namespace LFSCK engine scan the striped directory for the first time,
5636 * it will be regarded as a normal directory. As the LFSCK processing, some
5637 * other LFSCK instance on other MDT will find the shard of this striped dir,
5638 * and find that the master MDT-object of the striped directory lost its LMV
5639 * EA, then such remote LFSCK instance will regenerate the master LMV EA and
5640 * notify the LFSCK instance on this MDT to rescan the striped directory.
5642 * \param[in] env pointer to the thread context
5643 * \param[in] com pointer to the lfsck component
5644 * \param[in] llu the lfsck_lmv_unit that contains the striped directory
5647 * \retval positive number for success
5648 * \retval 0 for LFSCK stopped/paused
5649 * \retval negative error number on failure
5651 static int lfsck_namespace_rescan_striped_dir(const struct lu_env *env,
5652 struct lfsck_component *com,
5653 struct lfsck_lmv_unit *llu)
5655 struct lfsck_thread_info *info = lfsck_env_info(env);
5656 struct lfsck_instance *lfsck = com->lc_lfsck;
5657 struct lfsck_assistant_data *lad = com->lc_data;
5658 struct dt_object *dir;
5659 const struct dt_it_ops *iops;
5661 struct lu_dirent *ent =
5662 (struct lu_dirent *)info->lti_key;
5663 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5664 struct ptlrpc_thread *thread = &lfsck->li_thread;
5665 struct lfsck_namespace_req *lnr;
5666 struct lfsck_assistant_req *lar;
5671 LASSERT(list_empty(&lad->lad_req_list));
5673 lfsck->li_lmv = &llu->llu_lmv;
5674 lfsck->li_obj_dir = lfsck_object_get(llu->llu_obj);
5675 rc = lfsck_open_dir(env, lfsck, 0);
5679 dir = lfsck->li_obj_dir;
5680 di = lfsck->li_di_dir;
5681 iops = &dir->do_index_ops->dio_it;
5683 rc = iops->rec(env, di, (struct dt_rec *)ent,
5684 lfsck->li_args_dir);
5686 rc = lfsck_unpack_ent(ent, &lfsck->li_cookie_dir,
5690 if (bk->lb_param & LPF_FAILOUT)
5696 if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen))
5699 lnr = lfsck_namespace_assistant_req_init(lfsck, ent, type);
5701 if (bk->lb_param & LPF_FAILOUT)
5702 GOTO(out, rc = PTR_ERR(lnr));
5707 lar = &lnr->lnr_lar;
5708 rc = lfsck_namespace_assistant_handler_p1(env, com, lar);
5709 lfsck_namespace_assistant_req_fini(env, lar);
5710 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
5713 if (unlikely(!thread_is_running(thread)))
5717 rc = iops->next(env, di);
5721 lfsck_close_dir(env, lfsck, rc);
5725 /* The close_dir() may insert a dummy lnr in the lad->lad_req_list. */
5726 if (list_empty(&lad->lad_req_list))
5729 spin_lock(&lad->lad_lock);
5730 lar = list_entry(lad->lad_req_list.next, struct lfsck_assistant_req,
5732 list_del_init(&lar->lar_list);
5733 spin_unlock(&lad->lad_lock);
5735 rc = lfsck_namespace_assistant_handler_p1(env, com, lar);
5736 lfsck_namespace_assistant_req_fini(env, lar);
5738 RETURN(rc == 0 ? 1 : rc);
5741 static int lfsck_namespace_assistant_handler_p2(const struct lu_env *env,
5742 struct lfsck_component *com)
5744 struct lfsck_instance *lfsck = com->lc_lfsck;
5745 struct ptlrpc_thread *thread = &lfsck->li_thread;
5746 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5747 struct lfsck_namespace *ns = com->lc_file_ram;
5748 struct dt_object *obj = com->lc_obj;
5749 const struct dt_it_ops *iops = &obj->do_index_ops->dio_it;
5750 struct dt_object *target;
5758 while (!list_empty(&lfsck->li_list_lmv)) {
5759 struct lfsck_lmv_unit *llu;
5761 spin_lock(&lfsck->li_lock);
5762 llu = list_entry(lfsck->li_list_lmv.next,
5763 struct lfsck_lmv_unit, llu_link);
5764 list_del_init(&llu->llu_link);
5765 spin_unlock(&lfsck->li_lock);
5767 rc = lfsck_namespace_rescan_striped_dir(env, com, llu);
5772 CDEBUG(D_LFSCK, "%s: namespace LFSCK phase2 scan start\n",
5773 lfsck_lfsck2name(lfsck));
5775 lfsck_namespace_scan_local_lpf(env, com);
5777 com->lc_new_checked = 0;
5778 com->lc_new_scanned = 0;
5779 com->lc_time_last_checkpoint = cfs_time_current();
5780 com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
5781 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
