X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Flfsck%2Flfsck_namespace.c;h=931ffa29a18eaa6a982f1b38d9ab977e13625e18;hp=b5e5d80c5024ef80c0af7af1b26ae1bf62fcddea;hb=445da16c2ac0475b1c1077c822800b68cdbb7ce3;hpb=d0fbde388136728285c9dc74f639508a01efc881 diff --git a/lustre/lfsck/lfsck_namespace.c b/lustre/lfsck/lfsck_namespace.c index b5e5d80..931ffa2 100644 --- a/lustre/lfsck/lfsck_namespace.c +++ b/lustre/lfsck/lfsck_namespace.c @@ -20,7 +20,7 @@ * GPL HEADER END */ /* - * Copyright (c) 2012, 2013, Intel Corporation. + * Copyright (c) 2013, 2015, Intel Corporation. */ /* * lustre/lfsck/lfsck_namespace.c @@ -41,7 +41,15 @@ #include "lfsck_internal.h" -#define LFSCK_NAMESPACE_MAGIC 0xA0629D03 +#define LFSCK_NAMESPACE_MAGIC_V1 0xA0629D03 +#define LFSCK_NAMESPACE_MAGIC_V2 0xA0621A0B + +/* For Lustre-2.x (x <= 6), the namespace LFSCK used LFSCK_NAMESPACE_MAGIC_V1 + * as the trace file magic. When downgrade to such old release, the old LFSCK + * will not recognize the new LFSCK_NAMESPACE_MAGIC_V2 in the new trace file, + * then it will reset the whole LFSCK, and will not cause start failure. The + * similar case will happen when upgrade from such old release. */ +#define LFSCK_NAMESPACE_MAGIC LFSCK_NAMESPACE_MAGIC_V2 enum lfsck_nameentry_check { LFSCK_NAMEENTRY_DEAD = 1, /* The object has been unlinked. */ @@ -49,10 +57,9 @@ enum lfsck_nameentry_check { LFSCK_NAMEENTRY_RECREATED = 3, /* The entry has been recreated. */ }; -static const char lfsck_namespace_name[] = "lfsck_namespace"; - static struct lfsck_namespace_req * lfsck_namespace_assistant_req_init(struct lfsck_instance *lfsck, + struct lfsck_assistant_object *lso, struct lu_dirent *ent, __u16 type) { struct lfsck_namespace_req *lnr; @@ -64,10 +71,9 @@ lfsck_namespace_assistant_req_init(struct lfsck_instance *lfsck, return ERR_PTR(-ENOMEM); INIT_LIST_HEAD(&lnr->lnr_lar.lar_list); - lu_object_get(&lfsck->li_obj_dir->do_lu); - lnr->lnr_obj = lfsck->li_obj_dir; + lnr->lnr_lar.lar_parent = lfsck_assistant_object_get(lso); + lnr->lnr_lmv = lfsck_lmv_get(lfsck->li_lmv); lnr->lnr_fid = ent->lde_fid; - lnr->lnr_oit_cookie = lfsck->li_pos_current.lp_oit_cookie; lnr->lnr_dir_cookie = ent->lde_hash; lnr->lnr_attr = ent->lde_attrs; lnr->lnr_size = size; @@ -84,7 +90,10 @@ static void lfsck_namespace_assistant_req_fini(const struct lu_env *env, struct lfsck_namespace_req *lnr = container_of0(lar, struct lfsck_namespace_req, lnr_lar); - lu_object_put(env, &lnr->lnr_obj->do_lu); + if (lnr->lnr_lmv != NULL) + lfsck_lmv_put(env, lnr->lnr_lmv); + + lfsck_assistant_object_put(env, lar->lar_parent); OBD_FREE(lnr, lnr->lnr_size); } @@ -131,6 +140,30 @@ static void lfsck_namespace_le_to_cpu(struct lfsck_namespace *dst, dst->ln_bad_type_repaired = le64_to_cpu(src->ln_bad_type_repaired); dst->ln_lost_dirent_repaired = le64_to_cpu(src->ln_lost_dirent_repaired); + dst->ln_striped_dirs_scanned = + le64_to_cpu(src->ln_striped_dirs_scanned); + dst->ln_striped_dirs_repaired = + le64_to_cpu(src->ln_striped_dirs_repaired); + dst->ln_striped_dirs_failed = + le64_to_cpu(src->ln_striped_dirs_failed); + dst->ln_striped_dirs_disabled = + le64_to_cpu(src->ln_striped_dirs_disabled); + dst->ln_striped_dirs_skipped = + le64_to_cpu(src->ln_striped_dirs_skipped); + dst->ln_striped_shards_scanned = + le64_to_cpu(src->ln_striped_shards_scanned); + dst->ln_striped_shards_repaired = + le64_to_cpu(src->ln_striped_shards_repaired); + dst->ln_striped_shards_failed = + le64_to_cpu(src->ln_striped_shards_failed); + dst->ln_striped_shards_skipped = + le64_to_cpu(src->ln_striped_shards_skipped); + dst->ln_name_hash_repaired = le64_to_cpu(src->ln_name_hash_repaired); + dst->ln_local_lpf_scanned = le64_to_cpu(src->ln_local_lpf_scanned); + dst->ln_local_lpf_moved = le64_to_cpu(src->ln_local_lpf_moved); + dst->ln_local_lpf_skipped = le64_to_cpu(src->ln_local_lpf_skipped); + dst->ln_local_lpf_failed = le64_to_cpu(src->ln_local_lpf_failed); + dst->ln_bitmap_size = le32_to_cpu(src->ln_bitmap_size); } static void lfsck_namespace_cpu_to_le(struct lfsck_namespace *dst, @@ -176,6 +209,30 @@ static void lfsck_namespace_cpu_to_le(struct lfsck_namespace *dst, dst->ln_bad_type_repaired = cpu_to_le64(src->ln_bad_type_repaired); dst->ln_lost_dirent_repaired = cpu_to_le64(src->ln_lost_dirent_repaired); + dst->ln_striped_dirs_scanned = + cpu_to_le64(src->ln_striped_dirs_scanned); + dst->ln_striped_dirs_repaired = + cpu_to_le64(src->ln_striped_dirs_repaired); + dst->ln_striped_dirs_failed = + cpu_to_le64(src->ln_striped_dirs_failed); + dst->ln_striped_dirs_disabled = + cpu_to_le64(src->ln_striped_dirs_disabled); + dst->ln_striped_dirs_skipped = + cpu_to_le64(src->ln_striped_dirs_skipped); + dst->ln_striped_shards_scanned = + cpu_to_le64(src->ln_striped_shards_scanned); + dst->ln_striped_shards_repaired = + cpu_to_le64(src->ln_striped_shards_repaired); + dst->ln_striped_shards_failed = + cpu_to_le64(src->ln_striped_shards_failed); + dst->ln_striped_shards_skipped = + cpu_to_le64(src->ln_striped_shards_skipped); + dst->ln_name_hash_repaired = cpu_to_le64(src->ln_name_hash_repaired); + dst->ln_local_lpf_scanned = cpu_to_le64(src->ln_local_lpf_scanned); + dst->ln_local_lpf_moved = cpu_to_le64(src->ln_local_lpf_moved); + dst->ln_local_lpf_skipped = cpu_to_le64(src->ln_local_lpf_skipped); + dst->ln_local_lpf_failed = cpu_to_le64(src->ln_local_lpf_failed); + dst->ln_bitmap_size = cpu_to_le32(src->ln_bitmap_size); } static void lfsck_namespace_record_failure(const struct lu_env *env, @@ -191,7 +248,7 @@ static void lfsck_namespace_record_failure(const struct lu_env *env, ns->ln_pos_first_inconsistent = pos; CDEBUG(D_LFSCK, "%s: namespace LFSCK hit first non-repaired " - "inconsistency at the pos ["LPU64", "DFID", "LPX64"]\n", + "inconsistency at the pos [%llu, "DFID", %#llx]\n", lfsck_lfsck2name(lfsck), ns->ln_pos_first_inconsistent.lp_oit_cookie, PFID(&ns->ln_pos_first_inconsistent.lp_dir_parent), @@ -200,9 +257,81 @@ static void lfsck_namespace_record_failure(const struct lu_env *env, } /** - * \retval +ve: the lfsck_namespace is broken, the caller should reset it. - * \retval 0: succeed. - * \retval -ve: failed cases. + * Load the MDT bitmap from the lfsck_namespace trace file. + * + * \param[in] env pointer to the thread context + * \param[in] com pointer to the lfsck component + * + * \retval 0 for success + * \retval negative error number on failure or data corruption + */ +static int lfsck_namespace_load_bitmap(const struct lu_env *env, + struct lfsck_component *com) +{ + struct dt_object *obj = com->lc_obj; + struct lfsck_assistant_data *lad = com->lc_data; + struct lfsck_namespace *ns = com->lc_file_ram; + struct cfs_bitmap *bitmap = lad->lad_bitmap; + ssize_t size; + __u32 nbits; + int rc; + ENTRY; + + if (com->lc_lfsck->li_mdt_descs.ltd_tgts_bitmap->size > + ns->ln_bitmap_size) + nbits = com->lc_lfsck->li_mdt_descs.ltd_tgts_bitmap->size; + else + nbits = ns->ln_bitmap_size; + + if (unlikely(nbits < BITS_PER_LONG)) + nbits = BITS_PER_LONG; + + if (nbits > bitmap->size) { + __u32 new_bits = bitmap->size; + struct cfs_bitmap *new_bitmap; + + while (new_bits < nbits) + new_bits <<= 1; + + new_bitmap = CFS_ALLOCATE_BITMAP(new_bits); + if (new_bitmap == NULL) + RETURN(-ENOMEM); + + lad->lad_bitmap = new_bitmap; + CFS_FREE_BITMAP(bitmap); + bitmap = new_bitmap; + } + + if (ns->ln_bitmap_size == 0) { + lad->lad_incomplete = 0; + CFS_RESET_BITMAP(bitmap); + + RETURN(0); + } + + size = (ns->ln_bitmap_size + 7) >> 3; + rc = dt_xattr_get(env, obj, + lfsck_buf_get(env, bitmap->data, size), + XATTR_NAME_LFSCK_BITMAP); + if (rc != size) + RETURN(rc >= 0 ? -EINVAL : rc); + + if (cfs_bitmap_check_empty(bitmap)) + lad->lad_incomplete = 0; + else + lad->lad_incomplete = 1; + + RETURN(0); +} + +/** + * Load namespace LFSCK statistics information from the trace file. + * + * \param[in] env pointer to the thread context + * \param[in] com pointer to the lfsck component + * + * \retval 0 for success + * \retval negative error number on failure */ static int lfsck_namespace_load(const struct lu_env *env, struct lfsck_component *com) @@ -212,7 +341,7 @@ static int lfsck_namespace_load(const struct lu_env *env, rc = dt_xattr_get(env, com->lc_obj, lfsck_buf_get(env, com->lc_file_disk, len), - XATTR_NAME_LFSCK_NAMESPACE, BYPASS_CAPA); + XATTR_NAME_LFSCK_NAMESPACE); if (rc == len) { struct lfsck_namespace *ns = com->lc_file_ram; @@ -222,7 +351,7 @@ static int lfsck_namespace_load(const struct lu_env *env, CDEBUG(D_LFSCK, "%s: invalid lfsck_namespace magic " "%#x != %#x\n", lfsck_lfsck2name(com->lc_lfsck), ns->ln_magic, LFSCK_NAMESPACE_MAGIC); - rc = 1; + rc = -ESTALE; } else { rc = 0; } @@ -231,24 +360,39 @@ static int lfsck_namespace_load(const struct lu_env *env, "expected = %d: rc = %d\n", lfsck_lfsck2name(com->lc_lfsck), len, rc); if (rc >= 0) - rc = 1; + rc = -ESTALE; } + return rc; } static int lfsck_namespace_store(const struct lu_env *env, - struct lfsck_component *com, bool init) + struct lfsck_component *com) { - struct dt_object *obj = com->lc_obj; - struct lfsck_instance *lfsck = com->lc_lfsck; - struct thandle *handle; - int len = com->lc_file_size; - int rc; + struct dt_object *obj = com->lc_obj; + struct lfsck_instance *lfsck = com->lc_lfsck; + struct lfsck_namespace *ns = com->lc_file_ram; + struct lfsck_assistant_data *lad = com->lc_data; + struct dt_device *dev = lfsck_obj2dev(obj); + struct cfs_bitmap *bitmap = NULL; + struct thandle *handle; + __u32 nbits = 0; + int len = com->lc_file_size; + int rc; ENTRY; + if (lad != NULL) { + bitmap = lad->lad_bitmap; + nbits = bitmap->size; + + LASSERT(nbits > 0); + LASSERTF((nbits & 7) == 0, "Invalid nbits %u\n", nbits); + } + + ns->ln_bitmap_size = nbits; lfsck_namespace_cpu_to_le((struct lfsck_namespace *)com->lc_file_disk, - (struct lfsck_namespace *)com->lc_file_ram); - handle = dt_trans_create(env, lfsck->li_bottom); + ns); + handle = dt_trans_create(env, dev); if (IS_ERR(handle)) GOTO(log, rc = PTR_ERR(handle)); @@ -258,20 +402,30 @@ static int lfsck_namespace_store(const struct lu_env *env, if (rc != 0) GOTO(out, rc); - rc = dt_trans_start_local(env, lfsck->li_bottom, handle); + if (bitmap != NULL) { + rc = dt_declare_xattr_set(env, obj, + lfsck_buf_get(env, bitmap->data, nbits >> 3), + XATTR_NAME_LFSCK_BITMAP, 0, handle); + if (rc != 0) + GOTO(out, rc); + } + + rc = dt_trans_start_local(env, dev, handle); if (rc != 0) GOTO(out, rc); rc = dt_xattr_set(env, obj, lfsck_buf_get(env, com->lc_file_disk, len), - XATTR_NAME_LFSCK_NAMESPACE, - init ? LU_XATTR_CREATE : LU_XATTR_REPLACE, - handle, BYPASS_CAPA); + XATTR_NAME_LFSCK_NAMESPACE, 0, handle); + if (rc == 0 && bitmap != NULL) + rc = dt_xattr_set(env, obj, + lfsck_buf_get(env, bitmap->data, nbits >> 3), + XATTR_NAME_LFSCK_BITMAP, 0, handle); GOTO(out, rc); out: - dt_trans_stop(env, lfsck->li_bottom, handle); + dt_trans_stop(env, dev, handle); log: if (rc != 0) @@ -280,6 +434,64 @@ log: return rc; } +static struct dt_object * +lfsck_namespace_load_one_trace_file(const struct lu_env *env, + struct lfsck_component *com, + struct dt_object *parent, + const char *name, bool reset) +{ + struct lfsck_instance *lfsck = com->lc_lfsck; + struct dt_object *obj; + int rc; + + if (reset) { + rc = local_object_unlink(env, lfsck->li_bottom, parent, name); + if (rc != 0 && rc != -ENOENT) + return ERR_PTR(rc); + } + + obj = local_index_find_or_create(env, lfsck->li_los, parent, name, + S_IFREG | S_IRUGO | S_IWUSR, + &dt_lfsck_features); + + return obj; +} + +static int lfsck_namespace_load_sub_trace_files(const struct lu_env *env, + struct lfsck_component *com, + bool reset) +{ + char *name = lfsck_env_info(env)->lti_key; + struct lfsck_sub_trace_obj *lsto; + struct dt_object *obj; + int rc; + int i; + + for (i = 0, lsto = &com->lc_sub_trace_objs[0]; + i < LFSCK_STF_COUNT; i++, lsto++) { + snprintf(name, NAME_MAX, "%s_%02d", LFSCK_NAMESPACE, i); + if (lsto->lsto_obj != NULL) { + if (!reset) + continue; + + lfsck_object_put(env, lsto->lsto_obj); + lsto->lsto_obj = NULL; + } + + obj = lfsck_namespace_load_one_trace_file(env, com, + com->lc_lfsck->li_lfsck_dir, name, reset); + if (IS_ERR(obj)) + return PTR_ERR(obj); + + lsto->lsto_obj = obj; + rc = obj->do_ops->do_index_try(env, obj, &dt_lfsck_features); + if (rc != 0) + return rc; + } + + return 0; +} + static int lfsck_namespace_init(const struct lu_env *env, struct lfsck_component *com) { @@ -290,21 +502,24 @@ static int lfsck_namespace_init(const struct lu_env *env, ns->ln_magic = LFSCK_NAMESPACE_MAGIC; ns->ln_status = LS_INIT; down_write(&com->lc_sem); - rc = lfsck_namespace_store(env, com, true); + rc = lfsck_namespace_store(env, com); up_write(&com->lc_sem); + if (rc == 0) + rc = lfsck_namespace_load_sub_trace_files(env, com, true); + return rc; } /** - * Update the namespace LFSCK tracing file for the given @fid + * Update the namespace LFSCK trace file for the given @fid * * \param[in] env pointer to the thread context * \param[in] com pointer to the lfsck component * \param[in] fid the fid which flags to be updated in the lfsck - * tracing file + * trace file * \param[in] add true if add new flags, otherwise remove flags * - * \retval 0 for succeed or nothing to be done + * \retval 0 for success or nothing to be done * \retval negative error number on failure */ int lfsck_namespace_trace_update(const struct lu_env *env, @@ -313,10 +528,11 @@ int lfsck_namespace_trace_update(const struct lu_env *env, const __u8 flags, bool add) { struct lfsck_instance *lfsck = com->lc_lfsck; - struct dt_object *obj = com->lc_obj; + struct dt_object *obj; struct lu_fid *key = &lfsck_env_info(env)->lti_fid3; - struct dt_device *dev = lfsck->li_bottom; + struct dt_device *dev; struct thandle *th = NULL; + int idx; int rc = 0; __u8 old = 0; __u8 new = 0; @@ -324,10 +540,16 @@ int lfsck_namespace_trace_update(const struct lu_env *env, LASSERT(flags != 0); - down_write(&com->lc_sem); + if (unlikely(!fid_is_sane(fid))) + RETURN(0); + + idx = lfsck_sub_trace_file_fid2idx(fid); + obj = com->lc_sub_trace_objs[idx].lsto_obj; + dev = lfsck_obj2dev(obj); + mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex); fid_cpu_to_be(key, fid); rc = dt_lookup(env, obj, (struct dt_rec *)&old, - (const struct dt_key *)key, BYPASS_CAPA); + (const struct dt_key *)key); if (rc == -ENOENT) { if (!add) GOTO(unlock, rc = 0); @@ -374,15 +596,14 @@ int lfsck_namespace_trace_update(const struct lu_env *env, GOTO(log, rc); if (old != 0) { - rc = dt_delete(env, obj, (const struct dt_key *)key, - th, BYPASS_CAPA); + rc = dt_delete(env, obj, (const struct dt_key *)key, th); if (rc != 0) GOTO(log, rc); } if (new != 0) { rc = dt_insert(env, obj, (const struct dt_rec *)&new, - (const struct dt_key *)key, th, BYPASS_CAPA, 1); + (const struct dt_key *)key, th, 1); if (rc != 0) GOTO(log, rc); } @@ -394,19 +615,19 @@ log: dt_trans_stop(env, dev, th); CDEBUG(D_LFSCK, "%s: namespace LFSCK %s flags for "DFID" in the " - "tracing file, flags %x, old %x, new %x: rc = %d\n", + "trace file, flags %x, old %x, new %x: rc = %d\n", lfsck_lfsck2name(lfsck), add ? "add" : "del", PFID(fid), (__u32)flags, (__u32)old, (__u32)new, rc); unlock: - up_write(&com->lc_sem); + mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex); return rc; } -static int lfsck_namespace_check_exist(const struct lu_env *env, - struct dt_object *dir, - struct dt_object *obj, const char *name) +int lfsck_namespace_check_exist(const struct lu_env *env, + struct dt_object *dir, + struct dt_object *obj, const char *name) { struct lu_fid *fid = &lfsck_env_info(env)->lti_fid; int rc; @@ -416,7 +637,7 @@ static int lfsck_namespace_check_exist(const struct lu_env *env, RETURN(LFSCK_NAMEENTRY_DEAD); rc = dt_lookup(env, dir, (struct dt_rec *)fid, - (const struct dt_key *)name, BYPASS_CAPA); + (const struct dt_key *)name); if (rc == -ENOENT) RETURN(LFSCK_NAMEENTRY_REMOVED); @@ -437,11 +658,9 @@ static int lfsck_declare_namespace_exec_dir(const struct lu_env *env, /* For destroying all invalid linkEA entries. */ rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, handle); - if (rc != 0) - return rc; - - /* For insert new linkEA entry. */ - rc = dt_declare_xattr_set(env, obj, + if (rc == 0) + /* For insert new linkEA entry. */ + rc = dt_declare_xattr_set(env, obj, lfsck_buf_get_const(env, NULL, DEFAULT_LINKEA_SIZE), XATTR_NAME_LINK, 0, handle); return rc; @@ -458,22 +677,26 @@ int __lfsck_links_read(const struct lu_env *env, struct dt_object *obj, if (!dt_object_exists(obj)) return -ENOENT; - rc = dt_xattr_get(env, obj, ldata->ld_buf, XATTR_NAME_LINK, BYPASS_CAPA); + rc = dt_xattr_get(env, obj, ldata->ld_buf, XATTR_NAME_LINK); if (rc == -ERANGE) { /* Buf was too small, figure out what we need. */ - rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LINK, - BYPASS_CAPA); - if (rc <= 0) + rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LINK); + if (unlikely(rc == 0)) + return -ENODATA; + + if (rc < 0) return rc; lu_buf_realloc(ldata->ld_buf, rc); if (ldata->ld_buf->lb_buf == NULL) return -ENOMEM; - rc = dt_xattr_get(env, obj, ldata->ld_buf, XATTR_NAME_LINK, - BYPASS_CAPA); + rc = dt_xattr_get(env, obj, ldata->ld_buf, XATTR_NAME_LINK); } + if (unlikely(rc == 0)) + return -ENODATA; + if (rc > 0) rc = linkea_init(ldata); @@ -497,7 +720,7 @@ static int lfsck_namespace_links_remove(const struct lu_env *env, struct dt_object *obj) { struct lfsck_instance *lfsck = com->lc_lfsck; - struct dt_device *dev = lfsck->li_bottom; + struct dt_device *dev = lfsck_obj2dev(obj); struct thandle *th = NULL; int rc = 0; ENTRY; @@ -523,7 +746,7 @@ static int lfsck_namespace_links_remove(const struct lu_env *env, if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN) GOTO(unlock, rc = 0); - rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, th, BYPASS_CAPA); + rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, th); GOTO(unlock, rc); @@ -554,20 +777,62 @@ static int lfsck_links_write(const struct lu_env *env, struct dt_object *obj, ldata->ld_buf->lb_buf, ldata->ld_leh->leh_len); - return dt_xattr_set(env, obj, buf, XATTR_NAME_LINK, 0, handle, - BYPASS_CAPA); + return dt_xattr_set(env, obj, buf, XATTR_NAME_LINK, 0, handle); } -static void lfsck_namespace_unpack_linkea_entry(struct linkea_data *ldata, - struct lu_name *cname, - struct lu_fid *pfid, - char *buf) +static int lfsck_namespace_unpack_linkea_entry(struct linkea_data *ldata, + struct lu_name *cname, + struct lu_fid *pfid, + char *buf, const int buflen) { linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, cname, pfid); + if (unlikely(ldata->ld_reclen <= 0 || + ldata->ld_reclen + sizeof(struct link_ea_header) > + ldata->ld_leh->leh_len || + cname->ln_namelen <= 0 || + cname->ln_namelen > NAME_MAX || + cname->ln_namelen >= buflen || + !fid_is_sane(pfid))) + return -EINVAL; + /* To guarantee the 'name' is terminated with '0'. */ memcpy(buf, cname->ln_name, cname->ln_namelen); buf[cname->ln_namelen] = 0; cname->ln_name = buf; + + return 0; +} + +static void lfsck_linkea_del_buf(struct linkea_data *ldata, + const struct lu_name *lname) +{ + LASSERT(ldata->ld_leh != NULL && ldata->ld_lee != NULL); + + /* If current record is corrupted, all the subsequent + * records will be dropped. */ + if (unlikely(ldata->ld_reclen <= 0 || + ldata->ld_reclen + sizeof(struct link_ea_header) > + ldata->ld_leh->leh_len)) { + void *ptr = ldata->ld_lee; + + ldata->ld_leh->leh_len = sizeof(struct link_ea_header); + ldata->ld_leh->leh_reccount = 0; + linkea_first_entry(ldata); + while (ldata->ld_lee != NULL && + (char *)ldata->ld_lee < (char *)ptr) { + int reclen = (ldata->ld_lee->lee_reclen[0] << 8) | + ldata->ld_lee->lee_reclen[1]; + + ldata->ld_leh->leh_len += reclen; + ldata->ld_leh->leh_reccount++; + ldata->ld_lee = (struct link_ea_entry *) + ((char *)ldata->ld_lee + reclen); + } + + ldata->ld_lee = NULL; + } else { + linkea_del_buf(ldata, lname); + } } static int lfsck_namespace_filter_linkea_entry(struct linkea_data *ldata, @@ -591,7 +856,7 @@ static int lfsck_namespace_filter_linkea_entry(struct linkea_data *ldata, if (!remove) break; - linkea_del_buf(ldata, cname); + lfsck_linkea_del_buf(ldata, cname); } else { linkea_next_entry(ldata); } @@ -626,6 +891,8 @@ static int lfsck_namespace_filter_linkea_entry(struct linkea_data *ldata, * type "O": The MDT-object has no linkEA, and there is no name * entry that references the MDT-object. * + * type "S": The orphan MDT-object is a shard of a striped directory + * * \see lfsck_layout_recreate_parent() for more types. * * The orphan name will be like: @@ -647,16 +914,17 @@ static int lfsck_namespace_insert_orphan(const struct lu_env *env, struct lfsck_thread_info *info = lfsck_env_info(env); struct lu_name *cname = &info->lti_name; struct dt_insert_rec *rec = &info->lti_dt_rec; - struct lu_fid *tfid = &info->lti_fid5; + struct lu_attr *la = &info->lti_la2; const struct lu_fid *cfid = lfsck_dto2fid(orphan); const struct lu_fid *pfid; + struct lu_fid tfid; struct lfsck_instance *lfsck = com->lc_lfsck; - struct dt_device *dev = lfsck->li_bottom; + struct dt_device *dev = lfsck_obj2dev(orphan); struct dt_object *parent; struct thandle *th = NULL; - struct lustre_handle plh = { 0 }; + struct lfsck_lock_handle *pllh = &info->lti_llh; struct lustre_handle clh = { 0 }; - struct linkea_data ldata = { 0 }; + struct linkea_data ldata2 = { NULL }; struct lu_buf linkea_buf; int namelen; int idx = 0; @@ -665,53 +933,66 @@ static int lfsck_namespace_insert_orphan(const struct lu_env *env, ENTRY; cname->ln_name = NULL; - /* Create .lustre/lost+found/MDTxxxx when needed. */ - if (unlikely(lfsck->li_lpf_obj == NULL)) { - rc = lfsck_create_lpf(env, lfsck); - if (rc != 0) - GOTO(log, rc); - } + if (unlikely(lfsck->li_lpf_obj == NULL)) + GOTO(log, rc = -ENXIO); parent = lfsck->li_lpf_obj; pfid = lfsck_dto2fid(parent); - /* Hold update lock on the parent to prevent others to access. */ - rc = lfsck_ibits_lock(env, lfsck, parent, &plh, - MDS_INODELOCK_UPDATE, LCK_EX); - if (rc != 0) - GOTO(log, rc); - +again: do { namelen = snprintf(info->lti_key, NAME_MAX, DFID"%s-%s-%d", PFID(cfid), infix, type, idx++); - rc = dt_lookup(env, parent, (struct dt_rec *)tfid, - (const struct dt_key *)info->lti_key, - BYPASS_CAPA); + rc = dt_lookup(env, parent, (struct dt_rec *)&tfid, + (const struct dt_key *)info->lti_key); if (rc != 0 && rc != -ENOENT) GOTO(log, rc); - if (unlikely(rc == 0 && lu_fid_eq(cfid, tfid))) + if (unlikely(rc == 0 && lu_fid_eq(cfid, &tfid))) exist = true; } while (rc == 0 && !exist); + rc = lfsck_lock(env, lfsck, parent, info->lti_key, pllh, + MDS_INODELOCK_UPDATE, LCK_PW); + if (rc != 0) + GOTO(log, rc); + + /* Re-check whether the name conflict with othrs after taken + * the ldlm lock. */ + rc = dt_lookup(env, parent, (struct dt_rec *)&tfid, + (const struct dt_key *)info->lti_key); + if (rc == 0) { + if (!lu_fid_eq(cfid, &tfid)) { + exist = false; + lfsck_unlock(pllh); + goto again; + } + + exist = true; + } else if (rc != -ENOENT) { + GOTO(log, rc); + } else { + exist = false; + } + cname->ln_name = info->lti_key; cname->ln_namelen = namelen; - rc = linkea_data_new(&ldata, &info->lti_linkea_buf2); + rc = linkea_data_new(&ldata2, &info->lti_linkea_buf2); if (rc != 0) GOTO(log, rc); - rc = linkea_add_buf(&ldata, cname, pfid); + rc = linkea_add_buf(&ldata2, cname, pfid); if (rc != 0) GOTO(log, rc); rc = lfsck_ibits_lock(env, lfsck, orphan, &clh, - MDS_INODELOCK_UPDATE | MDS_INODELOCK_LOOKUP, - LCK_EX); + MDS_INODELOCK_UPDATE | MDS_INODELOCK_LOOKUP | + MDS_INODELOCK_XATTR, LCK_EX); if (rc != 0) GOTO(log, rc); - lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf, - ldata.ld_leh->leh_len); + lfsck_buf_init(&linkea_buf, ldata2.ld_buf->lb_buf, + ldata2.ld_leh->leh_len); th = dt_trans_create(env, dev); if (IS_ERR(th)) GOTO(log, rc = PTR_ERR(th)); @@ -751,38 +1032,44 @@ static int lfsck_namespace_insert_orphan(const struct lu_env *env, } } + memset(la, 0, sizeof(*la)); + la->la_ctime = cfs_time_current_sec(); + la->la_valid = LA_CTIME; + rc = dt_declare_attr_set(env, orphan, la, th); + if (rc != 0) + GOTO(stop, rc); + rc = dt_trans_start_local(env, dev, th); if (rc != 0) GOTO(stop, rc); dt_write_lock(env, orphan, 0); - rc = lfsck_links_read(env, orphan, &ldata); + rc = lfsck_links_read2(env, orphan, &ldata2); if (likely((rc == -ENODATA) || (rc == -EINVAL) || - (rc == 0 && ldata.ld_leh->leh_reccount == 0))) { + (rc == 0 && ldata2.ld_leh != NULL && + ldata2.ld_leh->leh_reccount == 0))) { if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN) GOTO(unlock, rc = 1); if (S_ISDIR(lfsck_object_type(orphan))) { rc = dt_delete(env, orphan, - (const struct dt_key *)dotdot, th, - BYPASS_CAPA); + (const struct dt_key *)dotdot, th); if (rc != 0) GOTO(unlock, rc); rec->rec_type = S_IFDIR; rec->rec_fid = pfid; rc = dt_insert(env, orphan, (const struct dt_rec *)rec, - (const struct dt_key *)dotdot, th, - BYPASS_CAPA, 1); + (const struct dt_key *)dotdot, th, 1); if (rc != 0) GOTO(unlock, rc); } rc = dt_xattr_set(env, orphan, &linkea_buf, XATTR_NAME_LINK, 0, - th, BYPASS_CAPA); + th); } else { if (rc == 0 && count != NULL) - *count = ldata.ld_leh->leh_reccount; + *count = ldata2.ld_leh->leh_reccount; GOTO(unlock, rc); } @@ -792,8 +1079,7 @@ static int lfsck_namespace_insert_orphan(const struct lu_env *env, rec->rec_type = lfsck_object_type(orphan) & S_IFMT; rec->rec_fid = cfid; rc = dt_insert(env, parent, (const struct dt_rec *)rec, - (const struct dt_key *)cname->ln_name, - th, BYPASS_CAPA, 1); + (const struct dt_key *)cname->ln_name, th, 1); if (rc == 0 && S_ISDIR(rec->rec_type)) { dt_write_lock(env, parent, 0); rc = dt_ref_add(env, parent, th); @@ -801,6 +1087,9 @@ static int lfsck_namespace_insert_orphan(const struct lu_env *env, } } + if (rc == 0) + rc = dt_attr_set(env, orphan, la, th); + GOTO(stop, rc = (rc == 0 ? 1 : rc)); unlock: @@ -811,7 +1100,7 @@ stop: log: lfsck_ibits_unlock(&clh, LCK_EX); - lfsck_ibits_unlock(&plh, LCK_EX); + lfsck_unlock(pllh); CDEBUG(D_LFSCK, "%s: namespace LFSCK insert orphan for the " "object "DFID", name = %s: rc = %d\n", lfsck_lfsck2name(lfsck), PFID(cfid), @@ -858,21 +1147,34 @@ static int lfsck_namespace_insert_normal(const struct lu_env *env, struct lu_attr *la = &info->lti_la; struct dt_insert_rec *rec = &info->lti_dt_rec; struct lfsck_instance *lfsck = com->lc_lfsck; + /* The child and its name may be on different MDTs. */ + const struct lu_fid *pfid = lfsck_dto2fid(parent); + const struct lu_fid *cfid = lfsck_dto2fid(child); struct dt_device *dev = lfsck->li_next; struct thandle *th = NULL; - struct lustre_handle lh = { 0 }; + struct lfsck_lock_handle *llh = &info->lti_llh; int rc = 0; ENTRY; + /* @parent/@child may be based on lfsck->li_bottom, + * but here we need the object based on the lfsck->li_next. */ + + parent = lfsck_object_locate(dev, parent); + if (IS_ERR(parent)) + GOTO(log, rc = PTR_ERR(parent)); + if (unlikely(!dt_try_as_dir(env, parent))) GOTO(log, rc = -ENOTDIR); + child = lfsck_object_locate(dev, child); + if (IS_ERR(child)) + GOTO(log, rc = PTR_ERR(child)); + if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN) GOTO(log, rc = 1); - /* Hold update lock on the parent to prevent others to access. */ - rc = lfsck_ibits_lock(env, lfsck, parent, &lh, - MDS_INODELOCK_UPDATE, LCK_EX); + rc = lfsck_lock(env, lfsck, parent, name, llh, + MDS_INODELOCK_UPDATE, LCK_PW); if (rc != 0) GOTO(log, rc); @@ -881,7 +1183,7 @@ static int lfsck_namespace_insert_normal(const struct lu_env *env, GOTO(unlock, rc = PTR_ERR(th)); rec->rec_type = lfsck_object_type(child) & S_IFMT; - rec->rec_fid = lfsck_dto2fid(child); + rec->rec_fid = cfid; rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec, (const struct dt_key *)name, th); if (rc != 0) @@ -900,12 +1202,16 @@ static int lfsck_namespace_insert_normal(const struct lu_env *env, if (rc != 0) GOTO(stop, rc); + rc = dt_declare_attr_set(env, child, la, th); + if (rc != 0) + GOTO(stop, rc); + rc = dt_trans_start_local(env, dev, th); if (rc != 0) GOTO(stop, rc); rc = dt_insert(env, parent, (const struct dt_rec *)rec, - (const struct dt_key *)name, th, BYPASS_CAPA, 1); + (const struct dt_key *)name, th, 1); if (rc != 0) GOTO(stop, rc); @@ -918,7 +1224,11 @@ static int lfsck_namespace_insert_normal(const struct lu_env *env, } la->la_ctime = cfs_time_current_sec(); - rc = dt_attr_set(env, parent, la, th, BYPASS_CAPA); + rc = dt_attr_set(env, parent, la, th); + if (rc != 0) + GOTO(stop, rc); + + rc = dt_attr_set(env, child, la, th); GOTO(stop, rc = (rc == 0 ? 1 : rc)); @@ -926,14 +1236,13 @@ stop: dt_trans_stop(env, dev, th); unlock: - lfsck_ibits_unlock(&lh, LCK_EX); + lfsck_unlock(llh); log: CDEBUG(D_LFSCK, "%s: namespace LFSCK insert object "DFID" with " "the name %s and type %o to the parent "DFID": rc = %d\n", - lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(child)), name, - lfsck_object_type(child) & S_IFMT, - PFID(lfsck_dto2fid(parent)), rc); + lfsck_lfsck2name(lfsck), PFID(cfid), name, + lfsck_object_type(child) & S_IFMT, PFID(pfid), rc); if (rc != 0) { struct lfsck_namespace *ns = com->lc_file_ram; @@ -947,106 +1256,7 @@ log: } /** - * Create the specified orphan MDT-object on remote MDT. - * - * The LFSCK instance on this MDT will send LFSCK RPC to remote MDT to - * ask the remote LFSCK instance to create the specified orphan object - * under .lustre/lost+found/MDTxxxx/ directory with the name: - * ${FID}-P-${conflict_version}. - * - * \param[in] env pointer to the thread context - * \param[in] com pointer to the lfsck component - * \param[in] orphan pointer to the orphan MDT-object - * \param[in] type the orphan's type to be created - * - * type "P": The orphan object to be created was a parent directory - * of some DMT-object which linkEA shows that the @orphan - * object is missing. - * - * \see lfsck_layout_recreate_parent() for more types. - * - * \retval positive number for repaired cases - * \retval 0 if needs to repair nothing - * \retval negative error number on failure - */ -static int lfsck_namespace_create_orphan_remote(const struct lu_env *env, - struct lfsck_component *com, - struct dt_object *orphan, - __u32 type) -{ - struct lfsck_thread_info *info = lfsck_env_info(env); - struct lfsck_request *lr = &info->lti_lr; - struct lu_seq_range *range = &info->lti_range; - const struct lu_fid *fid = lfsck_dto2fid(orphan); - struct lfsck_namespace *ns = com->lc_file_ram; - struct lfsck_instance *lfsck = com->lc_lfsck; - struct seq_server_site *ss = - lu_site2seq(lfsck->li_bottom->dd_lu_dev.ld_site); - struct lfsck_tgt_desc *ltd = NULL; - struct ptlrpc_request *req = NULL; - int rc; - ENTRY; - - if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN) - GOTO(out, rc = 1); - - fld_range_set_mdt(range); - rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range); - if (rc != 0) - GOTO(out, rc); - - ltd = lfsck_tgt_get(&lfsck->li_mdt_descs, range->lsr_index); - if (ltd == NULL) { - ns->ln_flags |= LF_INCOMPLETE; - - GOTO(out, rc = -ENODEV); - } - - req = ptlrpc_request_alloc(class_exp2cliimp(ltd->ltd_exp), - &RQF_LFSCK_NOTIFY); - if (req == NULL) - GOTO(out, rc = -ENOMEM); - - rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY); - if (rc != 0) { - ptlrpc_request_free(req); - - GOTO(out, rc); - } - - lr = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST); - memset(lr, 0, sizeof(*lr)); - lr->lr_event = LE_CREATE_ORPHAN; - lr->lr_index = lfsck_dev_idx(lfsck->li_bottom); - lr->lr_active = LFSCK_TYPE_NAMESPACE; - lr->lr_fid = *fid; - lr->lr_type = type; - - ptlrpc_request_set_replen(req); - rc = ptlrpc_queue_wait(req); - ptlrpc_req_finished(req); - - if (rc == 0) - rc = 1; - else if (rc == -EEXIST) - rc = 0; - - GOTO(out, rc); - -out: - CDEBUG(D_LFSCK, "%s: namespace LFSCK create object " - DFID" on the MDT %x remotely: rc = %d\n", - lfsck_lfsck2name(lfsck), PFID(fid), - ltd != NULL ? ltd->ltd_index : -1, rc); - - if (ltd != NULL) - lfsck_tgt_put(ltd); - - return rc; -} - -/** - * Create the specified orphan MDT-object locally. + * Create the specified orphan directory. * * For the case that the parent MDT-object stored in some MDT-object's * linkEA entry is lost, the LFSCK will re-create the parent object as @@ -1056,21 +1266,15 @@ out: * \param[in] env pointer to the thread context * \param[in] com pointer to the lfsck component * \param[in] orphan pointer to the orphan MDT-object to be created - * \param[in] type the orphan's type to be created - * - * type "P": The orphan object to be created was a parent directory - * of some DMT-object which linkEA shows that the @orphan - * object is missing. - * - * \see lfsck_layout_recreate_parent() for more types. + * \param[in] lmv pointer to master LMV EA that will be set to the orphan * * \retval positive number for repaired cases * \retval negative error number on failure */ -static int lfsck_namespace_create_orphan_local(const struct lu_env *env, - struct lfsck_component *com, - struct dt_object *orphan, - __u32 type) +static int lfsck_namespace_create_orphan_dir(const struct lu_env *env, + struct lfsck_component *com, + struct dt_object *orphan, + struct lmv_mds_md_v1 *lmv) { struct lfsck_thread_info *info = lfsck_env_info(env); struct lu_attr *la = &info->lti_la; @@ -1078,79 +1282,110 @@ static int lfsck_namespace_create_orphan_local(const struct lu_env *env, struct dt_object_format *dof = &info->lti_dof; struct lu_name *cname = &info->lti_name2; struct dt_insert_rec *rec = &info->lti_dt_rec; - struct lu_fid *tfid = &info->lti_fid; + struct lmv_mds_md_v1 *lmv2 = &info->lti_lmv2; const struct lu_fid *cfid = lfsck_dto2fid(orphan); - const struct lu_fid *pfid; + struct lu_fid tfid; struct lfsck_instance *lfsck = com->lc_lfsck; - struct dt_device *dev = lfsck->li_bottom; + struct lfsck_namespace *ns = com->lc_file_ram; + struct dt_device *dev = lfsck_obj2dev(orphan); struct dt_object *parent = NULL; - struct dt_object *child = NULL; struct thandle *th = NULL; - struct lustre_handle lh = { 0 }; - struct linkea_data ldata = { 0 }; + struct lfsck_lock_handle *llh = &info->lti_llh; + struct linkea_data ldata = { NULL }; struct lu_buf linkea_buf; + struct lu_buf lmv_buf; char name[32]; int namelen; int idx = 0; int rc = 0; + int rc1 = 0; ENTRY; LASSERT(!dt_object_exists(orphan)); - LASSERT(!dt_object_remote(orphan)); - - /* @orphan maybe not attached to lfsck->li_bottom */ - child = lfsck_object_find_by_dev(env, dev, cfid); - if (IS_ERR(child)) - GOTO(log, rc = PTR_ERR(child)); cname->ln_name = NULL; if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN) GOTO(log, rc = 1); - /* Create .lustre/lost+found/MDTxxxx when needed. */ - if (unlikely(lfsck->li_lpf_obj == NULL)) { - rc = lfsck_create_lpf(env, lfsck); + if (dt_object_remote(orphan)) { + LASSERT(lfsck->li_lpf_root_obj != NULL); + + idx = lfsck_find_mdt_idx_by_fid(env, lfsck, cfid); + if (idx < 0) + GOTO(log, rc = idx); + + snprintf(name, 8, "MDT%04x", idx); + rc = dt_lookup(env, lfsck->li_lpf_root_obj, + (struct dt_rec *)&tfid, + (const struct dt_key *)name); if (rc != 0) - GOTO(log, rc); + GOTO(log, rc = (rc == -ENOENT ? -ENXIO : rc)); + + parent = lfsck_object_find_bottom(env, lfsck, &tfid); + if (IS_ERR(parent)) + GOTO(log, rc = PTR_ERR(parent)); + + if (unlikely(!dt_try_as_dir(env, parent))) + GOTO(log, rc = -ENOTDIR); + } else { + if (unlikely(lfsck->li_lpf_obj == NULL)) + GOTO(log, rc = -ENXIO); + + parent = lfsck->li_lpf_obj; } - parent = lfsck->li_lpf_obj; - pfid = lfsck_dto2fid(parent); + dev = lfsck_find_dev_by_fid(env, lfsck, cfid); + if (IS_ERR(dev)) + GOTO(log, rc = PTR_ERR(dev)); - /* Hold update lock on the parent to prevent others to access. */ - rc = lfsck_ibits_lock(env, lfsck, parent, &lh, - MDS_INODELOCK_UPDATE, LCK_EX); - if (rc != 0) - GOTO(log, rc); + idx = 0; +again: do { namelen = snprintf(name, 31, DFID"-P-%d", PFID(cfid), idx++); - rc = dt_lookup(env, parent, (struct dt_rec *)tfid, - (const struct dt_key *)name, BYPASS_CAPA); + rc = dt_lookup(env, parent, (struct dt_rec *)&tfid, + (const struct dt_key *)name); if (rc != 0 && rc != -ENOENT) - GOTO(unlock1, rc); + GOTO(log, rc); } while (rc == 0); - cname->ln_name = name; - cname->ln_namelen = namelen; + rc = lfsck_lock(env, lfsck, parent, name, llh, + MDS_INODELOCK_UPDATE, LCK_PW); + if (rc != 0) + GOTO(log, rc); - memset(la, 0, sizeof(*la)); - la->la_mode = type | (S_ISDIR(type) ? 