5783 di = iops->init(env, obj, 0, BYPASS_CAPA);
5785 RETURN(PTR_ERR(di));
5787 fid_cpu_to_be(&fid, &ns->ln_fid_latest_scanned_phase2);
5788 rc = iops->get(env, di, (const struct dt_key *)&fid);
5792 /* Skip the start one, which either has been processed or non-exist. */
5793 rc = iops->next(env, di);
5798 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY3) &&
5800 struct l_wait_info lwi;
5802 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val),
5804 l_wait_event(thread->t_ctl_waitq,
5805 !thread_is_running(thread),
5808 if (unlikely(!thread_is_running(thread)))
5812 key = iops->key(env, di);
5813 fid_be_to_cpu(&fid, (const struct lu_fid *)key);
5814 if (!fid_is_sane(&fid)) {
5819 target = lfsck_object_find_by_dev(env, lfsck->li_bottom, &fid);
5820 if (IS_ERR(target)) {
5821 rc = PTR_ERR(target);
5825 if (dt_object_exists(target)) {
5826 rc = iops->rec(env, di, (struct dt_rec *)&flags, 0);
5828 rc = lfsck_namespace_double_scan_one(env, com,
5835 lfsck_object_put(env, target);
5838 down_write(&com->lc_sem);
5839 com->lc_new_checked++;
5840 com->lc_new_scanned++;
5841 ns->ln_fid_latest_scanned_phase2 = fid;
5843 ns->ln_objs_repaired_phase2++;
5845 ns->ln_objs_failed_phase2++;
5846 up_write(&com->lc_sem);
5848 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
5851 if (unlikely(cfs_time_beforeq(com->lc_time_next_checkpoint,
5852 cfs_time_current())) &&
5853 com->lc_new_checked != 0) {
5854 down_write(&com->lc_sem);
5855 ns->ln_run_time_phase2 +=
5856 cfs_duration_sec(cfs_time_current() +
5857 HALF_SEC - com->lc_time_last_checkpoint);
5858 ns->ln_time_last_checkpoint = cfs_time_current_sec();
5859 ns->ln_objs_checked_phase2 += com->lc_new_checked;
5860 com->lc_new_checked = 0;
5861 rc = lfsck_namespace_store(env, com, false);
5862 up_write(&com->lc_sem);
5866 com->lc_time_last_checkpoint = cfs_time_current();
5867 com->lc_time_next_checkpoint =
5868 com->lc_time_last_checkpoint +
5869 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
5872 lfsck_control_speed_by_self(com);
5873 if (unlikely(!thread_is_running(thread)))
5876 rc = iops->next(env, di);
5885 iops->fini(env, di);
5887 CDEBUG(D_LFSCK, "%s: namespace LFSCK phase2 scan stop: rc = %d\n",
5888 lfsck_lfsck2name(lfsck), rc);
5893 static void lfsck_namespace_assistant_fill_pos(const struct lu_env *env,
5894 struct lfsck_component *com,
5895 struct lfsck_position *pos)
5897 struct lfsck_assistant_data *lad = com->lc_data;
5898 struct lfsck_namespace_req *lnr;
5900 if (list_empty(&lad->lad_req_list))
5903 lnr = list_entry(lad->lad_req_list.next,
5904 struct lfsck_namespace_req,
5906 pos->lp_oit_cookie = lnr->lnr_oit_cookie;
5907 pos->lp_dir_cookie = lnr->lnr_dir_cookie - 1;
5908 pos->lp_dir_parent = *lfsck_dto2fid(lnr->lnr_obj);
5911 static int lfsck_namespace_double_scan_result(const struct lu_env *env,
5912 struct lfsck_component *com,
5915 struct lfsck_instance *lfsck = com->lc_lfsck;
5916 struct lfsck_namespace *ns = com->lc_file_ram;
5918 down_write(&com->lc_sem);
5919 ns->ln_run_time_phase2 += cfs_duration_sec(cfs_time_current() +
5920 HALF_SEC - lfsck->li_time_last_checkpoint);
5921 ns->ln_time_last_checkpoint = cfs_time_current_sec();
5922 ns->ln_objs_checked_phase2 += com->lc_new_checked;
5923 com->lc_new_checked = 0;
5926 if (ns->ln_flags & LF_INCOMPLETE)
5927 ns->ln_status = LS_PARTIAL;
5929 ns->ln_status = LS_COMPLETED;
5930 if (!(lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN))
5931 ns->ln_flags &= ~(LF_SCANNED_ONCE | LF_INCONSISTENT);
5932 ns->ln_time_last_complete = ns->ln_time_last_checkpoint;
5933 ns->ln_success_count++;
5934 } else if (rc == 0) {
5935 if (lfsck->li_status != 0)
5936 ns->ln_status = lfsck->li_status;
5938 ns->ln_status = LS_STOPPED;
5940 ns->ln_status = LS_FAILED;
5943 rc = lfsck_namespace_store(env, com, false);
5944 up_write(&com->lc_sem);
5950 lfsck_namespace_assistant_sync_failures_interpret(const struct lu_env *env,
5951 struct ptlrpc_request *req,
5958 * Notify remote LFSCK instances about former failures.
5960 * The local LFSCK instance has recorded which MDTs have ever failed to respond
5961 * some LFSCK verification requests (maybe because of network issues or the MDT
5962 * itself trouble). During the respond gap the MDT may missed some name entries
5963 * verification, then the MDT cannot know whether related MDT-objects have been
5964 * referenced by related name entries or not, then in the second-stage scanning,
5965 * these MDT-objects will be regarded as orphan, if the MDT-object contains bad
5966 * linkEA for back reference, then it will misguide the LFSCK to generate wrong
5967 * name entry for repairing the orphan.
5969 * To avoid above trouble, when layout LFSCK finishes the first-stage scanning,
5970 * it will scan the bitmap for the ever failed MDTs, and notify them that they
5971 * have ever missed some name entries verification and should skip the handling
5972 * for orphan MDT-objects.
5974 * \param[in] env pointer to the thread context
5975 * \param[in] com pointer to the lfsck component
5976 * \param[in] lr pointer to the lfsck request
5978 static void lfsck_namespace_assistant_sync_failures(const struct lu_env *env,
5979 struct lfsck_component *com,
5980 struct lfsck_request *lr)
5982 struct lfsck_async_interpret_args *laia =
5983 &lfsck_env_info(env)->lti_laia2;
5984 struct lfsck_assistant_data *lad = com->lc_data;
5985 struct lfsck_namespace *ns = com->lc_file_ram;
5986 struct lfsck_instance *lfsck = com->lc_lfsck;
5987 struct lfsck_tgt_descs *ltds = &lfsck->li_mdt_descs;
5988 struct lfsck_tgt_desc *ltd;
5989 struct ptlrpc_request_set *set;
5993 set = ptlrpc_prep_set();
5995 GOTO(out, rc = -ENOMEM);
5997 lr->lr_flags2 = ns->ln_flags | LF_INCOMPLETE;
5998 memset(laia, 0, sizeof(*laia));
5999 lad->lad_touch_gen++;
6001 spin_lock(<ds->ltd_lock);
6002 while (!list_empty(&lad->lad_mdt_list)) {
6003 ltd = list_entry(lad->lad_mdt_list.next,
6004 struct lfsck_tgt_desc,
6005 ltd_namespace_list);
6006 if (ltd->ltd_namespace_gen == lad->lad_touch_gen)
6009 ltd->ltd_namespace_gen = lad->lad_touch_gen;
6010 list_move_tail(<d->ltd_namespace_list,
6011 &lad->lad_mdt_list);
6012 if (!lad->lad_incomplete ||
6013 !cfs_bitmap_check(lad->lad_bitmap, ltd->ltd_index)) {
6014 ltd->ltd_namespace_failed = 0;
6018 ltd->ltd_namespace_failed = 1;
6019 spin_unlock(<ds->ltd_lock);
6020 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
6021 lfsck_namespace_assistant_sync_failures_interpret,
6022 laia, LFSCK_NOTIFY);
6024 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail "
6025 "to sync failure with MDT %x: rc = %d\n",
6026 lfsck_lfsck2name(lfsck), ltd->ltd_index, rc);
6028 spin_lock(<ds->ltd_lock);
6030 spin_unlock(<ds->ltd_lock);
6032 rc = ptlrpc_set_wait(set);
6033 ptlrpc_set_destroy(set);
6039 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail "
6040 "to sync failure with MDTs, and related MDTs "
6041 "may handle orphan improperly: rc = %d\n",
6042 lfsck_lfsck2name(lfsck), rc);
6047 struct lfsck_assistant_operations lfsck_namespace_assistant_ops = {
6048 .la_handler_p1 = lfsck_namespace_assistant_handler_p1,
6049 .la_handler_p2 = lfsck_namespace_assistant_handler_p2,
6050 .la_fill_pos = lfsck_namespace_assistant_fill_pos,
6051 .la_double_scan_result = lfsck_namespace_double_scan_result,
6052 .la_req_fini = lfsck_namespace_assistant_req_fini,
6053 .la_sync_failures = lfsck_namespace_assistant_sync_failures,
6057 * Verify the specified linkEA entry for the given directory object.