0700 : 0600); + /* Re-check whether the name conflict with othrs after taken + * the ldlm lock. */ + rc = dt_lookup(env, parent, (struct dt_rec *)&tfid, + (const struct dt_key *)name); + if (unlikely(rc == 0)) { + lfsck_unlock(llh); + goto again; + } + + if (rc != -ENOENT) + GOTO(unlock1, rc); + + cname->ln_name = name; + cname->ln_namelen = namelen; + + memset(la, 0, sizeof(*la)); + la->la_mode = S_IFDIR | 0700; la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID | LA_ATIME | LA_MTIME | LA_CTIME; - child->do_ops->do_ah_init(env, hint, parent, child, - la->la_mode & S_IFMT); + orphan->do_ops->do_ah_init(env, hint, parent, orphan, + la->la_mode & S_IFMT); memset(dof, 0, sizeof(*dof)); - dof->dof_type = dt_mode_to_dft(type); + dof->dof_type = dt_mode_to_dft(S_IFDIR); rc = linkea_data_new(&ldata, &info->lti_linkea_buf2); if (rc != 0) GOTO(unlock1, rc); - rc = linkea_add_buf(&ldata, cname, pfid); + rc = linkea_add_buf(&ldata, cname, lfsck_dto2fid(parent)); if (rc != 0) GOTO(unlock1, rc); @@ -1158,25 +1393,57 @@ static int lfsck_namespace_create_orphan_local(const struct lu_env *env, if (IS_ERR(th)) GOTO(unlock1, rc = PTR_ERR(th)); - rc = dt_declare_create(env, child, la, hint, dof, th); - if (rc == 0 && S_ISDIR(type)) - rc = dt_declare_ref_add(env, child, th); + /* Sync the remote transaction to guarantee that the subsequent + * lock against the @orphan can find the @orphan in time. */ + if (dt_object_remote(orphan)) + th->th_sync = 1; + rc = dt_declare_create(env, orphan, la, hint, dof, th); if (rc != 0) GOTO(stop, rc); + if (unlikely(!dt_try_as_dir(env, orphan))) + GOTO(stop, rc = -ENOTDIR); + + rc = dt_declare_ref_add(env, orphan, th); + if (rc != 0) + GOTO(stop, rc); + + rec->rec_type = S_IFDIR; + rec->rec_fid = cfid; + rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec, + (const struct dt_key *)dot, th); + if (rc != 0) + GOTO(stop, rc); + + rec->rec_fid = lfsck_dto2fid(parent); + rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec, + (const struct dt_key *)dotdot, th); + if (rc != 0) + GOTO(stop, rc); + + if (lmv != NULL) { + lmv->lmv_magic = LMV_MAGIC; + lmv->lmv_master_mdt_index = lfsck_dev_idx(lfsck); + lfsck_lmv_header_cpu_to_le(lmv2, lmv); + lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2)); + rc = dt_declare_xattr_set(env, orphan, &lmv_buf, + XATTR_NAME_LMV, 0, th); + if (rc != 0) + GOTO(stop, rc); + } + lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf, ldata.ld_leh->leh_len); - rc = dt_declare_xattr_set(env, child, &linkea_buf, + rc = dt_declare_xattr_set(env, orphan, &linkea_buf, XATTR_NAME_LINK, 0, th); if (rc != 0) GOTO(stop, rc); - rec->rec_type = type; rec->rec_fid = cfid; rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec, (const struct dt_key *)name, th); - if (rc == 0 && S_ISDIR(type)) + if (rc == 0) rc = dt_declare_ref_add(env, parent, th); if (rc != 0) @@ -1186,45 +1453,43 @@ static int lfsck_namespace_create_orphan_local(const struct lu_env *env, if (rc != 0) GOTO(stop, rc); - dt_write_lock(env, child, 0); - rc = dt_create(env, child, la, hint, dof, th); + dt_write_lock(env, orphan, 0); + rc = dt_create(env, orphan, la, hint, dof, th); if (rc != 0) GOTO(unlock2, rc); - if (S_ISDIR(type)) { - if (unlikely(!dt_try_as_dir(env, child))) - GOTO(unlock2, rc = -ENOTDIR); + rc = dt_ref_add(env, orphan, th); + if (rc != 0) + GOTO(unlock2, rc); - rec->rec_type = S_IFDIR; - rec->rec_fid = cfid; - rc = dt_insert(env, child, (const struct dt_rec *)rec, - (const struct dt_key *)dot, th, BYPASS_CAPA, 1); - if (rc != 0) - GOTO(unlock2, rc); + rec->rec_fid = cfid; + rc = dt_insert(env, orphan, (const struct dt_rec *)rec, + (const struct dt_key *)dot, th, 1); + if (rc != 0) + GOTO(unlock2, rc); - rec->rec_fid = pfid; - rc = dt_insert(env, child, (const struct dt_rec *)rec, - (const struct dt_key *)dotdot, th, - BYPASS_CAPA, 1); - if (rc != 0) - GOTO(unlock2, rc); + rec->rec_fid = lfsck_dto2fid(parent); + rc = dt_insert(env, orphan, (const struct dt_rec *)rec, + (const struct dt_key *)dotdot, th, 1); + if (rc != 0) + GOTO(unlock2, rc); - rc = dt_ref_add(env, child, th); + if (lmv != NULL) { + rc = dt_xattr_set(env, orphan, &lmv_buf, XATTR_NAME_LMV, 0, th); if (rc != 0) GOTO(unlock2, rc); } - rc = dt_xattr_set(env, child, &linkea_buf, - XATTR_NAME_LINK, 0, th, BYPASS_CAPA); - dt_write_unlock(env, child); + rc = dt_xattr_set(env, orphan, &linkea_buf, + XATTR_NAME_LINK, 0, th); + dt_write_unlock(env, orphan); if (rc != 0) GOTO(stop, rc); - rec->rec_type = type; rec->rec_fid = cfid; rc = dt_insert(env, parent, (const struct dt_rec *)rec, - (const struct dt_key *)name, th, BYPASS_CAPA, 1); - if (rc == 0 && S_ISDIR(type)) { + (const struct dt_key *)name, th, 1); + if (rc == 0) { dt_write_lock(env, parent, 0); rc = dt_ref_add(env, parent, th); dt_write_unlock(env, parent); @@ -1233,61 +1498,24 @@ static int lfsck_namespace_create_orphan_local(const struct lu_env *env, GOTO(stop, rc = (rc == 0 ? 1 : rc)); unlock2: - dt_write_unlock(env, child); + dt_write_unlock(env, orphan); stop: - dt_trans_stop(env, dev, th); + rc1 = dt_trans_stop(env, dev, th); + if (rc1 != 0 && rc > 0) + rc = rc1; unlock1: - lfsck_ibits_unlock(&lh, LCK_EX); + lfsck_unlock(llh); log: - CDEBUG(D_LFSCK, "%s: namespace LFSCK create orphan locally for " - "the object "DFID", name = %s, type %o: rc = %d\n", + CDEBUG(D_LFSCK, "%s: namespace LFSCK create orphan dir for " + "the object "DFID", name = %s: rc = %d\n", lfsck_lfsck2name(lfsck), PFID(cfid), - cname->ln_name != NULL ? cname->ln_name : "", type, rc); - - if (child != NULL && !IS_ERR(child)) - lfsck_object_put(env, child); - - return rc; -} - -/** - * Create the specified orphan MDT-object. - * - * For the case that the parent MDT-object stored in some MDT-object's - * linkEA entry is lost, the LFSCK will re-create the parent object as - * an orphan and insert it into .lustre/lost+found/MDTxxxx/ directory - * with the name: ${FID}-P-${conflict_version}. - * - * \param[in] env pointer to the thread context - * \param[in] com pointer to the lfsck component - * \param[in] orphan pointer to the orphan MDT-object - * - * type "P": The orphan object to be created was a parent directory - * of some DMT-object which linkEA shows that the @orphan - * object is missing. - * - * \see lfsck_layout_recreate_parent() for more types. - * - * \retval positive number for repaired cases - * \retval 0 if needs to repair nothing - * \retval negative error number on failure - */ -static int lfsck_namespace_create_orphan(const struct lu_env *env, - struct lfsck_component *com, - struct dt_object *orphan) -{ - struct lfsck_namespace *ns = com->lc_file_ram; - int rc; + cname->ln_name != NULL ? cname->ln_name : "", rc); - if (dt_object_remote(orphan)) - rc = lfsck_namespace_create_orphan_remote(env, com, orphan, - S_IFDIR); - else - rc = lfsck_namespace_create_orphan_local(env, com, orphan, - S_IFDIR); + if (parent != NULL && !IS_ERR(parent) && parent != lfsck->li_lpf_obj) + lfsck_object_put(env, parent); if (rc != 0) ns->ln_flags |= LF_INCONSISTENT; @@ -1324,36 +1552,43 @@ static int lfsck_namespace_shrink_linkea(const struct lu_env *env, bool next) { struct lfsck_instance *lfsck = com->lc_lfsck; - struct dt_device *dev = lfsck->li_bottom; + struct dt_device *dev = lfsck_obj2dev(obj); struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram; struct thandle *th = NULL; struct lustre_handle lh = { 0 }; - struct linkea_data ldata_new = { 0 }; + struct linkea_data ldata_new = { NULL }; struct lu_buf linkea_buf; + int buflen = 0; int rc = 0; ENTRY; rc = lfsck_ibits_lock(env, lfsck, obj, &lh, - MDS_INODELOCK_UPDATE | - MDS_INODELOCK_XATTR, LCK_EX); + MDS_INODELOCK_UPDATE | MDS_INODELOCK_XATTR, + LCK_EX); if (rc != 0) GOTO(log, rc); if (next) - linkea_del_buf(ldata, cname); + lfsck_linkea_del_buf(ldata, cname); else lfsck_namespace_filter_linkea_entry(ldata, cname, pfid, true); - lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf, - ldata->ld_leh->leh_len); + if (ldata->ld_leh->leh_reccount > 0) { + lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf, + ldata->ld_leh->leh_len); + buflen = linkea_buf.lb_len; + } again: th = dt_trans_create(env, dev); if (IS_ERR(th)) GOTO(unlock1, rc = PTR_ERR(th)); - rc = dt_declare_xattr_set(env, obj, &linkea_buf, - XATTR_NAME_LINK, 0, th); + if (buflen != 0) + rc = dt_declare_xattr_set(env, obj, &linkea_buf, + XATTR_NAME_LINK, 0, th); + else + rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, th); if (rc != 0) GOTO(stop, rc); @@ -1367,7 +1602,8 @@ again: rc = lfsck_links_read2(env, obj, &ldata_new); if (rc != 0) - GOTO(unlock2, rc); + GOTO(unlock2, + rc = (rc == -ENODATA ? 0 : rc)); /* The specified linkEA entry has been removed by race. */ rc = linkea_links_find(&ldata_new, cname, pfid); @@ -1378,12 +1614,12 @@ again: GOTO(unlock2, rc = 1); if (next) - linkea_del_buf(&ldata_new, cname); + lfsck_linkea_del_buf(&ldata_new, cname); else lfsck_namespace_filter_linkea_entry(&ldata_new, cname, pfid, true); - if (linkea_buf.lb_len < ldata_new.ld_leh->leh_len) { + if (buflen < ldata_new.ld_leh->leh_len) { dt_write_unlock(env, obj); dt_trans_stop(env, dev, th); lfsck_buf_init(&linkea_buf, ldata_new.ld_buf->lb_buf, @@ -1391,10 +1627,14 @@ again: goto again; } - lfsck_buf_init(&linkea_buf, ldata_new.ld_buf->lb_buf, - ldata_new.ld_leh->leh_len); - rc = dt_xattr_set(env, obj, &linkea_buf, - XATTR_NAME_LINK, 0, th, BYPASS_CAPA); + if (ldata_new.ld_leh->leh_reccount > 0) { + lfsck_buf_init(&linkea_buf, ldata_new.ld_buf->lb_buf, + ldata_new.ld_leh->leh_len); + rc = dt_xattr_set(env, obj, &linkea_buf, + XATTR_NAME_LINK, 0, th); + } else { + rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, th); + } GOTO(unlock2, rc = (rc == 0 ? 1 : rc)); @@ -1450,20 +1690,21 @@ static int lfsck_namespace_shrink_linkea_cond(const struct lu_env *env, struct lu_name *cname, struct lu_fid *pfid) { - struct lu_fid *cfid = &lfsck_env_info(env)->lti_fid3; - struct lustre_handle lh = { 0 }; - int rc; + struct lfsck_thread_info *info = lfsck_env_info(env); + struct lu_fid *cfid = &info->lti_fid3; + struct lfsck_lock_handle *llh = &info->lti_llh; + int rc; ENTRY; - rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh, - MDS_INODELOCK_UPDATE, LCK_EX); + rc = lfsck_lock(env, com->lc_lfsck, parent, cname->ln_name, llh, + MDS_INODELOCK_UPDATE, LCK_PR); if (rc != 0) RETURN(rc); dt_read_lock(env, parent, 0); if (unlikely(lfsck_is_dead_obj(parent))) { dt_read_unlock(env, parent); - lfsck_ibits_unlock(&lh, LCK_EX); + lfsck_unlock(llh); rc = lfsck_namespace_shrink_linkea(env, com, child, ldata, cname, pfid, true); @@ -1471,8 +1712,7 @@ static int lfsck_namespace_shrink_linkea_cond(const struct lu_env *env, } rc = dt_lookup(env, parent, (struct dt_rec *)cfid, - (const struct dt_key *)cname->ln_name, - BYPASS_CAPA); + (const struct dt_key *)cname->ln_name); dt_read_unlock(env, parent); /* It is safe to release the ldlm lock, because when the logic come @@ -1482,7 +1722,7 @@ static int lfsck_namespace_shrink_linkea_cond(const struct lu_env *env, * has removed the specified linkEA entry by race, then it is OK, * because the subsequent lfsck_namespace_shrink_linkea() can handle * such case. */ - lfsck_ibits_unlock(&lh, LCK_EX); + lfsck_unlock(llh); if (rc == -ENOENT) { rc = lfsck_namespace_shrink_linkea(env, com, child, ldata, cname, pfid, true); @@ -1537,23 +1777,35 @@ static int lfsck_namespace_replace_cond(const struct lu_env *env, const struct lu_name *cname) { struct lfsck_thread_info *info = lfsck_env_info(env); - struct lu_fid *tfid = &info->lti_fid5; struct lu_attr *la = &info->lti_la; struct dt_insert_rec *rec = &info->lti_dt_rec; + struct lu_fid tfid; struct lfsck_instance *lfsck = com->lc_lfsck; + /* The child and its name may be on different MDTs. */ struct dt_device *dev = lfsck->li_next; const char *name = cname->ln_name; - struct dt_object *obj = NULL; - struct lustre_handle plh = { 0 }; + const struct lu_fid *pfid = lfsck_dto2fid(parent); + struct dt_object *cobj = NULL; + struct lfsck_lock_handle *pllh = &info->lti_llh; struct lustre_handle clh = { 0 }; - struct linkea_data ldata = { 0 }; + struct linkea_data ldata = { NULL }; struct thandle *th = NULL; bool exist = true; int rc = 0; ENTRY; - rc = lfsck_ibits_lock(env, lfsck, parent, &plh, - MDS_INODELOCK_UPDATE, LCK_EX); + /* @parent/@child may be based on lfsck->li_bottom, + * but here we need the object based on the lfsck->li_next. */ + + parent = lfsck_object_locate(dev, parent); + if (IS_ERR(parent)) + GOTO(log, rc = PTR_ERR(parent)); + + if (unlikely(!dt_try_as_dir(env, parent))) + GOTO(log, rc = -ENOTDIR); + + rc = lfsck_lock(env, lfsck, parent, name, pllh, + MDS_INODELOCK_UPDATE, LCK_PW); if (rc != 0) GOTO(log, rc); @@ -1562,9 +1814,9 @@ static int lfsck_namespace_replace_cond(const struct lu_env *env, goto replace; } - obj = lfsck_object_find(env, lfsck, cfid); - if (IS_ERR(obj)) { - rc = PTR_ERR(obj); + cobj = lfsck_object_find_by_dev(env, dev, cfid); + if (IS_ERR(cobj)) { + rc = PTR_ERR(cobj); if (rc == -ENOENT) { exist = false; goto replace; @@ -1573,13 +1825,13 @@ static int lfsck_namespace_replace_cond(const struct lu_env *env, GOTO(log, rc); } - if (!dt_object_exists(obj)) { + if (!dt_object_exists(cobj)) { exist = false; goto replace; } - rc = dt_lookup(env, parent, (struct dt_rec *)tfid, - (const struct dt_key *)name, BYPASS_CAPA); + rc = dt_lookup(env, parent, (struct dt_rec *)&tfid, + (const struct dt_key *)name); if (rc == -ENOENT) { exist = false; goto replace; @@ -1589,22 +1841,23 @@ static int lfsck_namespace_replace_cond(const struct lu_env *env, GOTO(log, rc); /* Someone changed the name entry, cannot replace it. */ - if (!lu_fid_eq(cfid, tfid)) + if (!lu_fid_eq(cfid, &tfid)) GOTO(log, rc = 0); /* lock the object to be destroyed. */ - rc = lfsck_ibits_lock(env, lfsck, obj, &clh, + rc = lfsck_ibits_lock(env, lfsck, cobj, &clh, MDS_INODELOCK_UPDATE | - MDS_INODELOCK_XATTR, LCK_EX); + MDS_INODELOCK_UPDATE | MDS_INODELOCK_XATTR, + LCK_EX); if (rc != 0) GOTO(log, rc); - if (unlikely(lfsck_is_dead_obj(obj))) { + if (unlikely(lfsck_is_dead_obj(cobj))) { exist = false; goto replace; } - rc = dt_attr_get(env, obj, la, BYPASS_CAPA); + rc = dt_attr_get(env, cobj, la); if (rc != 0) GOTO(log, rc); @@ -1614,8 +1867,7 @@ static int lfsck_namespace_replace_cond(const struct lu_env *env, GOTO(log, rc); if (S_ISREG(la->la_mode)) { - rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LOV, - BYPASS_CAPA); + rc = dt_xattr_get(env, cobj, &LU_BUF_NULL, XATTR_NAME_LOV); /* If someone has created related OST-object(s), * then keep it. */ if ((rc > 0) || (rc < 0 && rc != -ENODATA)) @@ -1634,7 +1886,7 @@ replace: if (rc != 0) GOTO(log, rc); - rc = linkea_links_find(&ldata, cname, lfsck_dto2fid(parent)); + rc = linkea_links_find(&ldata, cname, pfid); /* Someone moved the child, no need to replace. */ if (rc != 0) GOTO(log, rc = 0); @@ -1647,7 +1899,7 @@ replace: GOTO(log, rc = PTR_ERR(th)); if (exist) { - rc = dt_declare_destroy(env, obj, th); + rc = dt_declare_destroy(env, cobj, th); if (rc != 0) GOTO(stop, rc); } @@ -1663,22 +1915,23 @@ replace: if (rc != 0) GOTO(stop, rc); - rc = dt_trans_start(env, dev, th); + rc = dt_trans_start_local(env, dev, th); if (rc != 0) GOTO(stop, rc); if (exist) { - rc = dt_destroy(env, obj, th); + rc = dt_destroy(env, cobj, th); if (rc != 0) GOTO(stop, rc); } /* The old name entry maybe not exist. */ - dt_delete(env, parent, (const struct dt_key *)name, th, - BYPASS_CAPA); + rc = dt_delete(env, parent, (const struct dt_key *)name, th); + if (rc != 0 && rc != -ENOENT) + GOTO(stop, rc); rc = dt_insert(env, parent, (const struct dt_rec *)rec, - (const struct dt_key *)name, th, BYPASS_CAPA, 1); + (const struct dt_key *)name, th, 1); GOTO(stop, rc = (rc == 0 ? 1 : rc)); @@ -1687,16 +1940,16 @@ stop: log: lfsck_ibits_unlock(&clh, LCK_EX); - lfsck_ibits_unlock(&plh, LCK_EX); - if (obj != NULL && !IS_ERR(obj)) - lfsck_object_put(env, obj); + lfsck_unlock(pllh); + + if (cobj != NULL && !IS_ERR(cobj)) + lfsck_object_put(env, cobj); CDEBUG(D_LFSCK, "%s: namespace LFSCK conditionally destroy the " "object "DFID" because of conflict with the object "DFID " under the parent "DFID" with name %s: rc = %d\n", lfsck_lfsck2name(lfsck), PFID(cfid), - PFID(lfsck_dto2fid(child)), PFID(lfsck_dto2fid(parent)), - name, rc); + PFID(lfsck_dto2fid(child)), PFID(pfid), name, rc); return rc; } @@ -1721,14 +1974,12 @@ int lfsck_namespace_rebuild_linkea(const struct lu_env *env, struct linkea_data *ldata) { struct lfsck_instance *lfsck = com->lc_lfsck; - struct dt_device *dev = lfsck->li_bottom; + struct dt_device *dev = lfsck_obj2dev(obj); struct thandle *th = NULL; struct lu_buf linkea_buf; int rc = 0; ENTRY; - LASSERT(!dt_object_remote(obj)); - th = dt_trans_create(env, dev); if (IS_ERR(th)) GOTO(log, rc = PTR_ERR(th)); @@ -1752,7 +2003,7 @@ int lfsck_namespace_rebuild_linkea(const struct lu_env *env, GOTO(unlock, rc = 1); rc = dt_xattr_set(env, obj, &linkea_buf, - XATTR_NAME_LINK, 0, th, BYPASS_CAPA); + XATTR_NAME_LINK, 0, th); GOTO(unlock, rc = (rc == 0 ? 1 : rc)); @@ -1806,20 +2057,30 @@ int lfsck_namespace_repair_dirent(const struct lu_env *env, { struct lfsck_thread_info *info = lfsck_env_info(env); struct dt_insert_rec *rec = &info->lti_dt_rec; + const struct lu_fid *pfid = lfsck_dto2fid(parent); const struct lu_fid *cfid = lfsck_dto2fid(child); - struct lu_fid *tfid = &info->lti_fid5; + struct lu_fid tfid; struct lfsck_instance *lfsck = com->lc_lfsck; struct dt_device *dev = lfsck->li_next; struct thandle *th = NULL; + struct lfsck_lock_handle *llh = &info->lti_llh; struct lustre_handle lh = { 0 }; int rc = 0; ENTRY; + parent = lfsck_object_locate(dev, parent); + if (IS_ERR(parent)) + GOTO(log, rc = PTR_ERR(parent)); + if (unlikely(!dt_try_as_dir(env, parent))) GOTO(log, rc = -ENOTDIR); - rc = lfsck_ibits_lock(env, lfsck, parent, &lh, - MDS_INODELOCK_UPDATE, LCK_EX); + if (!update || strcmp(name, name2) == 0) + rc = lfsck_lock(env, lfsck, parent, name, llh, + MDS_INODELOCK_UPDATE, LCK_PW); + else + rc = lfsck_ibits_lock(env, lfsck, parent, &lh, + MDS_INODELOCK_UPDATE, LCK_PW); if (rc != 0) GOTO(log, rc); @@ -1841,19 +2102,20 @@ int lfsck_namespace_repair_dirent(const struct lu_env *env, GOTO(stop, rc); } - if (dec) { + if (dec && S_ISDIR(type)) { rc = dt_declare_ref_del(env, parent, th); if (rc != 0) GOTO(stop, rc); } - rc = dt_trans_start(env, dev, th); + rc = dt_trans_start_local(env, dev, th); if (rc != 0) GOTO(stop, rc); + dt_write_lock(env, parent, 0); - rc = dt_lookup(env, parent, (struct dt_rec *)tfid, - (const struct dt_key *)name, BYPASS_CAPA); + rc = dt_lookup(env, parent, (struct dt_rec *)&tfid, + (const struct dt_key *)name); /* Someone has removed the bad name entry by race. */ if (rc == -ENOENT) GOTO(unlock2, rc = 0); @@ -1863,27 +2125,25 @@ int lfsck_namespace_repair_dirent(const struct lu_env *env, /* Someone has removed the bad name entry and reused it for other * object by race. */ - if (!lu_fid_eq(tfid, cfid)) + if (!lu_fid_eq(&tfid, cfid)) GOTO(unlock2, rc = 0); if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN) GOTO(unlock2, rc = 1); - rc = dt_delete(env, parent, (const struct dt_key *)name, th, - BYPASS_CAPA); + rc = dt_delete(env, parent, (const struct dt_key *)name, th); if (rc != 0) GOTO(unlock2, rc); if (update) { rc = dt_insert(env, parent, (const struct dt_rec *)rec, - (const struct dt_key *)name2, th, - BYPASS_CAPA, 1); + (const struct dt_key *)name2, th, 1); if (rc != 0) GOTO(unlock2, rc); } - if (dec) { + if (dec && S_ISDIR(type)) { rc = dt_ref_del(env, parent, th); if (rc != 0) GOTO(unlock2, rc); @@ -1898,21 +2158,23 @@ stop: dt_trans_stop(env, dev, th); /* We are not sure whether the child will become orphan or not. - * Record it in the LFSCK tracing file for further checking in + * Record it in the LFSCK trace file for further checking in * the second-stage scanning. */ if (!update && !dec && rc == 0) lfsck_namespace_trace_update(env, com, cfid, LNTF_CHECK_LINKEA, true); unlock1: - lfsck_ibits_unlock(&lh, LCK_EX); + /* It is harmless even if unlock the unused lock_handle */ + lfsck_ibits_unlock(&lh, LCK_PW); + lfsck_unlock(llh); log: CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found bad name " "entry for: parent "DFID", child "DFID", name %s, type " "in name entry %o, type claimed by child %o. repair it " - "by %s with new name2 %s: rc = %d\n", lfsck_lfsck2name(lfsck), - PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)), + "by %s with new name2 %s: rc = %d\n", + lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid), name, type, update ? lfsck_object_type(child) : 0, update ? "updating" : "removing", name2, rc); @@ -1952,9 +2214,9 @@ static int lfsck_namespace_repair_unmatched_pairs(const struct lu_env *env, struct lfsck_thread_info *info = lfsck_env_info(env); struct dt_insert_rec *rec = &info->lti_dt_rec; struct lfsck_instance *lfsck = com->lc_lfsck; - struct dt_device *dev = lfsck->li_bottom; + struct dt_device *dev = lfsck_obj2dev(obj); struct thandle *th = NULL; - struct linkea_data ldata = { 0 }; + struct linkea_data ldata = { NULL }; struct lu_buf linkea_buf; int rc = 0; ENTRY; @@ -2005,16 +2267,15 @@ static int lfsck_namespace_repair_unmatched_pairs(const struct lu_env *env, GOTO(unlock, rc = 1); /* The old ".." name entry maybe not exist. */ - dt_delete(env, obj, (const struct dt_key *)dotdot, th, - BYPASS_CAPA); + dt_delete(env, obj, (const struct dt_key *)dotdot, th); rc = dt_insert(env, obj, (const struct dt_rec *)rec, - (const struct dt_key *)dotdot, th, BYPASS_CAPA, 1); + (const struct dt_key *)dotdot, th, 1); if (rc != 0) GOTO(unlock, rc); rc = dt_xattr_set(env, obj, &linkea_buf, - XATTR_NAME_LINK, 0, th, BYPASS_CAPA); + XATTR_NAME_LINK, 0, th); GOTO(unlock, rc = (rc == 0 ? 1 : rc)); @@ -2068,6 +2329,7 @@ lfsck_namespace_dsd_orphan(const struct lu_env *env, enum lfsck_namespace_inconsistency_type *type) { struct lfsck_thread_info *info = lfsck_env_info(env); + struct lfsck_namespace *ns = com->lc_file_ram; int rc; ENTRY; @@ -2078,6 +2340,17 @@ lfsck_namespace_dsd_orphan(const struct lu_env *env, RETURN(rc); *type = LNIT_MUL_REF; + + /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has + * ever tried to verify some remote MDT-object that resides on this + * MDT, but this MDT failed to respond such request. So means there + * may be some remote name entry on other MDT that references this + * object with another name, so we cannot know whether this linkEA + * is valid or not. So keep it there and maybe resolved when next + * LFSCK run. */ + if (ns->ln_flags & LF_INCOMPLETE) + RETURN(0); + /* The unique linkEA is invalid, even if the ".." name entry may be * valid, we still cannot know via which name entry this directory * will be referenced. Then handle it as pure orphan. */ @@ -2107,6 +2380,7 @@ lfsck_namespace_dsd_orphan(const struct lu_env *env, * \param[out] type to tell the caller what the inconsistency is * \param[in] retry if found inconsistency, but the caller does not hold * ldlm lock on the @child, then set @retry as true + * \param[in] unknown set if does not know how to repair the inconsistency * * \retval positive number for repaired cases * \retval 0 if nothing to be repaired @@ -2120,20 +2394,24 @@ lfsck_namespace_dsd_single(const struct lu_env *env, struct linkea_data *ldata, struct lustre_handle *lh, enum lfsck_namespace_inconsistency_type *type, - bool *retry) + bool *retry, bool *unknown) { struct lfsck_thread_info *info = lfsck_env_info(env); struct lu_name *cname = &info->lti_name; const struct lu_fid *cfid = lfsck_dto2fid(child); - struct lu_fid *tfid = &info->lti_fid3; + struct lu_fid tfid; + struct lfsck_namespace *ns = com->lc_file_ram; struct lfsck_instance *lfsck = com->lc_lfsck; struct dt_object *parent = NULL; + struct lmv_mds_md_v1 *lmv; int rc = 0; ENTRY; - lfsck_namespace_unpack_linkea_entry(ldata, cname, tfid, info->lti_key); + rc = lfsck_namespace_unpack_linkea_entry(ldata, cname, &tfid, + info->lti_key, + sizeof(info->lti_key)); /* The unique linkEA entry with bad parent will be handled as orphan. */ - if (!fid_is_sane(tfid)) { + if (rc != 0) { if (!lustre_handle_is_used(lh) && retry != NULL) *retry = true; else @@ -2143,7 +2421,7 @@ lfsck_namespace_dsd_single(const struct lu_env *env, GOTO(out, rc); } - parent = lfsck_object_find_bottom(env, lfsck, tfid); + parent = lfsck_object_find_bottom(env, lfsck, &tfid); if (IS_ERR(parent)) GOTO(out, rc = PTR_ERR(parent)); @@ -2153,6 +2431,16 @@ lfsck_namespace_dsd_single(const struct lu_env *env, * name entry the child will be referenced, since all known entries * have been verified during the first-stage scanning. */ if (!dt_object_exists(parent)) { + /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT + * has ever tried to verify some remote MDT-object that resides + * on this MDT, but this MDT failed to respond such request. So + * means there may be some remote name entry on other MDT that + * references this object with another name, so we cannot know + * whether this linkEA is valid or not. So keep it there and + * maybe resolved when next LFSCK run. */ + if (ns->ln_flags & LF_INCOMPLETE) + GOTO(out, rc = 0); + if (!lustre_handle_is_used(lh) && retry != NULL) { *retry = true; @@ -2162,8 +2450,33 @@ lfsck_namespace_dsd_single(const struct lu_env *env, lfsck_ibits_unlock(lh, LCK_EX); lost_parent: + lmv = &info->lti_lmv; + rc = lfsck_read_stripe_lmv(env, child, lmv); + if (rc != 0 && rc != -ENODATA) + GOTO(out, rc); + + if (rc == -ENODATA || lmv->lmv_magic != LMV_MAGIC_STRIPE) { + lmv = NULL; + } else if (lfsck_shard_name_to_index(env, + cname->ln_name, cname->ln_namelen, + S_IFDIR, cfid) < 0) { + /* It is an invalid name entry, we + * cannot trust the parent also. */ + rc = lfsck_namespace_shrink_linkea(env, com, child, + ldata, cname, &tfid, true); + if (rc < 0) + GOTO(out, rc); + + snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf), + "-"DFID, PFID(pfid)); + rc = lfsck_namespace_insert_orphan(env, com, child, + info->lti_tmpbuf, "S", NULL); + + GOTO(out, rc); + } + /* Create the lost parent as an orphan. */ - rc = lfsck_namespace_create_orphan(env, com, parent); + rc = lfsck_namespace_create_orphan_dir(env, com, parent, lmv); if (rc >= 0) { /* Add the missing name entry to the parent. */ rc = lfsck_namespace_insert_normal(env, com, parent, @@ -2179,7 +2492,7 @@ lost_parent: * current system to be consistent. */ rc = lfsck_namespace_shrink_linkea(env, com, child, ldata, - cname, tfid, true); + cname, &tfid, true); if (rc >= 0) { snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf), @@ -2192,7 +2505,7 @@ lost_parent: } GOTO(out, rc); - } + } /* !dt_object_exists(parent) */ /* The unique linkEA entry with bad parent will be handled as orphan. */ if (unlikely(!dt_try_as_dir(env, parent))) { @@ -2205,9 +2518,19 @@ lost_parent: GOTO(out, rc); } - rc = dt_lookup(env, parent, (struct dt_rec *)tfid, - (const struct dt_key *)cname->ln_name, BYPASS_CAPA); + rc = dt_lookup(env, parent, (struct dt_rec *)&tfid, + (const struct dt_key *)cname->ln_name); if (rc == -ENOENT) { + /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT + * has ever tried to verify some remote MDT-object that resides + * on this MDT, but this MDT failed to respond such request. So + * means there may be some remote name entry on other MDT that + * references this object with another name, so we cannot know + * whether this linkEA is valid or not. So keep it there and + * maybe resolved when next LFSCK run. */ + if (ns->ln_flags & LF_INCOMPLETE) + GOTO(out, rc = 0); + if (!lustre_handle_is_used(lh) && retry != NULL) { *retry = true; @@ -2215,6 +2538,28 @@ lost_parent: } lfsck_ibits_unlock(lh, LCK_EX); + rc = lfsck_namespace_check_name(env, parent, child, cname); + if (rc == -ENOENT) + goto lost_parent; + + if (rc < 0) + GOTO(out, rc); + + /* It is an invalid name entry, drop it. */ + if (unlikely(rc > 0)) { + rc = lfsck_namespace_shrink_linkea(env, com, child, + ldata, cname, &tfid, true); + if (rc >= 0) { + snprintf(info->lti_tmpbuf, + sizeof(info->lti_tmpbuf), + "-"DFID, PFID(pfid)); + rc = lfsck_namespace_insert_orphan(env, com, + child, info->lti_tmpbuf, "D", NULL); + } + + GOTO(out, rc); + } + /* Add the missing name entry back to the namespace. */ rc = lfsck_namespace_insert_normal(env, com, parent, child, cname->ln_name); @@ -2235,7 +2580,7 @@ lost_parent: * internal status of create operation. Under such * case, nothing to be done. */ rc = lfsck_namespace_shrink_linkea_cond(env, com, - parent, child, ldata, cname, tfid); + parent, child, ldata, cname, &tfid); if (rc >= 0) { snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf), @@ -2246,12 +2591,12 @@ lost_parent: } GOTO(out, rc); - } + } /* rc == -ENOENT */ if (rc != 0) GOTO(out, rc); - if (!lu_fid_eq(tfid, cfid)) { + if (!lu_fid_eq(&tfid, cfid)) { if (!lustre_handle_is_used(lh) && retry != NULL) { *retry = true; @@ -2263,7 +2608,7 @@ lost_parent: * may be created by the LFSCK for repairing dangling * name entry. Try to replace it. */ rc = lfsck_namespace_replace_cond(env, com, parent, child, - tfid, cname); + &tfid, cname); if (rc == 0) rc = lfsck_namespace_dsd_orphan(env, com, child, pfid, lh, type); @@ -2271,6 +2616,19 @@ lost_parent: GOTO(out, rc); } + /* Zero FID may because the remote directroy object has invalid linkEA, + * or lost linkEA. Under such case, the LFSCK on this MDT does not know + * how to repair the inconsistency, but the namespace LFSCK on the MDT + * where its name entry resides may has more information (name, FID) to + * repair such inconsistency. So here, keep the inconsistency to avoid + * some imporper repairing. */ + if (fid_is_zero(pfid)) { + if (unknown) + *unknown = true; + + GOTO(out, rc = 0); + } + /* The ".." name entry is wrong, update it. */ if (!lu_fid_eq(pfid, lfsck_dto2fid(parent))) { if (!lustre_handle_is_used(lh) && retry != NULL) { @@ -2311,6 +2669,7 @@ out: * \param[in,out] lh ldlm lock handler for the given @child * \param[out] type to tell the caller what the inconsistency is * \param[in] lpf true if the ".." entry is under lost+found/MDTxxxx/ + * \param[in] unknown set if does not know how to repair the inconsistency * * \retval positive number for repaired cases * \retval 0 if nothing to be repaired @@ -2324,35 +2683,37 @@ lfsck_namespace_dsd_multiple(const struct lu_env *env, struct linkea_data *ldata, struct lustre_handle *lh, enum lfsck_namespace_inconsistency_type *type, - bool lpf) + bool lpf, bool *unknown) { struct lfsck_thread_info *info = lfsck_env_info(env); struct lu_name *cname = &info->lti_name; const struct lu_fid *cfid = lfsck_dto2fid(child); - struct lu_fid *tfid = &info->lti_fid3; - struct lu_fid *pfid2 = &info->lti_fid4; + struct lu_fid *pfid2 = &info->lti_fid3; + struct lu_fid tfid; struct lfsck_namespace *ns = com->lc_file_ram; struct lfsck_instance *lfsck = com->lc_lfsck; struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram; struct dt_object *parent = NULL; - struct linkea_data ldata_new = { 0 }; - int count = 0; + struct linkea_data ldata_new = { NULL }; + int dirent_count = 0; int rc = 0; bool once = true; ENTRY; again: while (ldata->ld_lee != NULL) { - lfsck_namespace_unpack_linkea_entry(ldata, cname, tfid, - info->lti_key); - /* Drop repeated linkEA entries. */ - lfsck_namespace_filter_linkea_entry(ldata, cname, tfid, true); + rc = lfsck_namespace_unpack_linkea_entry(ldata, cname, &tfid, + info->lti_key, + sizeof(info->lti_key)); /* Drop invalid linkEA entry. */ - if (!fid_is_sane(tfid)) { - linkea_del_buf(ldata, cname); + if (rc != 0) { + lfsck_linkea_del_buf(ldata, cname); continue; } + /* Drop repeated linkEA entries. */ + lfsck_namespace_filter_linkea_entry(ldata, cname, &tfid, true); + /* If current dotdot is the .lustre/lost+found/MDTxxxx/, * then it is possible that: the directry object has ever * been lost, but its name entry was there. In the former @@ -2366,12 +2727,12 @@ again: * When the LFSCK runs again, if the dangling name is still * there, the LFSCK should move the orphan directory object * back to the normal namespace. */ - if (!lpf && !lu_fid_eq(pfid, tfid) && once) { + if (!lpf && !lu_fid_eq(pfid, &tfid) && once) { linkea_next_entry(ldata); continue; } - parent = lfsck_object_find_bottom(env, lfsck, tfid); + parent = lfsck_object_find_bottom(env, lfsck, &tfid); if (IS_ERR(parent)) RETURN(PTR_ERR(parent)); @@ -2382,7 +2743,7 @@ again: * there is still other chance to make the * child to be visible via other parent, then * remove this linkEA entry. */ - linkea_del_buf(ldata, cname); + lfsck_linkea_del_buf(ldata, cname); continue; } @@ -2392,13 +2753,12 @@ again: /* The linkEA entry with bad parent will be removed. */ if (unlikely(!dt_try_as_dir(env, parent))) { lfsck_object_put(env, parent); - linkea_del_buf(ldata, cname); + lfsck_linkea_del_buf(ldata, cname); continue; } - rc = dt_lookup(env, parent, (struct dt_rec *)tfid, - (const struct dt_key *)cname->ln_name, - BYPASS_CAPA); + rc = dt_lookup(env, parent, (struct dt_rec *)&tfid, + (const struct dt_key *)cname->ln_name); *pfid2 = *lfsck_dto2fid(parent); if (rc == -ENOENT) { lfsck_object_put(env, parent); @@ -2412,8 +2772,12 @@ again: RETURN(rc); } - if (lu_fid_eq(tfid, cfid)) { + if (lu_fid_eq(&tfid, cfid)) { lfsck_object_put(env, parent); + /* If the parent (that is declared via linkEA entry) + * directory contains the specified child, but such + * parent does not match the dotdot name entry, then + * trust the linkEA. */ if (!lu_fid_eq(pfid, pfid2)) { *type = LNIT_UNMATCHED_PAIRS; rc = lfsck_namespace_repair_unmatched_pairs(env, @@ -2439,18 +2803,19 @@ rebuild: if (rc < 0) RETURN(rc); - linkea_del_buf(ldata, cname); + lfsck_linkea_del_buf(ldata, cname); linkea_first_entry(ldata); /* There may be some invalid dangling name entries under * other parent directories, remove all of them. */ while (ldata->ld_lee != NULL) { - lfsck_namespace_unpack_linkea_entry(ldata, - cname, tfid, info->lti_key); - if (!fid_is_sane(tfid)) + rc = lfsck_namespace_unpack_linkea_entry(ldata, + cname, &tfid, info->lti_key, + sizeof(info->lti_key)); + if (rc != 0) goto next; parent = lfsck_object_find_bottom(env, lfsck, - tfid); + &tfid); if (IS_ERR(parent)) { rc = PTR_ERR(parent); if (rc != -ENOENT && @@ -2476,23 +2841,23 @@ rebuild: goto next; } - count += rc; + dirent_count += rc; next: - linkea_del_buf(ldata, cname); + lfsck_linkea_del_buf(ldata, cname); } - ns->ln_dirent_repaired += count; + ns->ln_dirent_repaired += dirent_count; RETURN(rc); - } + } /* lu_fid_eq(&tfid, lfsck_dto2fid(child)) */ lfsck_ibits_unlock(lh, LCK_EX); /* The name entry references another MDT-object that may be * created by the LFSCK for repairing dangling name entry. * Try to replace it. */ rc = lfsck_namespace_replace_cond(env, com, parent, child, - tfid, cname); + &tfid, cname); lfsck_object_put(env, parent); if (rc < 0) RETURN(rc); @@ -2500,12 +2865,13 @@ next: if (rc > 0) goto rebuild; - linkea_del_buf(ldata, cname); - } + lfsck_linkea_del_buf(ldata, cname); + } /* while (ldata->ld_lee != NULL) */ + linkea_first_entry(ldata); if (ldata->ld_leh->leh_reccount == 1) { rc = lfsck_namespace_dsd_single(env, com, child, pfid, ldata, - lh, type, NULL); + lh, type, NULL, unknown); RETURN(rc); } @@ -2519,7 +2885,6 @@ next: RETURN(rc); } - linkea_first_entry(ldata); /* If the dangling name entry for the orphan directory object has * been remvoed, then just check whether the directory object is * still under the .lustre/lost+found/MDTxxxx/ or not. */ @@ -2540,71 +2905,206 @@ next: } /** - * Double scan the directory object for namespace LFSCK. - * - * This function will verify the pairs in the namespace tree: - * the parent references the child via some name entry that should be in the - * child's linkEA entry, the child should back references the parent via its - * ".." name entry. - * - * The LFSCK will scan every linkEA entry in turn until find out the first - * matched pairs. If found, then all other linkEA entries will be dropped. - * If all the linkEA entries cannot match the ".." name entry, then there - * are serveral possible cases: - * - * 1) If there is only one linkEA entry, then trust it as long as the PFID - * in the linkEA entry is valid. - * - * 2) If there are multiple linkEA entries, then try to find the linkEA - * that matches the ".." name entry. If found, then all other entries - * are invalid; otherwise, it is quite possible that the ".." name entry - * is corrupted. Under such case, the LFSCK will rebuild the ".." name - * entry according to the first valid linkEA entry (both the parent and - * the name entry should exist). + * Repair the object's nlink attribute. * - * 3) If the directory object has no (valid) linkEA entry, then the - * directory object will be handled as pure orphan and inserted - * in the .lustre/lost+found/MDTxxxx/ with the name: - * ${self_FID}-${PFID}-D-${conflict_version} + * If all the known name entries have been verified, then the object's hard + * link attribute should match the object's linkEA entries count unless the + * object's has too much hard link to be recorded in the linkEA. Such cases + * should have been marked in the LFSCK trace file. Otherwise, trust the + * linkEA to update the object's nlink attribute. * * \param[in] env pointer to the thread context * \param[in] com pointer to the lfsck component - * \param[in] child pointer to the directory object to be handled - * \param[in] flags to indicate the specical checking on the @child + * \param[in] obj pointer to the dt_object to be handled + * \param[in,out] la pointer to buffer to object's attribute before + * and after the repairing * * \retval positive number for repaired cases * \retval 0 if nothing to be repaired * \retval negative error number on failure */ -static int lfsck_namespace_double_scan_dir(const struct lu_env *env, - struct lfsck_component *com, - struct dt_object *child, __u8 flags) +static int lfsck_namespace_repair_nlink(const struct lu_env *env, + struct lfsck_component *com, + struct dt_object *obj, + struct lu_attr *la) { - struct lfsck_thread_info *info = lfsck_env_info(env); - const struct lu_fid *cfid = lfsck_dto2fid(child); - struct lu_fid *pfid = &info->lti_fid2; - struct lfsck_namespace *ns = com->lc_file_ram; - struct lfsck_instance *lfsck = com->lc_lfsck; - struct lustre_handle lh = { 0 }; - struct linkea_data ldata = { 0 }; - bool unknown = false; - bool lpf = false; - bool retry = false; - enum lfsck_namespace_inconsistency_type type = LNIT_BAD_LINKEA; - int rc = 0; + struct lfsck_thread_info *info = lfsck_env_info(env); + struct lu_fid *tfid = &info->lti_fid3; + struct lfsck_namespace *ns = com->lc_file_ram; + struct lfsck_instance *lfsck = com->lc_lfsck; + struct dt_device *dev = lfsck_obj2dev(obj); + const struct lu_fid *cfid = lfsck_dto2fid(obj); + struct thandle *th = NULL; + struct linkea_data ldata = { NULL }; + struct lustre_handle lh = { 0 }; + __u32 old = la->la_nlink; + int idx; + int rc = 0; + __u8 flags; ENTRY; - LASSERT(!dt_object_remote(child)); + LASSERT(!dt_object_remote(obj)); - if (!