6058 * If the object has no such linkEA entry or it has more other linkEA
6059 * entries, then re-generate the linkEA with the given information.
6061 * \param[in] env pointer to the thread context
6062 * \param[in] dev pointer to the dt_device
6063 * \param[in] obj pointer to the dt_object to be handled
6064 * \param[in] cname the name for the child in the parent directory
6065 * \param[in] pfid the parent directory's FID for the linkEA
6067 * \retval 0 for success
6068 * \retval negative error number on failure
6070 int lfsck_verify_linkea(const struct lu_env *env, struct dt_device *dev,
6071 struct dt_object *obj, const struct lu_name *cname,
6072 const struct lu_fid *pfid)
6074 struct linkea_data ldata = { NULL };
6075 struct lu_buf linkea_buf;
6078 int fl = LU_XATTR_CREATE;
6082 LASSERT(S_ISDIR(lfsck_object_type(obj)));
6084 rc = lfsck_links_read(env, obj, &ldata);
6085 if (rc == -ENODATA) {
6087 } else if (rc == 0) {
6088 fl = LU_XATTR_REPLACE;
6089 if (ldata.ld_leh->leh_reccount != 1) {
6092 rc = linkea_links_find(&ldata, cname, pfid);
6101 rc = linkea_data_new(&ldata, &lfsck_env_info(env)->lti_linkea_buf);
6105 rc = linkea_add_buf(&ldata, cname, pfid);
6109 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
6110 ldata.ld_leh->leh_len);
6111 th = dt_trans_create(env, dev);
6113 RETURN(PTR_ERR(th));
6115 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
6116 XATTR_NAME_LINK, fl, th);
6120 rc = dt_trans_start_local(env, dev, th);
6124 dt_write_lock(env, obj, 0);
6125 rc = dt_xattr_set(env, obj, &linkea_buf,
6126 XATTR_NAME_LINK, fl, th, BYPASS_CAPA);
6127 dt_write_unlock(env, obj);
6132 dt_trans_stop(env, dev, th);
6137 * Get the name and parent directory's FID from the first linkEA entry.
6139 * \param[in] env pointer to the thread context
6140 * \param[in] obj pointer to the object which get linkEA from
6141 * \param[out] name pointer to the buffer to hold the name
6142 * in the first linkEA entry
6143 * \param[out] pfid pointer to the buffer to hold the parent
6144 * directory's FID in the first linkEA entry
6146 * \retval 0 for success
6147 * \retval negative error number on failure
6149 int lfsck_links_get_first(const struct lu_env *env, struct dt_object *obj,
6150 char *name, struct lu_fid *pfid)
6152 struct lu_name *cname = &lfsck_env_info(env)->lti_name;
6153 struct linkea_data ldata = { NULL };
6156 rc = lfsck_links_read(env, obj, &ldata);
6160 linkea_first_entry(&ldata);
6161 if (ldata.ld_lee == NULL)
6164 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, cname, pfid);
6165 /* To guarantee the 'name' is terminated with '0'. */
6166 memcpy(name, cname->ln_name, cname->ln_namelen);
6167 name[cname->ln_namelen] = 0;
6173 * Update the object's name entry with the given FID.
6175 * \param[in] env pointer to the thread context
6176 * \param[in] lfsck pointer to the lfsck instance
6177 * \param[in] parent pointer to the parent directory that holds
6179 * \param[in] name the name for the entry to be updated
6180 * \param[in] pfid the new PFID for the name entry
6181 * \param[in] type the type for the name entry to be updated
6183 * \retval 0 for success
6184 * \retval negative error number on failure
6186 int lfsck_update_name_entry(const struct lu_env *env,
6187 struct lfsck_instance *lfsck,
6188 struct dt_object *parent, const char *name,
6189 const struct lu_fid *pfid, __u32 type)
6191 struct dt_insert_rec *rec = &lfsck_env_info(env)->lti_dt_rec;
6192 struct dt_device *dev = lfsck->li_next;
6193 struct lustre_handle lh = { 0 };
6199 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
6200 MDS_INODELOCK_UPDATE, LCK_EX);
6204 th = dt_trans_create(env, dev);
6206 GOTO(unlock, rc = PTR_ERR(th));
6208 rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
6212 rec->rec_type = type;
6213 rec->rec_fid = pfid;
6214 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
6215 (const struct dt_key *)name, th);
6219 rc = dt_declare_ref_add(env, parent, th);
6223 rc = dt_trans_start(env, dev, th);
6227 rc = dt_delete(env, parent, (const struct dt_key *)name, th,
6229 if (rc == -ENOENT) {
6237 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
6238 (const struct dt_key *)name, th, BYPASS_CAPA, 1);
6239 if (rc == 0 && S_ISDIR(type) && !exists) {
6240 dt_write_lock(env, parent, 0);
6241 rc = dt_ref_add(env, parent, th);
6242 dt_write_unlock(env, parent);
6248 dt_trans_stop(env, dev, th);
6251 lfsck_ibits_unlock(&lh, LCK_EX);
6253 CDEBUG(D_LFSCK, "%s: update name entry "DFID"/%s with the FID "DFID
6254 " and the type %o: rc = %d\n", lfsck_lfsck2name(lfsck),
6255 PFID(lfsck_dto2fid(parent)), name, PFID(pfid), type, rc);
6260 int lfsck_namespace_setup(const struct lu_env *env,
6261 struct lfsck_instance *lfsck)
6263 struct lfsck_component *com;
6264 struct lfsck_namespace *ns;
6265 struct dt_object *root = NULL;
6266 struct dt_object *obj;
6270 LASSERT(lfsck->li_master);
6276 INIT_LIST_HEAD(&com->lc_link);
6277 INIT_LIST_HEAD(&com->lc_link_dir);
6278 init_rwsem(&com->lc_sem);
6279 atomic_set(&com->lc_ref, 1);
6280 com->lc_lfsck = lfsck;
6281 com->lc_type = LFSCK_TYPE_NAMESPACE;
6282 com->lc_ops = &lfsck_namespace_ops;
6283 com->lc_data = lfsck_assistant_data_init(
6284 &lfsck_namespace_assistant_ops,
6286 if (com->lc_data == NULL)
6287 GOTO(out, rc = -ENOMEM);
6289 com->lc_file_size = sizeof(struct lfsck_namespace);
6290 OBD_ALLOC(com->lc_file_ram, com->lc_file_size);
6291 if (com->lc_file_ram == NULL)
6292 GOTO(out, rc = -ENOMEM);
6294 OBD_ALLOC(com->lc_file_disk, com->lc_file_size);
6295 if (com->lc_file_disk == NULL)
6296 GOTO(out, rc = -ENOMEM);
6298 root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
6300 GOTO(out, rc = PTR_ERR(root));
6302 if (unlikely(!dt_try_as_dir(env, root)))
6303 GOTO(out, rc = -ENOTDIR);
6305 obj = local_index_find_or_create(env, lfsck->li_los, root,
6306 lfsck_namespace_name,
6307 S_IFREG | S_IRUGO | S_IWUSR,
6308 &dt_lfsck_features);
6310 GOTO(out, rc = PTR_ERR(obj));
6313 rc = obj->do_ops->do_index_try(env, obj, &dt_lfsck_features);
6317 rc = lfsck_namespace_load(env, com);
6319 rc = lfsck_namespace_init(env, com);
6321 rc = lfsck_namespace_reset(env, com, true);
6325 ns = com->lc_file_ram;
6326 switch (ns->ln_status) {
6331 spin_lock(&lfsck->li_lock);
6332 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
6333 spin_unlock(&lfsck->li_lock);
6336 CERROR("%s: unknown lfsck_namespace status %d\n",
6337 lfsck_lfsck2name(lfsck), ns->ln_status);
6339 case LS_SCANNING_PHASE1:
6340 case LS_SCANNING_PHASE2:
6341 /* No need to store the status to disk right now.
6342 * If the system crashed before the status stored,
6343 * it will be loaded back when next time. */
6344 ns->ln_status = LS_CRASHED;
6348 spin_lock(&lfsck->li_lock);
6349 list_add_tail(&com->lc_link, &lfsck->li_list_scan);
6350 list_add_tail(&com->lc_link_dir, &lfsck->li_list_dir);
6351 spin_unlock(&lfsck->li_lock);
6358 if (root != NULL && !IS_ERR(root))
6359 lu_object_put(env, &root->do_lu);
6361 lfsck_component_cleanup(env, com);
6362 CERROR("%s: fail to init namespace LFSCK component: rc = %d\n",
6363 lfsck_lfsck2name(lfsck), rc);