(lfsck->li_bookmark_ram.lb_param & LPF_ALL_TGT)) { - CDEBUG(D_LFSCK, "%s: some MDT(s) maybe NOT take part in the" - "the namespace LFSCK, then the LFSCK cannot guarantee" - "all the name entries have been verified in first-stage" - "scanning. So have to skip orphan related handling for" - "the directory object "DFID" with remote name entry\n", - lfsck_lfsck2name(lfsck), PFID(cfid)); + rc = lfsck_ibits_lock(env, lfsck, obj, &lh, + MDS_INODELOCK_UPDATE, LCK_PW); + if (rc != 0) + GOTO(log, rc); - RETURN(0); + th = dt_trans_create(env, dev); + if (IS_ERR(th)) + GOTO(log, rc = PTR_ERR(th)); + + la->la_valid = LA_NLINK; + rc = dt_declare_attr_set(env, obj, la, th); + if (rc != 0) + GOTO(stop, rc); + + rc = dt_trans_start_local(env, dev, th); + if (rc != 0) + GOTO(stop, rc); + + dt_write_lock(env, obj, 0); + /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has + * ever tried to verify some remote MDT-object that resides on this + * MDT, but this MDT failed to respond such request. So means there + * may be some remote name entry on other MDT that references this + * object with another name, so we cannot know whether this linkEA + * is valid or not. So keep it there and maybe resolved when next + * LFSCK run. */ + if (ns->ln_flags & LF_INCOMPLETE) + GOTO(unlock, rc = 0); + + fid_cpu_to_be(tfid, cfid); + idx = lfsck_sub_trace_file_fid2idx(cfid); + rc = dt_lookup(env, com->lc_sub_trace_objs[idx].lsto_obj, + (struct dt_rec *)&flags, (const struct dt_key *)tfid); + if (rc != 0) + GOTO(unlock, rc); + + if (flags & LNTF_SKIP_NLINK) + GOTO(unlock, rc = 0); + + rc = dt_attr_get(env, obj, la); + if (rc != 0) + GOTO(unlock, rc = (rc == -ENOENT ? 0 : rc)); + + rc = lfsck_links_read2(env, obj, &ldata); + if (rc != 0) + GOTO(unlock, rc = (rc == -ENODATA ? 0 : rc)); + + if (la->la_nlink == ldata.ld_leh->leh_reccount || + unlikely(la->la_nlink == 0)) + GOTO(unlock, rc = 0); + + la->la_nlink = ldata.ld_leh->leh_reccount; + if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN) + GOTO(unlock, rc = 1); + + rc = dt_attr_set(env, obj, la, th); + + GOTO(unlock, rc = (rc == 0 ? 1 : rc)); + +unlock: + dt_write_unlock(env, obj); + +stop: + dt_trans_stop(env, dev, th); + +log: + lfsck_ibits_unlock(&lh, LCK_PW); + CDEBUG(D_LFSCK, "%s: namespace LFSCK repaired the object "DFID"'s " + "nlink count from %u to %u: rc = %d\n", + lfsck_lfsck2name(lfsck), PFID(cfid), old, la->la_nlink, rc); + + if (rc != 0) + ns->ln_flags |= LF_INCONSISTENT; + + return rc; +} + +/** + * Double scan the directory object for namespace LFSCK. + * + * This function will verify the pairs in the namespace tree: + * the parent references the child via some name entry that should be in the + * child's linkEA entry, the child should back references the parent via its + * ".." name entry. + * + * The LFSCK will scan every linkEA entry in turn until find out the first + * matched pairs. If found, then all other linkEA entries will be dropped. + * If all the linkEA entries cannot match the ".." name entry, then there + * are serveral possible cases: + * + * 1) If there is only one linkEA entry, then trust it as long as the PFID + * in the linkEA entry is valid. + * + * 2) If there are multiple linkEA entries, then try to find the linkEA + * that matches the ".." name entry. If found, then all other entries + * are invalid; otherwise, it is quite possible that the ".." name entry + * is corrupted. Under such case, the LFSCK will rebuild the ".." name + * entry according to the first valid linkEA entry (both the parent and + * the name entry should exist). + * + * 3) If the directory object has no (valid) linkEA entry, then the + * directory object will be handled as pure orphan and inserted + * in the .lustre/lost+found/MDTxxxx/ with the name: + * ${self_FID}-${PFID}-D-${conflict_version} + * + * \param[in] env pointer to the thread context + * \param[in] com pointer to the lfsck component + * \param[in] child pointer to the directory object to be handled + * \param[in] flags to indicate the specical checking on the @child + * + * \retval positive number for repaired cases + * \retval 0 if nothing to be repaired + * \retval negative error number on failure + */ +static int lfsck_namespace_double_scan_dir(const struct lu_env *env, + struct lfsck_component *com, + struct dt_object *child, __u8 flags) +{ + struct lfsck_thread_info *info = lfsck_env_info(env); + const struct lu_fid *cfid = lfsck_dto2fid(child); + struct lu_fid *pfid = &info->lti_fid2; + struct lfsck_namespace *ns = com->lc_file_ram; + struct lfsck_instance *lfsck = com->lc_lfsck; + struct lustre_handle lh = { 0 }; + struct linkea_data ldata = { NULL }; + bool unknown = false; + bool lpf = false; + bool retry = false; + enum lfsck_namespace_inconsistency_type type = LNIT_BAD_LINKEA; + int rc = 0; + ENTRY; + + LASSERT(!dt_object_remote(child)); + + if (flags & LNTF_UNCERTAIN_LMV) { + if (flags & LNTF_RECHECK_NAME_HASH) { + rc = lfsck_namespace_scan_shard(env, com, child); + if (rc < 0) + RETURN(rc); + + ns->ln_striped_shards_scanned++; + } else { + ns->ln_striped_shards_skipped++; + } + } + + flags &= ~(LNTF_RECHECK_NAME_HASH | LNTF_UNCERTAIN_LMV); + if (flags == 0) + RETURN(0); + + if (flags & (LNTF_CHECK_LINKEA | LNTF_CHECK_PARENT) && + !(lfsck->li_bookmark_ram.lb_param & LPF_ALL_TGT)) { + CDEBUG(D_LFSCK, "%s: some MDT(s) maybe NOT take part in the" + "the namespace LFSCK, then the LFSCK cannot guarantee" + "all the name entries have been verified in first-stage" + "scanning. So have to skip orphan related handling for" + "the directory object "DFID" with remote name entry\n", + lfsck_lfsck2name(lfsck), PFID(cfid)); + + RETURN(0); } if (unlikely(!dt_try_as_dir(env, child))) @@ -2630,7 +3130,7 @@ lock: } rc = dt_lookup(env, child, (struct dt_rec *)pfid, - (const struct dt_key *)dotdot, BYPASS_CAPA); + (const struct dt_key *)dotdot); if (rc != 0) { if (rc != -ENOENT && rc != -ENODATA && rc != -EINVAL) { dt_read_unlock(env, child); @@ -2647,6 +3147,8 @@ lock: } else if (lfsck->li_lpf_obj != NULL && lu_fid_eq(pfid, lfsck_dto2fid(lfsck->li_lpf_obj))) { lpf = true; + } else if (unlikely(!fid_is_sane(pfid))) { + fid_zero(pfid); } rc = lfsck_links_read(env, child, &ldata); @@ -2681,6 +3183,20 @@ lock: * directory, then handle it as orphan. */ lfsck_ibits_unlock(&lh, LCK_EX); type = LNIT_MUL_REF; + + /* If the LFSCK is marked as LF_INCOMPLETE, + * then means some MDT has ever tried to + * verify some remote MDT-object that resides + * on this MDT, but this MDT failed to respond + * such request. So means there may be some + * remote name entry on other MDT that + * references this object with another name, + * so we cannot know whether this linkEA is + * valid or not. So keep it there and maybe + * resolved when next LFSCK run. */ + if (ns->ln_flags & LF_INCOMPLETE) + GOTO(out, rc = 0); + snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf), "-"DFID, PFID(pfid)); rc = lfsck_namespace_insert_orphan(env, com, child, @@ -2688,13 +3204,13 @@ lock: } GOTO(out, rc); - } + } /* rc != 0 */ linkea_first_entry(&ldata); /* This is the most common case: the object has unique linkEA entry. */ if (ldata.ld_leh->leh_reccount == 1) { rc = lfsck_namespace_dsd_single(env, com, child, pfid, &ldata, - &lh, &type, &retry); + &lh, &type, &retry, &unknown); if (retry) { LASSERT(!lustre_handle_is_used(&lh)); @@ -2726,7 +3242,7 @@ lock: * but the LFSCK cannot aware that at that time, then it adds * the bad linkEA entry for further processing. */ rc = lfsck_namespace_dsd_multiple(env, com, child, pfid, &ldata, - &lh, &type, lpf); + &lh, &type, lpf, &unknown); GOTO(out, rc); @@ -2785,7 +3301,7 @@ static int lfsck_namespace_double_scan_one(const struct lu_env *env, struct lfsck_instance *lfsck = com->lc_lfsck; struct lfsck_namespace *ns = com->lc_file_ram; struct dt_object *parent = NULL; - struct linkea_data ldata = { 0 }; + struct linkea_data ldata = { NULL }; bool repaired = false; int count = 0; int rc; @@ -2807,13 +3323,42 @@ static int lfsck_namespace_double_scan_one(const struct lu_env *env, rc = lfsck_links_read(env, child, &ldata); dt_read_unlock(env, child); + + if (rc == -EINVAL) { + struct lustre_handle lh = { 0 }; + + rc = lfsck_ibits_lock(env, com->lc_lfsck, child, &lh, + MDS_INODELOCK_UPDATE | + MDS_INODELOCK_XATTR, LCK_EX); + if (rc == 0) { + rc = lfsck_namespace_links_remove(env, com, child); + lfsck_ibits_unlock(&lh, LCK_EX); + } + + GOTO(out, rc = (rc == -ENOENT ? 0 : rc)); + } + if (rc != 0) GOTO(out, rc); linkea_first_entry(&ldata); while (ldata.ld_lee != NULL) { - lfsck_namespace_unpack_linkea_entry(&ldata, cname, pfid, - info->lti_key); + rc = lfsck_namespace_unpack_linkea_entry(&ldata, cname, pfid, + info->lti_key, + sizeof(info->lti_key)); + /* Invalid PFID in the linkEA entry. */ + if (rc != 0) { + rc = lfsck_namespace_shrink_linkea(env, com, child, + &ldata, cname, pfid, true); + if (rc < 0) + GOTO(out, rc); + + if (rc > 0) + repaired = true; + + continue; + } + rc = lfsck_namespace_filter_linkea_entry(&ldata, cname, pfid, false); /* Found repeated linkEA entries */ @@ -2831,19 +3376,6 @@ static int lfsck_namespace_double_scan_one(const struct lu_env *env, /* fall through */ } - /* Invalid PFID in the linkEA entry. */ - if (!fid_is_sane(pfid)) { - rc = lfsck_namespace_shrink_linkea(env, com, child, - &ldata, cname, pfid, true); - if (rc < 0) - GOTO(out, rc); - - if (rc > 0) - repaired = true; - - continue; - } - parent = lfsck_object_find_bottom(env, lfsck, pfid); if (IS_ERR(parent)) GOTO(out, rc = PTR_ERR(parent)); @@ -2859,9 +3391,25 @@ lost_parent: rc = lfsck_namespace_shrink_linkea(env, com, child, &ldata, cname, pfid, true); } else { + /* If the LFSCK is marked as LF_INCOMPLETE, + * then means some MDT has ever tried to + * verify some remote MDT-object that resides + * on this MDT, but this MDT failed to respond + * such request. So means there may be some + * remote name entry on other MDT that + * references this object with another name, + * so we cannot know whether this linkEA is + * valid or not. So keep it there and maybe + * resolved when next LFSCK run. */ + if (ns->ln_flags & LF_INCOMPLETE) { + lfsck_object_put(env, parent); + + GOTO(out, rc = 0); + } + /* Create the lost parent as an orphan. */ - rc = lfsck_namespace_create_orphan(env, com, - parent); + rc = lfsck_namespace_create_orphan_dir(env, com, + parent, NULL); if (rc < 0) { lfsck_object_put(env, parent); @@ -2898,7 +3446,7 @@ lost_parent: repaired = true; continue; - } + } /* !dt_object_exists(parent) */ /* The linkEA entry with bad parent will be removed. */ if (unlikely(!dt_try_as_dir(env, parent))) { @@ -2915,8 +3463,7 @@ lost_parent: } rc = dt_lookup(env, parent, (struct dt_rec *)cfid, - (const struct dt_key *)cname->ln_name, - BYPASS_CAPA); + (const struct dt_key *)cname->ln_name); if (rc != 0 && rc != -ENOENT) { lfsck_object_put(env, parent); @@ -2959,7 +3506,9 @@ lost_parent: continue; } - rc = dt_attr_get(env, child, la, BYPASS_CAPA); + /* The following handles -ENOENT case */ + + rc = dt_attr_get(env, child, la); if (rc != 0) GOTO(out, rc); @@ -2979,6 +3528,43 @@ lost_parent: continue; } + /* If the LFSCK is marked as LF_INCOMPLETE, then means some + * MDT has ever tried to verify some remote MDT-object that + * resides on this MDT, but this MDT failed to respond such + * request. So means there may be some remote name entry on + * other MDT that references this object with another name, + * so we cannot know whether this linkEA is valid or not. + * So keep it there and maybe resolved when next LFSCK run. */ + if (ns->ln_flags & LF_INCOMPLETE) { + lfsck_object_put(env, parent); + + GOTO(out, rc = 0); + } + + rc = lfsck_namespace_check_name(env, parent, child, cname); + if (rc == -ENOENT) + goto lost_parent; + + if (rc < 0) { + lfsck_object_put(env, parent); + + GOTO(out, rc); + } + + /* It is an invalid name entry, drop it. */ + if (unlikely(rc > 0)) { + lfsck_object_put(env, parent); + rc = lfsck_namespace_shrink_linkea(env, com, child, + &ldata, cname, pfid, true); + if (rc < 0) + GOTO(out, rc); + + if (rc > 0) + repaired = true; + + continue; + } + /* Add the missing name entry back to the namespace. */ rc = lfsck_namespace_insert_normal(env, com, parent, child, cname->ln_name); @@ -3017,33 +3603,55 @@ out: if (rc < 0 && rc != -ENODATA) return rc; - if (rc == 0) { - LASSERT(ldata.ld_leh != NULL); - + if (rc == 0 && ldata.ld_leh != NULL) count = ldata.ld_leh->leh_reccount; - } if (count == 0) { - /* If the child becomes orphan, then insert it into - * the global .lustre/lost+found/MDTxxxx directory. */ - rc = lfsck_namespace_insert_orphan(env, com, child, "", "O", - &count); - if (rc < 0) - return rc; + /* If the LFSCK is marked as LF_INCOMPLETE, then means some + * MDT has ever tried to verify some remote MDT-object that + * resides on this MDT, but this MDT failed to respond such + * request. So means there may be some remote name entry on + * other MDT that references this object with another name, + * so we cannot know whether this linkEA is valid or not. + * So keep it there and maybe resolved when next LFSCK run. */ + if (!(ns->ln_flags & LF_INCOMPLETE)) { + /* If the child becomes orphan, then insert it into + * the global .lustre/lost+found/MDTxxxx directory. */ + rc = lfsck_namespace_insert_orphan(env, com, child, + "", "O", &count); + if (rc < 0) + return rc; - if (rc > 0) { - ns->ln_mul_ref_repaired++; - repaired = true; + if (rc > 0) { + ns->ln_mul_ref_repaired++; + repaired = true; + } } - } - - rc = dt_attr_get(env, child, la, BYPASS_CAPA); - if (rc != 0) - return rc; + } else { + rc = dt_attr_get(env, child, la); + if (rc != 0) + return rc; - if (la->la_nlink != count) { - /* XXX: there will be other patch(es) for MDT-object - * hard links verification. */ + if (la->la_nlink != 0 && la->la_nlink != count) { + if (unlikely(!S_ISREG(lfsck_object_type(child)) && + !S_ISLNK(lfsck_object_type(child)))) { + CDEBUG(D_LFSCK, "%s: namespace LFSCK finds " + "the object "DFID"'s nlink count %d " + "does not match linkEA count %d, " + "type %o, skip it.\n", + lfsck_lfsck2name(lfsck), + PFID(lfsck_dto2fid(child)), + la->la_nlink, count, + lfsck_object_type(child)); + } else { + rc = lfsck_namespace_repair_nlink(env, com, + child, la); + if (rc > 0) { + ns->ln_objs_nlink_repaired++; + rc = 0; + } + } + } } if (repaired) { @@ -3064,48 +3672,126 @@ static void lfsck_namespace_dump_statistics(struct seq_file *m, __u32 time_phase1, __u32 time_phase2) { - seq_printf(m, "checked_phase1: "LPU64"\n" - "checked_phase2: "LPU64"\n" - "updated_phase1: "LPU64"\n" - "updated_phase2: "LPU64"\n" - "failed_phase1: "LPU64"\n" - "failed_phase2: "LPU64"\n" - "directories: "LPU64"\n" - "dirent_repaired: "LPU64"\n" - "linkea_repaired: "LPU64"\n" - "nlinks_repaired: "LPU64"\n" - "multiple_linked_checked: "LPU64"\n" - "multiple_linked_repaired: "LPU64"\n" - "unknown_inconsistency: "LPU64"\n" - "unmatched_pairs_repaired: "LPU64"\n" - "dangling_repaired: "LPU64"\n" - "multiple_referenced_repaired: "LPU64"\n" - "bad_file_type_repaired: "LPU64"\n" - "lost_dirent_repaired: "LPU64"\n" - "success_count: %u\n" - "run_time_phase1: %u seconds\n" - "run_time_phase2: %u seconds\n", - checked_phase1, - checked_phase2, - ns->ln_items_repaired, - ns->ln_objs_repaired_phase2, - ns->ln_items_failed, - ns->ln_objs_failed_phase2, - ns->ln_dirs_checked, - ns->ln_dirent_repaired, - ns->ln_linkea_repaired, - ns->ln_objs_nlink_repaired, - ns->ln_mul_linked_checked, - ns->ln_mul_linked_repaired, - ns->ln_unknown_inconsistency, - ns->ln_unmatched_pairs_repaired, - ns->ln_dangling_repaired, - ns->ln_mul_ref_repaired, - ns->ln_bad_type_repaired, - ns->ln_lost_dirent_repaired, - ns->ln_success_count, - time_phase1, - time_phase2); + seq_printf(m, "checked_phase1: %llu\n" + "checked_phase2: %llu\n" + "updated_phase1: %llu\n" + "updated_phase2: %llu\n" + "failed_phase1: %llu\n" + "failed_phase2: %llu\n" + "directories: %llu\n" + "dirent_repaired: %llu\n" + "linkea_repaired: %llu\n" + "nlinks_repaired: %llu\n" + "multiple_linked_checked: %llu\n" + "multiple_linked_repaired: %llu\n" + "unknown_inconsistency: %llu\n" + "unmatched_pairs_repaired: %llu\n" + "dangling_repaired: %llu\n" + "multiple_referenced_repaired: %llu\n" + "bad_file_type_repaired: %llu\n" + "lost_dirent_repaired: %llu\n" + "local_lost_found_scanned: %llu\n" + "local_lost_found_moved: %llu\n" + "local_lost_found_skipped: %llu\n" + "local_lost_found_failed: %llu\n" + "striped_dirs_scanned: %llu\n" + "striped_dirs_repaired: %llu\n" + "striped_dirs_failed: %llu\n" + "striped_dirs_disabled: %llu\n" + "striped_dirs_skipped: %llu\n" + "striped_shards_scanned: %llu\n" + "striped_shards_repaired: %llu\n" + "striped_shards_failed: %llu\n" + "striped_shards_skipped: %llu\n" + "name_hash_repaired: %llu\n" + "success_count: %u\n" + "run_time_phase1: %u seconds\n" + "run_time_phase2: %u seconds\n", + checked_phase1, + checked_phase2, + ns->ln_items_repaired, + ns->ln_objs_repaired_phase2, + ns->ln_items_failed, + ns->ln_objs_failed_phase2, + ns->ln_dirs_checked, + ns->ln_dirent_repaired, + ns->ln_linkea_repaired, + ns->ln_objs_nlink_repaired, + ns->ln_mul_linked_checked, + ns->ln_mul_linked_repaired, + ns->ln_unknown_inconsistency, + ns->ln_unmatched_pairs_repaired, + ns->ln_dangling_repaired, + ns->ln_mul_ref_repaired, + ns->ln_bad_type_repaired, + ns->ln_lost_dirent_repaired, + ns->ln_local_lpf_scanned, + ns->ln_local_lpf_moved, + ns->ln_local_lpf_skipped, + ns->ln_local_lpf_failed, + ns->ln_striped_dirs_scanned, + ns->ln_striped_dirs_repaired, + ns->ln_striped_dirs_failed, + ns->ln_striped_dirs_disabled, + ns->ln_striped_dirs_skipped, + ns->ln_striped_shards_scanned, + ns->ln_striped_shards_repaired, + ns->ln_striped_shards_failed, + ns->ln_striped_shards_skipped, + ns->ln_name_hash_repaired, + ns->ln_success_count, + time_phase1, + time_phase2); +} + +static void lfsck_namespace_release_lmv(const struct lu_env *env, + struct lfsck_component *com) +{ + struct lfsck_instance *lfsck = com->lc_lfsck; + struct lfsck_namespace *ns = com->lc_file_ram; + + while (!list_empty(&lfsck->li_list_lmv)) { + struct lfsck_lmv_unit *llu; + struct lfsck_lmv *llmv; + + llu = list_entry(lfsck->li_list_lmv.next, + struct lfsck_lmv_unit, llu_link); + llmv = &llu->llu_lmv; + + LASSERTF(atomic_read(&llmv->ll_ref) == 1, + "still in using: %u\n", + atomic_read(&llmv->ll_ref)); + + ns->ln_striped_dirs_skipped++; + lfsck_lmv_put(env, llmv); + } +} + +static int lfsck_namespace_check_for_double_scan(const struct lu_env *env, + struct lfsck_component *com, + struct dt_object *obj) +{ + struct lu_attr *la = &lfsck_env_info(env)->lti_la; + int rc; + + rc = dt_attr_get(env, obj, la); + if (rc != 0) + return rc; + + /* zero-linkEA object may be orphan, but it also maybe because + * of upgrading. Currently, we cannot record it for double scan. + * Because it may cause the LFSCK trace file to be too large. */ + + /* "la_ctime" == 1 means that it has ever been removed from + * backend /lost+found directory but not been added back to + * the normal namespace yet. */ + + if ((S_ISREG(lfsck_object_type(obj)) && la->la_nlink > 1) || + unlikely(la->la_ctime == 1)) + rc = lfsck_namespace_trace_update(env, com, lfsck_dto2fid(obj), + LNTF_CHECK_LINKEA, true); + + return rc; } /* namespace APIs */ @@ -3113,11 +3799,12 @@ static void lfsck_namespace_dump_statistics(struct seq_file *m, static int lfsck_namespace_reset(const struct lu_env *env, struct lfsck_component *com, bool init) { - struct lfsck_instance *lfsck = com->lc_lfsck; - struct lfsck_namespace *ns = com->lc_file_ram; - struct dt_object *root; - struct dt_object *dto; - int rc; + struct lfsck_instance *lfsck = com->lc_lfsck; + struct lfsck_namespace *ns = com->lc_file_ram; + struct lfsck_assistant_data *lad = com->lc_data; + struct dt_object *root; + struct dt_object *dto; + int rc; ENTRY; root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid); @@ -3141,26 +3828,22 @@ static int lfsck_namespace_reset(const struct lu_env *env, ns->ln_magic = LFSCK_NAMESPACE_MAGIC; ns->ln_status = LS_INIT; - rc = local_object_unlink(env, lfsck->li_bottom, root, - lfsck_namespace_name); - if (rc != 0) - GOTO(out, rc); - lfsck_object_put(env, com->lc_obj); com->lc_obj = NULL; - dto = local_index_find_or_create(env, lfsck->li_los, root, - lfsck_namespace_name, - S_IFREG | S_IRUGO | S_IWUSR, - &dt_lfsck_features); + dto = lfsck_namespace_load_one_trace_file(env, com, root, + LFSCK_NAMESPACE, true); if (IS_ERR(dto)) GOTO(out, rc = PTR_ERR(dto)); com->lc_obj = dto; - rc = dto->do_ops->do_index_try(env, dto, &dt_lfsck_features); + rc = lfsck_namespace_load_sub_trace_files(env, com, true); if (rc != 0) GOTO(out, rc); - rc = lfsck_namespace_store(env, com, true); + lad->lad_incomplete = 0; + CFS_RESET_BITMAP(lad->lad_bitmap); + + rc = lfsck_namespace_store(env, com); GOTO(out, rc); @@ -3168,7 +3851,7 @@ out: up_write(&com->lc_sem); put: - lu_object_put(env, &root->do_lu); + lfsck_object_put(env, root); log: CDEBUG(D_LFSCK, "%s: namespace LFSCK reset: rc = %d\n", lfsck_lfsck2name(lfsck), rc); @@ -3188,10 +3871,102 @@ lfsck_namespace_fail(const struct lu_env *env, struct lfsck_component *com, up_write(&com->lc_sem); } -static int lfsck_namespace_checkpoint(const struct lu_env *env, - struct lfsck_component *com, bool init) +static void lfsck_namespace_close_dir(const struct lu_env *env, + struct lfsck_component *com) { - struct lfsck_instance *lfsck = com->lc_lfsck; + struct lfsck_namespace *ns = com->lc_file_ram; + struct lfsck_assistant_data *lad = com->lc_data; + struct lfsck_assistant_object *lso = NULL; + struct lfsck_instance *lfsck = com->lc_lfsck; + struct lfsck_lmv *llmv = lfsck->li_lmv; + struct lfsck_namespace_req *lnr; + __u32 size = + sizeof(*lnr) + LFSCK_TMPBUF_LEN; + bool wakeup = false; + ENTRY; + + if (llmv == NULL) + RETURN_EXIT; + + OBD_ALLOC(lnr, size); + if (lnr == NULL) { + ns->ln_striped_dirs_skipped++; + + RETURN_EXIT; + } + + lso = lfsck_assistant_object_init(env, lfsck_dto2fid(lfsck->li_obj_dir), + NULL, lfsck->li_pos_current.lp_oit_cookie, true); + if (IS_ERR(lso)) { + OBD_FREE(lnr, size); + ns->ln_striped_dirs_skipped++; + + RETURN_EXIT; + } + + /* Generate a dummy request to indicate that all shards' name entry + * in this striped directory has been scanned for the first time. */ + INIT_LIST_HEAD(&lnr->lnr_lar.lar_list); + lnr->lnr_lar.lar_parent = lso; + lnr->lnr_lmv = lfsck_lmv_get(llmv); + lnr->lnr_fid = *lfsck_dto2fid(lfsck->li_obj_dir); + lnr->lnr_dir_cookie = MDS_DIR_END_OFF; + lnr->lnr_size = size; + + spin_lock(&lad->lad_lock); + if (lad->lad_assistant_status < 0) { + spin_unlock(&lad->lad_lock); + lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar); + ns->ln_striped_dirs_skipped++; + + RETURN_EXIT; + } + + list_add_tail(&lnr->lnr_lar.lar_list, &lad->lad_req_list); + if (lad->lad_prefetched == 0) + wakeup = true; + + lad->lad_prefetched++; + spin_unlock(&lad->lad_lock); + if (wakeup) + wake_up_all(&lad->lad_thread.t_ctl_waitq); + + EXIT; +} + +static int lfsck_namespace_open_dir(const struct lu_env *env, + struct lfsck_component *com) +{ + struct lfsck_instance *lfsck = com->lc_lfsck; + struct lfsck_namespace *ns = com->lc_file_ram; + struct lfsck_lmv *llmv = lfsck->li_lmv; + int rc = 0; + ENTRY; + + if (llmv == NULL) + RETURN(0); + + if (llmv->ll_lmv_master) { + struct lmv_mds_md_v1 *lmv = &llmv->ll_lmv; + + if (lmv->lmv_master_mdt_index != lfsck_dev_idx(lfsck)) { + lmv->lmv_master_mdt_index = + lfsck_dev_idx(lfsck); + ns->ln_flags |= LF_INCONSISTENT; + llmv->ll_lmv_updated = 1; + } + } else { + rc = lfsck_namespace_verify_stripe_slave(env, com, + lfsck->li_obj_dir, llmv); + } + + RETURN(rc > 0 ? 0 : rc); +} + +static int lfsck_namespace_checkpoint(const struct lu_env *env, + struct lfsck_component *com, bool init) +{ + struct lfsck_instance *lfsck = com->lc_lfsck; struct lfsck_namespace *ns = com->lc_file_ram; int rc; @@ -3213,15 +3988,15 @@ static int lfsck_namespace_checkpoint(const struct lu_env *env, com->lc_new_checked = 0; } - rc = lfsck_namespace_store(env, com, false); + rc = lfsck_namespace_store(env, com); up_write(&com->lc_sem); log: - CDEBUG(D_LFSCK, "%s: namespace LFSCK checkpoint at the pos ["LPU64 - ", "DFID", "LPX64"]: rc = %d\n", lfsck_lfsck2name(lfsck), - lfsck->li_pos_current.lp_oit_cookie, + CDEBUG(D_LFSCK, "%s: namespace LFSCK checkpoint at the pos [%llu" + ", "DFID", %#llx], status = %d: rc = %d\n", + lfsck_lfsck2name(lfsck), lfsck->li_pos_current.lp_oit_cookie, PFID(&lfsck->li_pos_current.lp_dir_parent), - lfsck->li_pos_current.lp_dir_cookie, rc); + lfsck->li_pos_current.lp_dir_cookie, ns->ln_status, rc); return rc > 0 ? 0 : rc; } @@ -3235,7 +4010,8 @@ static int lfsck_namespace_prep(const struct lu_env *env, struct lfsck_position *pos = &com->lc_pos_start; int rc; - if (ns->ln_status == LS_COMPLETED) { + rc = lfsck_namespace_load_bitmap(env, com); + if (rc != 0 || ns->ln_status == LS_COMPLETED) { rc = lfsck_namespace_reset(env, com, false); if (rc == 0) rc = lfsck_set_param(env, lfsck, lsp->lsp_start, true); @@ -3283,6 +4059,16 @@ static int lfsck_namespace_prep(const struct lu_env *env, ns->ln_mul_ref_repaired = 0; ns->ln_bad_type_repaired = 0; ns->ln_lost_dirent_repaired = 0; + ns->ln_striped_dirs_scanned = 0; + ns->ln_striped_dirs_repaired = 0; + ns->ln_striped_dirs_failed = 0; + ns->ln_striped_dirs_disabled = 0; + ns->ln_striped_dirs_skipped = 0; + ns->ln_striped_shards_scanned = 0; + ns->ln_striped_shards_repaired = 0; + ns->ln_striped_shards_failed = 0; + ns->ln_striped_shards_skipped = 0; + ns->ln_name_hash_repaired = 0; fid_zero(&ns->ln_fid_latest_scanned_phase2); if (list_empty(&com->lc_link_dir)) list_add_tail(&com->lc_link_dir, @@ -3308,8 +4094,8 @@ static int lfsck_namespace_prep(const struct lu_env *env, rc = lfsck_start_assistant(env, com, lsp); - CDEBUG(D_LFSCK, "%s: namespace LFSCK prep done, start pos ["LPU64", " - DFID", "LPX64"]: rc = %d\n", + CDEBUG(D_LFSCK, "%s: namespace LFSCK prep done, start pos [%llu, " + DFID", %#llx]: rc = %d\n", lfsck_lfsck2name(lfsck), pos->lp_oit_cookie, PFID(&pos->lp_dir_parent), pos->lp_dir_cookie, rc); @@ -3324,15 +4110,12 @@ static int lfsck_namespace_exec_oit(const struct lu_env *env, struct lfsck_namespace *ns = com->lc_file_ram; struct lfsck_instance *lfsck = com->lc_lfsck; const struct lu_fid *fid = lfsck_dto2fid(obj); - struct lu_attr *la = &info->lti_la; struct lu_fid *pfid = &info->lti_fid2; struct lu_name *cname = &info->lti_name; struct lu_seq_range *range = &info->lti_range; - struct dt_device *dev = lfsck->li_bottom; - struct seq_server_site *ss = - lu_site2seq(dev->dd_lu_dev.ld_site); - struct linkea_data ldata = { 0 }; - __u32 idx = lfsck_dev_idx(dev); + struct seq_server_site *ss = lfsck_dev_site(lfsck); + struct linkea_data ldata = { NULL }; + __u32 idx = lfsck_dev_idx(lfsck); int rc; ENTRY; @@ -3360,20 +4143,8 @@ static int lfsck_namespace_exec_oit(const struct lu_env *env, GOTO(out, rc = (rc == -ENOENT ? 0 : rc)); } - /* zero-linkEA object may be orphan, but it also maybe because - * of upgrading. Currently, we cannot record it for double scan. - * Because it may cause the LFSCK tracing file to be too large. */ if (rc == -ENODATA) { - if (S_ISDIR(lfsck_object_type(obj))) - GOTO(out, rc = 0); - - rc = dt_attr_get(env, obj, la, BYPASS_CAPA); - if (rc != 0) - GOTO(out, rc); - - if (la->la_nlink > 1) - rc = lfsck_namespace_trace_update(env, com, fid, - LNTF_CHECK_LINKEA, true); + rc = lfsck_namespace_check_for_double_scan(env, com, obj); GOTO(out, rc); } @@ -3399,28 +4170,18 @@ static int lfsck_namespace_exec_oit(const struct lu_env *env, rc = fld_local_lookup(env, ss->ss_server_fld, fid_seq(pfid), range); if ((rc == -ENOENT) || - (rc == 0 && range->lsr_index != idx)) { + (rc == 0 && range->lsr_index != idx)) rc = lfsck_namespace_trace_update(env, com, fid, LNTF_CHECK_LINKEA, true); - } else { - if (S_ISDIR(lfsck_object_type(obj))) - GOTO(out, rc = 0); - - rc = dt_attr_get(env, obj, la, BYPASS_CAPA); - if (rc != 0) - GOTO(out, rc); - - if (la->la_nlink > 1) - rc = lfsck_namespace_trace_update(env, com, - fid, LNTF_CHECK_LINKEA, true); - } + else + rc = lfsck_namespace_check_for_double_scan(env, com, + obj); } GOTO(out, rc); out: down_write(&com->lc_sem); - com->lc_new_checked++; if (S_ISDIR(lfsck_object_type(obj))) ns->ln_dirs_checked++; if (rc != 0) @@ -3432,13 +4193,32 @@ out: static int lfsck_namespace_exec_dir(const struct lu_env *env, struct lfsck_component *com, + struct lfsck_assistant_object *lso, struct lu_dirent *ent, __u16 type) { - struct lfsck_assistant_data *lad = com->lc_data; + struct lfsck_assistant_data *lad = com->lc_data; + struct lfsck_instance *lfsck = com->lc_lfsck; struct lfsck_namespace_req *lnr; - bool wakeup = false; - - lnr = lfsck_namespace_assistant_req_init(com->lc_lfsck, ent, type); + struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram; + struct ptlrpc_thread *mthread = &lfsck->li_thread; + struct ptlrpc_thread *athread = &lad->lad_thread; + struct l_wait_info lwi = { 0 }; + bool wakeup = false; + + l_wait_event(mthread->t_ctl_waitq, + lad->lad_prefetched < bk->lb_async_windows || + !thread_is_running(mthread) || + thread_is_stopped(athread), + &lwi); + + if (unlikely(!thread_is_running(mthread)) || + thread_is_stopped(athread)) + return 0; + + if (unlikely(lfsck_is_dead_obj(lfsck->li_obj_dir))) + return 0; + + lnr = lfsck_namespace_assistant_req_init(com->lc_lfsck, lso, ent, type); if (IS_ERR(lnr)) { struct lfsck_namespace *ns = com->lc_file_ram; @@ -3481,6 +4261,8 @@ static int lfsck_namespace_post(const struct lu_env *env, lfsck_post_generic(env, com, &result); down_write(&com->lc_sem); + lfsck_namespace_release_lmv(env, com); + spin_lock(&lfsck->li_lock); if (!init) ns->ln_pos_last_checkpoint = lfsck->li_pos_checkpoint; @@ -3491,8 +4273,9 @@ static int lfsck_namespace_post(const struct lu_env *env, list_del_init(&com->lc_link_dir); list_move_tail(&com->lc_link, &lfsck->li_list_double_scan); } else if (result == 0) { - ns->ln_status = lfsck->li_status; - if (ns->ln_status == 0) + if (lfsck->li_status != 0) + ns->ln_status = lfsck->li_status; + else ns->ln_status = LS_STOPPED; if (ns->ln_status != LS_PAUSED) { list_del_init(&com->lc_link_dir); @@ -3513,7 +4296,7 @@ static int lfsck_namespace_post(const struct lu_env *env, com->lc_new_checked = 0; } - rc = lfsck_namespace_store(env, com, false); + rc = lfsck_namespace_store(env, com); up_write(&com->lc_sem); CDEBUG(D_LFSCK, "%s: namespace LFSCK post done: rc = %d\n", @@ -3522,14 +4305,13 @@ static int lfsck_namespace_post(const struct lu_env *env, RETURN(rc); } -static int +static void lfsck_namespace_dump(const struct lu_env *env, struct lfsck_component *com, struct seq_file *m) { struct lfsck_instance *lfsck = com->lc_lfsck; struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram; struct lfsck_namespace *ns = com->lc_file_ram; - int rc; down_read(&com->lc_sem); seq_printf(m, "name: lfsck_namespace\n" @@ -3538,45 +4320,25 @@ lfsck_namespace_dump(const struct lu_env *env, struct lfsck_component *com, "status: %s\n", ns->ln_magic, bk->lb_version, - lfsck_status2names(ns->ln_status)); + lfsck_status2name(ns->ln_status)); - rc = lfsck_bits_dump(m, ns->ln_flags, lfsck_flags_names, "flags"); - if (rc < 0) - goto out; + lfsck_bits_dump(m, ns->ln_flags, lfsck_flags_names, "flags"); - rc = lfsck_bits_dump(m, bk->lb_param, lfsck_param_names, "param"); - if (rc < 0) - goto out; + lfsck_bits_dump(m, bk->lb_param, lfsck_param_names, "param"); - rc = lfsck_time_dump(m, ns->ln_time_last_complete, - "time_since_last_completed"); - if (rc < 0) - goto out; + lfsck_time_dump(m, ns->ln_time_last_complete, "last_completed"); - rc = lfsck_time_dump(m, ns->ln_time_latest_start, - "time_since_latest_start"); - if (rc < 0) - goto out; + lfsck_time_dump(m, ns->ln_time_latest_start, "latest_start"); - rc = lfsck_time_dump(m, ns->ln_time_last_checkpoint, - "time_since_last_checkpoint"); - if (rc < 0) - goto out; + lfsck_time_dump(m, ns->ln_time_last_checkpoint, "last_checkpoint"); - rc = lfsck_pos_dump(m, &ns->ln_pos_latest_start, - "latest_start_position"); - if (rc < 0) - goto out; + lfsck_pos_dump(m, &ns->ln_pos_latest_start, "latest_start_position"); - rc = lfsck_pos_dump(m, &ns->ln_pos_last_checkpoint, - "last_checkpoint_position"); - if (rc < 0) - goto out; + lfsck_pos_dump(m, &ns->ln_pos_last_checkpoint, + "last_checkpoint_position"); - rc = lfsck_pos_dump(m, &ns->ln_pos_first_inconsistent, - "first_failure_position"); - if (rc < 0) - goto out; + lfsck_pos_dump(m, &ns->ln_pos_first_inconsistent, + "first_failure_position"); if (ns->ln_status == LS_SCANNING_PHASE1) { struct lfsck_position pos; @@ -3585,24 +4347,26 @@ lfsck_namespace_dump(const struct lu_env *env, struct lfsck_component *com, lfsck->li_time_last_checkpoint; __u64 checked = ns->ln_items_checked + com->lc_new_checked; __u64 speed = checked; - __u64 new_checked = com->lc_new_checked * HZ; + __u64 new_checked = com->lc_new_checked * + msecs_to_jiffies(MSEC_PER_SEC); __u32 rtime = ns->ln_run_time_phase1 + cfs_duration_sec(duration + HALF_SEC); if (duration != 0) do_div(new_checked, duration); + if (rtime != 0) do_div(speed, rtime); - lfsck_namespace_dump_statistics(m, ns, checked, - ns->ln_objs_checked_phase2, - rtime, ns->ln_run_time_phase2); - seq_printf(m, "average_speed_phase1: "LPU64" items/sec\n" - "average_speed_phase2: N/A\n" - "real_time_speed_phase1: "LPU64" items/sec\n" - "real_time_speed_phase2: N/A\n", - speed, - new_checked); + lfsck_namespace_dump_statistics(m, ns, checked, 0, rtime, 0); + seq_printf(m, "average_speed_phase1: %llu items/sec\n" + "average_speed_phase2: N/A\n" + "average_speed_total: %llu items/sec\n" + "real_time_speed_phase1: %llu items/sec\n" + "real_time_speed_phase2: N/A\n", + speed, + speed, + new_checked); LASSERT(lfsck->li_di_oit != NULL); @@ -3634,66 +4398,109 @@ lfsck_namespace_dump(const struct lu_env *env, struct lfsck_component *com, lfsck_pos_dump(m, &pos, "current_position"); } else if (ns->ln_status == LS_SCANNING_PHASE2) { cfs_duration_t duration = cfs_time_current() - - lfsck->li_time_last_checkpoint; + com->lc_time_last_checkpoint; __u64 checked = ns->ln_objs_checked_phase2 + com->lc_new_checked; __u64 speed1 = ns->ln_items_checked; __u64 speed2 = checked; - __u64 new_checked = com->lc_new_checked * HZ; + __u64 speed0 = speed1 + speed2; + __u64 new_checked = com->lc_new_checked * + msecs_to_jiffies(MSEC_PER_SEC); __u32 rtime = ns->ln_run_time_phase2 + cfs_duration_sec(duration + HALF_SEC); + __u32 time0 = ns->ln_run_time_phase1 + rtime; if (duration != 0) do_div(new_checked, duration); + if (ns->ln_run_time_phase1 != 0) do_div(speed1, ns->ln_run_time_phase1); + else if (ns->ln_items_checked != 0) + time0++; + if (rtime != 0) do_div(speed2, rtime); + else if (checked != 0) + time0++; + + if (time0 != 0) + do_div(speed0, time0); + lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked, checked, ns->ln_run_time_phase1, rtime); - - seq_printf(m, "average_speed_phase1: "LPU64" items/sec\n" - "average_speed_phase2: "LPU64" objs/sec\n" - "real_time_speed_phase1: N/A\n" - "real_time_speed_phase2: "LPU64" objs/sec\n" - "current_position: "DFID"\n", - speed1, - speed2, - new_checked, - PFID(&ns->ln_fid_latest_scanned_phase2)); + seq_printf(m, "average_speed_phase1: %llu items/sec\n" + "average_speed_phase2: %llu objs/sec\n" + "average_speed_total: %llu items/sec\n" + "real_time_speed_phase1: N/A\n" + "real_time_speed_phase2: %llu objs/sec\n" + "current_position: "DFID"\n", + speed1, + speed2, + speed0, + new_checked, + PFID(&ns->ln_fid_latest_scanned_phase2)); } else { __u64 speed1 = ns->ln_items_checked; __u64 speed2 = ns->ln_objs_checked_phase2; + __u64 speed0 = speed1 + speed2; + __u32 time0 = ns->ln_run_time_phase1 + ns->ln_run_time_phase2; if (ns->ln_run_time_phase1 != 0) do_div(speed1, ns->ln_run_time_phase1); + else if (ns->ln_items_checked != 0) + time0++; + if (ns->ln_run_time_phase2 != 0) do_div(speed2, ns->ln_run_time_phase2); + else if (ns->ln_objs_checked_phase2 != 0) + time0++; + + if (time0 != 0) + do_div(speed0, time0); + lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked, ns->ln_objs_checked_phase2, ns->ln_run_time_phase1, ns->ln_run_time_phase2); - - seq_printf(m, "average_speed_phase1: "LPU64" items/sec\n" - "average_speed_phase2: "LPU64" objs/sec\n" - "real_time_speed_phase1: N/A\n" - "real_time_speed_phase2: N/A\n" - "current_position: N/A\n", - speed1, - speed2); + seq_printf(m, "average_speed_phase1: %llu items/sec\n" + "average_speed_phase2: %llu objs/sec\n" + "average_speed_total: %llu items/sec\n" + "real_time_speed_phase1: N/A\n" + "real_time_speed_phase2: N/A\n" + "current_position: N/A\n", + speed1, + speed2, + speed0); } -out: + up_read(&com->lc_sem); - return 0; } static int lfsck_namespace_double_scan(const struct lu_env *env, struct lfsck_component *com) { - struct lfsck_namespace *ns = com->lc_file_ram; + struct lfsck_namespace *ns = com->lc_file_ram; + struct lfsck_assistant_data *lad = com->lc_data; + struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs; + struct lfsck_tgt_desc *ltd; + struct lfsck_tgt_desc *next; + int rc; - return lfsck_double_scan_generic(env, com, ns->ln_status); + rc = lfsck_double_scan_generic(env, com, ns->ln_status); + if (thread_is_stopped(&lad->lad_thread)) { + LASSERT(list_empty(&lad->lad_req_list)); + LASSERT(list_empty(&lad->lad_mdt_phase1_list)); + + spin_lock(<ds->ltd_lock); + list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list, + ltd_namespace_phase_list) { + list_del_init(<d->ltd_namespace_phase_list); + } + spin_unlock(<ds->ltd_lock); + } + + return rc; } static void lfsck_namespace_data_release(const struct lu_env *env, @@ -3710,6 +4517,7 @@ static void lfsck_namespace_data_release(const struct lu_env *env, LASSERT(list_empty(&lad->lad_req_list)); com->lc_data = NULL; + lfsck_namespace_release_lmv(env, com); spin_lock(<ds->ltd_lock); list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list, @@ -3726,55 +4534,159 @@ static void lfsck_namespace_data_release(const struct lu_env *env, } spin_unlock(<ds->ltd_lock); - CFS_FREE_BITMAP(lad->lad_bitmap); + if (likely(lad->lad_bitmap != NULL)) + CFS_FREE_BITMAP(lad->lad_bitmap); OBD_FREE_PTR(lad); } +static void lfsck_namespace_quit(const struct lu_env *env, + struct lfsck_component *com) +{ + struct lfsck_assistant_data *lad = com->lc_data; + struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs; + struct lfsck_tgt_desc *ltd; + struct lfsck_tgt_desc *next; + + LASSERT(lad != NULL); + + lfsck_quit_generic(env, com); + + LASSERT(thread_is_init(&lad->lad_thread) || + thread_is_stopped(&lad->lad_thread)); + LASSERT(list_empty(&lad->lad_req_list)); + + lfsck_namespace_release_lmv(env, com); + + spin_lock(<ds->ltd_lock); + list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list, + ltd_namespace_phase_list) { + list_del_init(<d->ltd_namespace_phase_list); + } + list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list, + ltd_namespace_phase_list) { + list_del_init(<d->ltd_namespace_phase_list); + } + spin_unlock(<ds->ltd_lock); +} + static int lfsck_namespace_in_notify(const struct lu_env *env, struct lfsck_component *com, - struct lfsck_request *lr) + struct lfsck_request *lr, + struct thandle *th) { struct lfsck_instance *lfsck = com->lc_lfsck; struct lfsck_namespace *ns = com->lc_file_ram; struct lfsck_assistant_data *lad = com->lc_data; struct lfsck_tgt_descs *ltds = &lfsck->li_mdt_descs; struct lfsck_tgt_desc *ltd; - int rc; + int rc = 0; bool fail = false; ENTRY; switch (lr->lr_event) { - case LE_CREATE_ORPHAN: { - struct dt_object *orphan = NULL; + case LE_SKIP_NLINK_DECLARE: { + struct dt_object *obj; + struct lu_fid *key = &lfsck_env_info(env)->lti_fid3; + int idx; + __u8 flags = 0; + + LASSERT(th != NULL); + + idx = lfsck_sub_trace_file_fid2idx(&lr->lr_fid); + obj = com->lc_sub_trace_objs[idx].lsto_obj; + fid_cpu_to_be(key, &lr->lr_fid); + mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex); + rc = dt_declare_delete(env, obj, + (const struct dt_key *)key, th); + if (rc == 0) + rc = dt_declare_insert(env, obj, + (const struct dt_rec *)&flags, + (const struct dt_key *)key, th); + mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex); + + RETURN(rc); + } + case LE_SKIP_NLINK: { + struct dt_object *obj; + struct lu_fid *key = &lfsck_env_info(env)->lti_fid3; + int idx; + __u8 flags = 0; + bool exist = false; + ENTRY; + + LASSERT(th != NULL); + + idx = lfsck_sub_trace_file_fid2idx(&lr->lr_fid); + obj = com->lc_sub_trace_objs[idx].lsto_obj; + fid_cpu_to_be(key, &lr->lr_fid); + mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex); + rc = dt_lookup(env, obj, (struct dt_rec *)&flags, + (const struct dt_key *)key); + if (rc == 0) { + if (flags & LNTF_SKIP_NLINK) { + mutex_unlock( + &com->lc_sub_trace_objs[idx].lsto_mutex); + + RETURN(0); + } + + exist = true; + } else if (rc != -ENOENT) { + GOTO(log, rc); + } + + flags |= LNTF_SKIP_NLINK; + if (exist) { + rc = dt_delete(env, obj, (const struct dt_key *)key, + th); + if (rc != 0) + GOTO(log, rc); + } + + rc = dt_insert(env, obj, (const struct dt_rec *)&flags, + (const struct dt_key *)key, th, 1); + + GOTO(log, rc); + +log: + mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex); + CDEBUG(D_LFSCK, "%s: RPC service thread mark the "DFID + " to be skipped for namespace double scan: rc = %d\n", + lfsck_lfsck2name(com->lc_lfsck), PFID(&lr->lr_fid), rc); - CDEBUG(D_LFSCK, "%s: namespace LFSCK handling notify from " - "MDT %x to create orphan"DFID" with type %o\n", - lfsck_lfsck2name(lfsck), lr->lr_index, - PFID(&lr->lr_fid), lr->lr_type); + if (rc != 0) + /* If we cannot record this object in the LFSCK tracing, + * we have to mark the LFSC as LF_INCOMPLETE, then the + * LFSCK will skip nlink attribute verification for + * all objects. */ + ns->ln_flags |= LF_INCOMPLETE; - orphan = lfsck_object_find(env, lfsck, &lr->lr_fid); - if (IS_ERR(orphan)) - GOTO(out_create, rc = PTR_ERR(orphan)); + return 0; + } + case LE_SET_LMV_MASTER: { + struct dt_object *obj; - if (dt_object_exists(orphan)) - GOTO(out_create, rc = -EEXIST); + obj = lfsck_object_find_bottom(env, lfsck, &lr->lr_fid); + if (IS_ERR(obj)) + RETURN(PTR_ERR(obj)); - rc = lfsck_namespace_create_orphan_local(env, com, orphan, - lr->lr_type); + if (likely(dt_object_exists(obj))) + rc = lfsck_namespace_notify_lmv_master_local(env, com, + obj); - GOTO(out_create, rc = (rc == 1) ? 0 : rc); + lfsck_object_put(env, obj); -out_create: - CDEBUG(D_LFSCK, "%s: namespace LFSCK handled notify from " - "MDT %x to create orphan"DFID" with type %o: rc = %d\n", - lfsck_lfsck2name(lfsck), lr->lr_index, - PFID(&lr->lr_fid), lr->lr_type, rc); + RETURN(rc > 0 ? 0 : rc); + } + case LE_SET_LMV_SLAVE: { + if (!(lr->lr_flags & LEF_RECHECK_NAME_HASH)) + ns->ln_striped_shards_repaired++; - if (orphan != NULL && !IS_ERR(orphan)) - lfsck_object_put(env, orphan); + rc = lfsck_namespace_trace_update(env, com, &lr->lr_fid, + LNTF_RECHECK_NAME_HASH, true); - return rc; + RETURN(rc > 0 ? 0 : rc); } case LE_PHASE1_DONE: case LE_PHASE2_DONE: @@ -3785,11 +4697,11 @@ out_create: } CDEBUG(D_LFSCK, "%s: namespace LFSCK handles notify %u from MDT %x, " - "status %d\n", lfsck_lfsck2name(lfsck), lr->lr_event, - lr->lr_index, lr->lr_status); + "status %d, flags %x\n", lfsck_lfsck2name(lfsck), lr->lr_event, + lr->lr_index, lr->lr_status, lr->lr_flags2); spin_lock(<ds->ltd_lock); - ltd = LTD_TGT(ltds, lr->lr_index); + ltd = lfsck_ltd2tgt(ltds, lr->lr_index); if (ltd == NULL) { spin_unlock(<ds->ltd_lock); @@ -3811,6 +4723,9 @@ out_create: break; } + if (lr->lr_flags2 & LF_INCOMPLETE) + ns->ln_flags |= LF_INCOMPLETE; + if (list_empty(<d->ltd_namespace_list)) list_add_tail(<d->ltd_namespace_list, &lad->lad_mdt_list); @@ -3851,17 +4766,88 @@ out_create: RETURN(0); } +static void lfsck_namespace_repaired(struct lfsck_namespace *ns, __u64 *count) +{ + *count += ns->ln_objs_nlink_repaired; + *count += ns->ln_dirent_repaired; + *count += ns->ln_linkea_repaired; + *count += ns->ln_mul_linked_repaired; + *count += ns->ln_unmatched_pairs_repaired; + *count += ns->ln_dangling_repaired; + *count += ns->ln_mul_ref_repaired; + *count += ns->ln_bad_type_repaired; + *count += ns->ln_lost_dirent_repaired; + *count += ns->ln_striped_dirs_disabled; + *count += ns->ln_striped_dirs_repaired; + *count += ns->ln_striped_shards_repaired; + *count += ns->ln_name_hash_repaired; + *count += ns->ln_local_lpf_moved; +} + +static int lfsck_namespace_query_all(const struct lu_env *env, + struct lfsck_component *com, + __u32 *mdts_count, __u64 *repaired) +{ + struct lfsck_namespace *ns = com->lc_file_ram; + struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs; + struct lfsck_tgt_desc *ltd; + int idx; + int rc; + ENTRY; + + rc = lfsck_query_all(env, com); + if (rc != 0) + RETURN(rc); + + down_read(<ds->ltd_rw_sem); + cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) { + ltd = lfsck_ltd2tgt(ltds, idx); + LASSERT(ltd != NULL); + + mdts_count[ltd->ltd_namespace_status]++; + *repaired += ltd->ltd_namespace_repaired; + } + up_read(<ds->ltd_rw_sem); + + down_read(&com->lc_sem); + mdts_count[ns->ln_status]++; + lfsck_namespace_repaired(ns, repaired); + up_read(&com->lc_sem); + + RETURN(0); +} + static int lfsck_namespace_query(const struct lu_env *env, - struct lfsck_component *com) + struct lfsck_component *com, + struct lfsck_request *req, + struct lfsck_reply *rep, + struct lfsck_query *que, int idx) { struct lfsck_namespace *ns = com->lc_file_ram; + int rc = 0; + + if (que != NULL) { + LASSERT(com->lc_lfsck->li_master); - return ns->ln_status; + rc = lfsck_namespace_query_all(env, com, + que->lu_mdts_count[idx], + &que->lu_repaired[idx]); + } else { + down_read(&com->lc_sem); + rep->lr_status = ns->ln_status; + if (req->lr_flags & LEF_QUERY_ALL) + lfsck_namespace_repaired(ns, &rep->lr_repaired); + up_read(&com->lc_sem); + } + + return rc; } static struct lfsck_operations lfsck_namespace_ops = { .lfsck_reset = lfsck_namespace_reset, .lfsck_fail = lfsck_namespace_fail, + .lfsck_close_dir = lfsck_namespace_close_dir, + .lfsck_open_dir = lfsck_namespace_open_dir, .lfsck_checkpoint = lfsck_namespace_checkpoint, .lfsck_prep = lfsck_namespace_prep, .lfsck_exec_oit = lfsck_namespace_exec_oit, @@ -3870,7 +4856,7 @@ static struct lfsck_operations lfsck_namespace_ops = { .lfsck_dump = lfsck_namespace_dump, .lfsck_double_scan = lfsck_namespace_double_scan, .lfsck_data_release = lfsck_namespace_data_release, - .lfsck_quit = lfsck_quit_generic, + .lfsck_quit = lfsck_namespace_quit, .lfsck_in_notify = lfsck_namespace_in_notify, .lfsck_query = lfsck_namespace_query, }; @@ -3890,6 +4876,8 @@ static struct lfsck_operations lfsck_namespace_ops = { * * \param[in] env pointer to the thread context * \param[in] com pointer to the lfsck component + * \param[in] parent pointer to the dir object that contains the dangling + * name entry * \param[in] child pointer to the object corresponding to the dangling * name entry * \param[in] lnr pointer to the namespace request that contains the @@ -3901,6 +4889,7 @@ static struct lfsck_operations lfsck_namespace_ops = { */ int lfsck_namespace_repair_dangling(const struct lu_env *env, struct lfsck_component *com, + struct dt_object *parent, struct dt_object *child, struct lfsck_namespace_req *lnr) { @@ -3909,14 +4898,17 @@ int lfsck_namespace_repair_dangling(const struct lu_env *env, struct dt_allocation_hint *hint = &info->lti_hint; struct dt_object_format *dof = &info->lti_dof; struct dt_insert_rec *rec = &info->lti_dt_rec; - struct dt_object *parent = lnr->lnr_obj; + struct lmv_mds_md_v1 *lmv2 = &info->lti_lmv2; const struct lu_name *cname; - struct linkea_data ldata = { 0 }; - struct lustre_handle lh = { 0 }; + const struct lu_fid *pfid = lfsck_dto2fid(parent); + const struct lu_fid *cfid = lfsck_dto2fid(child); + struct linkea_data ldata = { NULL }; + struct lfsck_lock_handle *llh = &info->lti_llh; struct lu_buf linkea_buf; + struct lu_buf lmv_buf; struct lfsck_instance *lfsck = com->lc_lfsck; struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram; - struct dt_device *dev = lfsck_obj2dt_dev(child); + struct dt_device *dev = lfsck->li_next; struct thandle *th = NULL; int rc = 0; __u16 type = lnr->lnr_type; @@ -3932,16 +4924,30 @@ int lfsck_namespace_repair_dangling(const struct lu_env *env, if (!create || bk->lb_param & LPF_DRYRUN) GOTO(log, rc = 0); + /* We may need to create the sub-objects of the @child via LOD, + * so make the modification based on lfsck->li_next. */ + + parent = lfsck_object_locate(dev, parent); + if (IS_ERR(parent)) + GOTO(log, rc = PTR_ERR(parent)); + + if (unlikely(!dt_try_as_dir(env, parent))) + GOTO(log, rc = -ENOTDIR); + + child = lfsck_object_locate(dev, child); + if (IS_ERR(child)) + GOTO(log, rc = PTR_ERR(child)); + rc = linkea_data_new(&ldata, &info->lti_linkea_buf2); if (rc != 0) GOTO(log, rc); - rc = linkea_add_buf(&ldata, cname, lfsck_dto2fid(parent)); + rc = linkea_add_buf(&ldata, cname, pfid); if (rc != 0) GOTO(log, rc); - rc = lfsck_ibits_lock(env, lfsck, parent, &lh, - MDS_INODELOCK_UPDATE, LCK_EX); + rc = lfsck_lock(env, lfsck, parent, lnr->lnr_name, llh, + MDS_INODELOCK_UPDATE, LCK_PR); if (rc != 0) GOTO(log, rc); @@ -3949,10 +4955,6 @@ int lfsck_namespace_repair_dangling(const struct lu_env *env, if (rc != 0) GOTO(log, rc); - th = dt_trans_create(env, dev); - if (IS_ERR(th)) - GOTO(log, rc = PTR_ERR(th)); - /* Set the ctime as zero, then others can know it is created for * repairing dangling name entry by LFSCK. And if the LFSCK made * wrong decision and the real MDT-object has been found later, @@ -3963,7 +4965,7 @@ int lfsck_namespace_repair_dangling(const struct lu_env *env, LA_ATIME | LA_MTIME | LA_CTIME; child->do_ops->do_ah_init(env, hint, parent, child, - la->la_mode & S_IFMT); + la->la_mode & S_IFMT); memset(dof, 0, sizeof(*dof)); dof->dof_type = dt_mode_to_dft(type); @@ -3971,6 +4973,10 @@ int lfsck_namespace_repair_dangling(const struct lu_env *env, * the MDT-object without stripes (dof->dof_reg.striped = 0). related * OST-objects will be created when write open. */ + th = dt_trans_create(env, dev); + if (IS_ERR(th)) + GOTO(log, rc = PTR_ERR(th)); + /* 1a. create child. */ rc = dt_declare_create(env, child, la, hint, dof, th); if (rc != 0) @@ -3980,30 +4986,52 @@ int lfsck_namespace_repair_dangling(const struct lu_env *env, if (unlikely(!dt_try_as_dir(env, child))) GOTO(stop, rc = -ENOTDIR); - /* 2a. insert dot into child dir */ + /* 2a. increase child nlink */ + rc = dt_declare_ref_add(env, child, th); + if (rc != 0) + GOTO(stop, rc); + + /* 3a. insert dot into child dir */ rec->rec_type = S_IFDIR; - rec->rec_fid = lfsck_dto2fid(child); + rec->rec_fid = cfid; rc = dt_declare_insert(env, child, (const struct dt_rec *)rec, (const struct dt_key *)dot, th); if (rc != 0) GOTO(stop, rc); - /* 3a. insert dotdot into child dir */ - rec->rec_fid = lfsck_dto2fid(parent); + /* 4a. insert dotdot into child dir */ + rec->rec_fid = pfid; rc = dt_declare_insert(env, child, (const struct dt_rec *)rec, (const struct dt_key *)dotdot, th); if (rc != 0) GOTO(stop, rc); - /* 4a. increase child nlink */ - rc = dt_declare_ref_add(env, child, th); - if (rc != 0) - GOTO(stop, rc); + /* 5a. generate slave LMV EA. */ + if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) { + int idx; + + idx = lfsck_shard_name_to_index(env, + lnr->lnr_name, lnr->lnr_namelen, + type, cfid); + if (unlikely(idx < 0)) + GOTO(stop, rc = idx); + + *lmv2 = lnr->lnr_lmv->ll_lmv; + lmv2->lmv_magic = LMV_MAGIC_STRIPE; + lmv2->lmv_master_mdt_index = idx; + + lfsck_lmv_header_cpu_to_le(lmv2, lmv2); + lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2)); + rc = dt_declare_xattr_set(env, child, &lmv_buf, + XATTR_NAME_LMV, 0, th); + if (rc != 0) + GOTO(stop, rc); + } } - /* 5a. insert linkEA for child */ + /* 6a. insert linkEA for child */ lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf, ldata.ld_leh->leh_len); rc = dt_declare_xattr_set(env, child, &linkea_buf, @@ -4011,7 +5039,7 @@ int lfsck_namespace_repair_dangling(const struct lu_env *env, if (rc != 0) GOTO(stop, rc); - rc = dt_trans_start(env, dev, th); + rc = dt_trans_start_local(env, dev, th); if (rc != 0) GOTO(stop, rc = (rc == -EEXIST ? 1 : rc)); @@ -4022,34 +5050,38 @@ int lfsck_namespace_repair_dangling(const struct lu_env *env, GOTO(unlock, rc = (rc == -EEXIST ? 1 : rc)); if (S_ISDIR(type)) { - if (unlikely(!dt_try_as_dir(env, child))) - GOTO(unlock, rc = -ENOTDIR); + /* 2b. increase child nlink */ + rc = dt_ref_add(env, child, th); + if (rc != 0) + GOTO(unlock, rc); - /* 2b. insert dot into child dir */ + /* 3b. insert dot into child dir */ rec->rec_type = S_IFDIR; - rec->rec_fid = lfsck_dto2fid(child); + rec->rec_fid = cfid; rc = dt_insert(env, child, (const struct dt_rec *)rec, - (const struct dt_key *)dot, th, BYPASS_CAPA, 1); + (const struct dt_key *)dot, th, 1); if (rc != 0) GOTO(unlock, rc); - /* 3b. insert dotdot into child dir */ - rec->rec_fid = lfsck_dto2fid(parent); + /* 4b. insert dotdot into child dir */ + rec->rec_fid = pfid; rc = dt_insert(env, child, (const struct dt_rec *)rec, - (const struct dt_key *)dotdot, th, - BYPASS_CAPA, 1); + (const struct dt_key *)dotdot, th, 1); if (rc != 0) GOTO(unlock, rc); - /* 4b. increase child nlink */ - rc = dt_ref_add(env, child, th); - if (rc != 0) - GOTO(unlock, rc); + /* 5b. generate slave LMV EA. */ + if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) { + rc = dt_xattr_set(env, child, &lmv_buf, XATTR_NAME_LMV, + 0, th); + if (rc != 0) + GOTO(unlock, rc); + } } - /* 5b. insert linkEA for child. */ + /* 6b. insert linkEA for child. */ rc = dt_xattr_set(env, child, &linkea_buf, - XATTR_NAME_LINK, 0, th, BYPASS_CAPA); + XATTR_NAME_LINK, 0, th); GOTO(unlock, rc); @@ -4060,13 +5092,12 @@ stop: dt_trans_stop(env, dev, th); log: - lfsck_ibits_unlock(&lh, LCK_EX); + lfsck_unlock(llh); CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found dangling " "reference for: parent "DFID", child "DFID", type %u, " "name %s. %s: rc = %d\n", lfsck_lfsck2name(lfsck), - PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)), - type, cname->ln_name, - create ? "Create the lost OST-object as required" : + PFID(pfid), PFID(cfid), type, cname->ln_name, + create ? "Create the lost MDT-object as required" : "Keep the MDT-object there by default", rc); if (rc <= 0) { @@ -4087,28 +5118,36 @@ static int lfsck_namespace_assistant_handler_p1(const struct lu_env *env, struct lfsck_instance *lfsck = com->lc_lfsck; struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram; struct lfsck_namespace *ns = com->lc_file_ram; - struct linkea_data ldata = { 0 }; + struct lfsck_assistant_data *lad = com->lc_data; + struct linkea_data ldata = { NULL }; const struct lu_name *cname; struct thandle *handle = NULL; struct lfsck_namespace_req *lnr = container_of0(lar, struct lfsck_namespace_req, lnr_lar); - struct dt_object *dir = lnr->lnr_obj; + struct dt_object *dir = NULL; struct dt_object *obj = NULL; - const struct lu_fid *pfid = lfsck_dto2fid(dir); - struct dt_device *dev; + struct lfsck_assistant_object *lso = lar->lar_parent; + const struct lu_fid *pfid = &lso->lso_fid; + struct dt_device *dev = NULL; struct lustre_handle lh = { 0 }; bool repaired = false; bool dtlocked = false; bool remove; bool newdata; bool log = false; - int idx; + bool bad_hash = false; + bool bad_linkea = false; + int idx = 0; int count = 0; - int rc; + int rc = 0; enum lfsck_namespace_inconsistency_type type = LNIT_NONE; ENTRY; - if (lnr->lnr_attr & LUDA_UPGRADE) { + if (lso->lso_dead) + RETURN(0); + + la->la_nlink = 0; + if (lnr->lnr_attr & LUDA_UPGRADE) { ns->ln_flags |= LF_UPGRADE; ns->ln_dirent_repaired++; repaired = true; @@ -4128,19 +5167,48 @@ static int lfsck_namespace_assistant_handler_p1(const struct lu_env *env, GOTO(out, rc); } - if (lnr->lnr_name[0] == '.' && - (lnr->lnr_namelen == 1 || fid_seq_is_dot(fid_seq(&lnr->lnr_fid)))) + if (unlikely(!fid_is_sane(&lnr->lnr_fid))) { + CDEBUG(D_LFSCK, "%s: dir scan find invalid FID "DFID + " for the name entry %.*s under "DFID"\n", + lfsck_lfsck2name(lfsck), PFID(&lnr->lnr_fid), + lnr->lnr_namelen, lnr->lnr_name, PFID(pfid)); + + if (strcmp(lnr->lnr_name, dotdot) != 0) + /* invalid FID means bad name entry, remove it. */ + type = LNIT_BAD_DIRENT; + else + /* If the parent FID is invalid, we cannot remove + * the ".." entry directly. */ + rc = lfsck_namespace_trace_update(env, com, pfid, + LNTF_CHECK_PARENT, true); + + GOTO(out, rc); + } + + if (unlikely(lnr->lnr_dir_cookie == MDS_DIR_END_OFF)) { + rc = lfsck_namespace_striped_dir_rescan(env, com, lnr); + + RETURN(rc); + } + + if (fid_seq_is_dot(fid_seq(&lnr->lnr_fid))) GOTO(out, rc = 0); + if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) { + rc = lfsck_namespace_handle_striped_master(env, com, lnr); + + RETURN(rc); + } + idx = lfsck_find_mdt_idx_by_fid(env, lfsck, &lnr->lnr_fid); if (idx < 0) GOTO(out, rc = idx); - if (idx == lfsck_dev_idx(lfsck->li_bottom)) { + if (idx == lfsck_dev_idx(lfsck)) { if (unlikely(strcmp(lnr->lnr_name, dotdot) == 0)) GOTO(out, rc = 0); - dev = lfsck->li_next; + dev = lfsck->li_bottom; } else { struct lfsck_tgt_desc *ltd; @@ -4154,12 +5222,12 @@ static int lfsck_namespace_assistant_handler_p1(const struct lu_env *env, GOTO(out, rc); } - ltd = LTD_TGT(&lfsck->li_mdt_descs, idx); + ltd = lfsck_ltd2tgt(&lfsck->li_mdt_descs, idx); if (unlikely(ltd == NULL)) { CDEBUG(D_LFSCK, "%s: cannot talk with MDT %x which " "did not join the namespace LFSCK\n", lfsck_lfsck2name(lfsck), idx); - ns->ln_flags |= LF_INCOMPLETE; + lfsck_lad_set_bitmap(env, com, idx); GOTO(out, rc = -ENODEV); } @@ -4175,10 +5243,26 @@ static int lfsck_namespace_assistant_handler_p1(const struct lu_env *env, if (dt_object_exists(obj) == 0) { dangling: + if (dir == NULL) { + dir = lfsck_assistant_object_load(env, lfsck, lso); + if (IS_ERR(dir)) { + rc = PTR_ERR(dir); + + GOTO(trace, rc == -ENOENT ? 0 : rc); + } + } + rc = lfsck_namespace_check_exist(env, dir, obj, lnr->lnr_name); if (rc == 0) { + if (!lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv, + lnr->lnr_name, lnr->lnr_namelen)) { + type = LNIT_BAD_DIRENT; + + GOTO(out, rc); + } + type = LNIT_DANGLING; - rc = lfsck_namespace_repair_dangling(env, com, + rc = lfsck_namespace_repair_dangling(env, com, dir, obj, lnr); if (rc == 0) repaired = true; @@ -4187,7 +5271,7 @@ dangling: GOTO(out, rc); } - if (!(bk->lb_param & LPF_DRYRUN) && repaired) { + if (!(bk->lb_param & LPF_DRYRUN) && lad->lad_advance_lock) { again: rc = lfsck_ibits_lock(env, lfsck, obj, &lh, @@ -4204,7 +5288,7 @@ again: if (rc != 0) GOTO(stop, rc); - rc = dt_trans_start(env, dev, handle); + rc = dt_trans_start_local(env, dev, handle); if (rc != 0) GOTO(stop, rc); @@ -4212,10 +5296,6 @@ again: dtlocked = true; } - rc = lfsck_namespace_check_exist(env, dir, obj, lnr->lnr_name); - if (rc != 0) - GOTO(stop, rc); - rc = lfsck_links_read(env, obj, &ldata); if (unlikely(rc == -ENOENT)) { if (handle != NULL) { @@ -4242,16 +5322,26 @@ again: type = LNIT_BAD_TYPE; } - goto record; + goto stop; } - ns->ln_flags |= LF_INCONSISTENT; + /* If the name entry hash does not match the slave striped + * directory, and the name entry does not match also, then + * it is quite possible that name entry is corrupted. */ + if (!lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv, + lnr->lnr_name, lnr->lnr_namelen)) { + ns->ln_flags |= LF_INCONSISTENT; + type = LNIT_BAD_DIRENT; + + GOTO(stop, rc = 0); + } /* If the file type stored in the name entry does not match * the file type claimed by the object, and the object does * not recognize the name entry, then it is quite possible * that the name entry is corrupted. */ if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type) { + ns->ln_flags |= LF_INCONSISTENT; type = LNIT_BAD_DIRENT; GOTO(stop, rc = 0); @@ -4274,7 +5364,6 @@ again: type = LNIT_BAD_TYPE; count = 1; - ns->ln_flags |= LF_INCONSISTENT; /* The magic crashed, we are not sure whether there are more * corrupt data in the linkea, so remove all linkea entries. */ remove = true; @@ -4285,27 +5374,53 @@ again: type = LNIT_BAD_TYPE; count = 1; - ns->ln_flags |= LF_UPGRADE; remove = false; newdata = true; nodata: if (bk->lb_param & LPF_DRYRUN) { + if (rc == -ENODATA) + ns->ln_flags |= LF_UPGRADE; + else + ns->ln_flags |= LF_INCONSISTENT; ns->ln_linkea_repaired++; repaired = true; log = true; - goto record; + goto stop; } - if (!lustre_handle_is_used(&lh)) + if (!lustre_handle_is_used(&lh)) { + remove = false; + newdata = false; + type = LNIT_NONE; + goto again; + } + + if (dir == NULL) { + dir = lfsck_assistant_object_load(env, lfsck, lso); + if (IS_ERR(dir)) { + rc = PTR_ERR(dir); + + GOTO(stop, rc == -ENOENT ? 0 : rc); + } + } + + rc = lfsck_namespace_check_exist(env, dir, obj, lnr->lnr_name); + if (rc != 0) + GOTO(stop, rc); + + bad_linkea = true; + if (!remove && newdata) + ns->ln_flags |= LF_UPGRADE; + else if (remove || !(ns->ln_flags & LF_UPGRADE)) + ns->ln_flags |= LF_INCONSISTENT; if (remove) { LASSERT(newdata); - rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, handle, - BYPASS_CAPA); - if (rc != 0) + rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, handle); + if (rc != 0 && rc != -ENOENT && rc != -ENODATA) GOTO(stop, rc); } @@ -4321,6 +5436,33 @@ nodata: GOTO(stop, rc); rc = lfsck_links_write(env, obj, &ldata, handle); + if (unlikely(rc == -ENOSPC) && + S_ISREG(lfsck_object_type(obj)) && !dt_object_remote(obj)) { + if (handle != NULL) { + LASSERT(dt_write_locked(env, obj)); + + dt_write_unlock(env, obj); + dtlocked = false; + + dt_trans_stop(env, dev, handle); + handle = NULL; + + lfsck_ibits_unlock(&lh, LCK_EX); + } + + rc = lfsck_namespace_trace_update(env, com, + &lnr->lnr_fid, LNTF_SKIP_NLINK, true); + if (rc != 0) + /* If we cannot record this object in the + * LFSCK tracing, we have to mark the LFSCK + * as LF_INCOMPLETE, then the LFSCK will + * skip nlink attribute verification for + * all objects. */ + ns->ln_flags |= LF_INCOMPLETE; + + GOTO(out, rc = 0); + } + if (rc != 0) GOTO(stop, rc); @@ -4335,35 +5477,6 @@ nodata: GOTO(stop, rc); } -record: - LASSERT(count > 0); - - rc = dt_attr_get(env, obj, la, BYPASS_CAPA); - if (rc != 0) - GOTO(stop, rc); - - if ((count == 1 && la->la_nlink == 1) || - S_ISDIR(lfsck_object_type(obj))) - /* Usually, it is for single linked object or dir, do nothing.*/ - GOTO(stop, rc); - - /* Following modification will be in another transaction. */ - if (handle != NULL) { - dt_write_unlock(env, obj); - dtlocked = false; - - dt_trans_stop(env, dev, handle); - handle = NULL; - - lfsck_ibits_unlock(&lh, LCK_EX); - } - - ns->ln_mul_linked_checked++; - rc = lfsck_namespace_trace_update(env, com, &lnr->lnr_fid, - LNTF_CHECK_LINKEA, true); - - GOTO(out, rc); - stop: if (dtlocked) dt_write_unlock(env, obj); @@ -4374,7 +5487,38 @@ stop: out: lfsck_ibits_unlock(&lh, LCK_EX); + if (!name_is_dot_or_dotdot(lnr->lnr_name, lnr->lnr_namelen) && + !lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv, + lnr->lnr_name, lnr->lnr_namelen) && + type != LNIT_BAD_DIRENT) { + ns->ln_flags |= LF_INCONSISTENT; + + log = false; + if (dir == NULL) { + dir = lfsck_assistant_object_load(env, lfsck, lso); + if (IS_ERR(dir)) { + rc = PTR_ERR(dir); + + GOTO(trace, rc == -ENOENT ? 0 : rc); + } + } + + rc = lfsck_namespace_repair_bad_name_hash(env, com, dir, + lnr->lnr_lmv, lnr->lnr_name); + if (rc == 0) + bad_hash = true; + } + if (rc >= 0) { + if (type != LNIT_NONE && dir == NULL) { + dir = lfsck_assistant_object_load(env, lfsck, lso); + if (IS_ERR(dir)) { + rc = PTR_ERR(dir); + + GOTO(trace, rc == -ENOENT ? 0 : rc); + } + } + switch (type) { case LNIT_BAD_TYPE: log = false; @@ -4399,30 +5543,38 @@ out: default: break; } + + if (obj != NULL && count == 1 && + S_ISREG(lfsck_object_type(obj))) + dt_attr_get(env, obj, la); } +trace: down_write(&com->lc_sem); if (rc < 0) { CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail to handle " "the entry: "DFID", parent "DFID", name %.*s: rc = %d\n", - lfsck_lfsck2name(lfsck), PFID(&lnr->lnr_fid), - PFID(lfsck_dto2fid(lnr->lnr_obj)), + lfsck_lfsck2name(lfsck), PFID(&lnr->lnr_fid), PFID(pfid), lnr->lnr_namelen, lnr->lnr_name, rc); lfsck_namespace_record_failure(env, lfsck, ns); + if ((rc == -ENOTCONN || rc == -ESHUTDOWN || rc == -EREMCHG || + rc == -ETIMEDOUT || rc == -EHOSTDOWN || + rc == -EHOSTUNREACH || rc == -EINPROGRESS) && + dev != NULL && dev != lfsck->li_bottom) + lfsck_lad_set_bitmap(env, com, idx); + if (!(bk->lb_param & LPF_FAILOUT)) rc = 0; } else { - if (log) - CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant " - "repaired the entry: "DFID", parent "DFID - ", name %.*s\n", lfsck_lfsck2name(lfsck), - PFID(&lnr->lnr_fid), - PFID(lfsck_dto2fid(lnr->lnr_obj)), - lnr->lnr_namelen, lnr->lnr_name); - if (repaired) { ns->ln_items_repaired++; + if (log) + CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant " + "repaired the entry: "DFID", parent "DFID + ", name %.*s\n", lfsck_lfsck2name(lfsck), + PFID(&lnr->lnr_fid), PFID(pfid), + lnr->lnr_namelen, lnr->lnr_name); switch (type) { case LNIT_DANGLING: @@ -4445,23 +5597,500 @@ out: false); } + if (bad_hash) { + ns->ln_name_hash_repaired++; + + /* Not count repeatedly. */ + if (!repaired) { + ns->ln_items_repaired++; + if (log) + CDEBUG(D_LFSCK, "%s: namespace LFSCK " + "assistant repaired the entry: " + DFID", parent "DFID + ", name %.*s\n", + lfsck_lfsck2name(lfsck), + PFID(&lnr->lnr_fid), PFID(pfid), + lnr->lnr_namelen, lnr->lnr_name); + } + + if (bk->lb_param & LPF_DRYRUN && + lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent)) + lfsck_pos_fill(env, lfsck, + &ns->ln_pos_first_inconsistent, + false); + } + rc = 0; } + + if (count > 1 || la->la_nlink > 1) + ns->ln_mul_linked_checked++; + up_write(&com->lc_sem); if (obj != NULL && !IS_ERR(obj)) lfsck_object_put(env, obj); + + if (dir != NULL && !IS_ERR(dir)) + lfsck_object_put(env, dir); + + lad->lad_advance_lock = bad_linkea; + return rc; } -static int lfsck_namespace_assistant_handler_p2(const struct lu_env *env, - struct lfsck_component *com) +/** + * Handle one orphan under the backend /lost+found directory + * + * Insert the orphan FID into the namespace LFSCK trace file for further + * processing (via the subsequent namespace LFSCK second-stage scanning). + * At the same time, remove the orphan name entry from backend /lost+found + * directory. There is an interval between the orphan name entry removed + * from the backend /lost+found directory and the orphan FID in the LFSCK + * trace file handled. In such interval, the LFSCK can be reset, then + * all the FIDs recorded in the namespace LFSCK trace file will be dropped. + * To guarantee that the orphans can be found when LFSCK run next time + * without e2fsck again, when remove the orphan name entry, the LFSCK + * will set the orphan's ctime attribute as 1. Since normal applications + * cannot change the object's ctime attribute as 1. Then when LFSCK run + * next time, it can record the object (that ctime is 1) in the namespace + * LFSCK trace file during the first-stage scanning. + * + * \param[in] env pointer to the thread context + * \param[in] com pointer to the lfsck component + * \param[in] parent pointer to the object for the backend /lost+found + * \param[in] ent pointer to the name entry for the target under the + * backend /lost+found + * + * \retval positive for repaired + * \retval 0 if needs to repair nothing + * \retval negative error number on failure + */ +static int lfsck_namespace_scan_local_lpf_one(const struct lu_env *env, + struct lfsck_component *com, + struct dt_object *parent, + struct lu_dirent *ent) +{ + struct lfsck_thread_info *info = lfsck_env_info(env); + struct lu_fid *key = &info->lti_fid; + struct lu_attr *la = &info->lti_la; + struct lfsck_instance *lfsck = com->lc_lfsck; + struct dt_object *obj; + struct dt_device *dev = lfsck->li_bottom; + struct dt_object *child = NULL; + struct thandle *th = NULL; + int idx; + int rc = 0; + __u8 flags = 0; + bool exist = false; + ENTRY; + + child = lfsck_object_find_by_dev(env, dev, &ent->lde_fid); + if (IS_ERR(child)) + RETURN(PTR_ERR(child)); + + LASSERT(dt_object_exists(child)); + LASSERT(!dt_object_remote(child)); + + idx = lfsck_sub_trace_file_fid2idx(&ent->lde_fid); + obj = com->lc_sub_trace_objs[idx].lsto_obj; + fid_cpu_to_be(key, &ent->lde_fid); + rc = dt_lookup(env, obj, (struct dt_rec *)&flags, + (const struct dt_key *)key); + if (rc == 0) { + exist = true; + flags |= LNTF_CHECK_ORPHAN; + } else if (rc == -ENOENT) { + flags = LNTF_CHECK_ORPHAN; + } else { + GOTO(out, rc); + } + + th = dt_trans_create(env, dev); + if (IS_ERR(th)) + GOTO(out, rc = PTR_ERR(th)); + + /* a1. remove name entry from backend /lost+found */ + rc = dt_declare_delete(env, parent, + (const struct dt_key *)ent->lde_name, th); + if (rc != 0) + GOTO(stop, rc); + + if (S_ISDIR(lfsck_object_type(child))) { + /* a2. decrease parent's nlink */ + rc = dt_declare_ref_del(env, parent, th); + if (rc != 0) + GOTO(stop, rc); + } + + if (exist) { + /* a3. remove child's FID from the LFSCK trace file. */ + rc = dt_declare_delete(env, obj, + (const struct dt_key *)key, th); + if (rc != 0) + GOTO(stop, rc); + } else { + /* a4. set child's ctime as 1 */ + memset(la, 0, sizeof(*la)); + la->la_ctime = 1; + la->la_valid = LA_CTIME; + rc = dt_declare_attr_set(env, child, la, th); + if (rc != 0) + GOTO(stop, rc); + } + + /* a5. insert child's FID into the LFSCK trace file. */ + rc = dt_declare_insert(env, obj, (const struct dt_rec *)&flags, + (const struct dt_key *)key, th); + if (rc != 0) + GOTO(stop, rc); + + rc = dt_trans_start_local(env, dev, th); + if (rc != 0) + GOTO(stop, rc); + + /* b1. remove name entry from backend /lost+found */ + rc = dt_delete(env, parent, (const struct dt_key *)ent->lde_name, th); + if (rc != 0) + GOTO(stop, rc); + + if (S_ISDIR(lfsck_object_type(child))) { + /* b2. decrease parent's nlink */ + dt_write_lock(env, parent, 0); + rc = dt_ref_del(env, parent, th); + dt_write_unlock(env, parent); + if (rc != 0) + GOTO(stop, rc); + } + + if (exist) { + /* a3. remove child's FID from the LFSCK trace file. */ + rc = dt_delete(env, obj, (const struct dt_key *)key, th); + if (rc != 0) + GOTO(stop, rc); + } else { + /* b4. set child's ctime as 1 */ + rc = dt_attr_set(env, child, la, th); + if (rc != 0) + GOTO(stop, rc); + } + + /* b5. insert child's FID into the LFSCK trace file. */ + rc = dt_insert(env, obj, (const struct dt_rec *)&flags, + (const struct dt_key *)key, th, 1); + + GOTO(stop, rc = (rc == 0 ? 1 : rc)); + +stop: + dt_trans_stop(env, dev, th); + +out: + lfsck_object_put(env, child); + + return rc; +} + +/** + * Handle orphans under the backend /lost+found directory + * + * Some backend checker, such as e2fsck for ldiskfs may find some orphans + * and put them under the backend /lost+found directory that is invisible + * to client. The LFSCK will scan such directory, for the original client + * visible orphans, add their fids into the namespace LFSCK trace file, + * then the subsenquent namespace LFSCK second-stage scanning can handle + * them as other objects to be double scanned: either move back to normal + * namespace, or to the global visible orphan directory: + * /ROOT/.lustre/lost+found/MDTxxxx/ + * + * \param[in] env pointer to the thread context + * \param[in] com pointer to the lfsck component + */ +static void lfsck_namespace_scan_local_lpf(const struct lu_env *env, + struct lfsck_component *com) +{ + struct lfsck_thread_info *info = lfsck_env_info(env); + struct lu_dirent *ent = + (struct lu_dirent *)info->lti_key; + struct lu_seq_range *range = &info->lti_range; + struct lfsck_instance *lfsck = com->lc_lfsck; + struct ptlrpc_thread *thread = &lfsck->li_thread; + struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram; + struct lfsck_namespace *ns = com->lc_file_ram; + struct dt_object *parent; + const struct dt_it_ops *iops; + struct dt_it *di; + struct seq_server_site *ss = lfsck_dev_site(lfsck); + __u64 cookie; + __u32 idx = lfsck_dev_idx(lfsck); + int rc = 0; + __u16 type; + ENTRY; + + parent = lfsck_object_find_by_dev(env, lfsck->li_bottom, + &LU_BACKEND_LPF_FID); + if (IS_ERR(parent)) { + CERROR("%s: fail to find backend /lost+found: rc = %ld\n", + lfsck_lfsck2name(lfsck), PTR_ERR(parent)); + RETURN_EXIT; + } + + /* It is normal that the /lost+found does not exist for ZFS backend. */ + if (!dt_object_exists(parent)) + GOTO(out, rc = 0); + + if (unlikely(!dt_try_as_dir(env, parent))) + GOTO(out, rc = -ENOTDIR); + + CDEBUG(D_LFSCK, "%s: start to scan backend /lost+found\n", + lfsck_lfsck2name(lfsck)); + + com->lc_new_scanned = 0; + iops = &parent->do_index_ops->dio_it; + di = iops->init(env, parent, LUDA_64BITHASH | LUDA_TYPE); + if (IS_ERR(di)) + GOTO(out, rc = PTR_ERR(di)); + + rc = iops->load(env, di, 0); + if (rc == 0) + rc = iops->next(env, di); + else if (rc > 0) + rc = 0; + + while (rc == 0) { + if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) && + unlikely(!thread_is_running(thread))) + break; + + rc = iops->rec(env, di, (struct dt_rec *)ent, + LUDA_64BITHASH | LUDA_TYPE); + if (rc == 0) + rc = lfsck_unpack_ent(ent, &cookie, &type); + + if (unlikely(rc != 0)) { + CDEBUG(D_LFSCK, "%s: fail to iterate backend " + "/lost+found: rc = %d\n", + lfsck_lfsck2name(lfsck), rc); + + goto skip; + } + + /* skip dot and dotdot entries */ + if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen)) + goto next; + + if (!fid_seq_in_fldb(fid_seq(&ent->lde_fid))) + goto skip; + + if (fid_is_norm(&ent->lde_fid)) { + fld_range_set_mdt(range); + rc = fld_local_lookup(env, ss->ss_server_fld, + fid_seq(&ent->lde_fid), range); + if (rc != 0) + goto skip; + } else if (idx != 0) { + /* If the returned FID is IGIF, then there are three + * possible cases: + * + * 1) The object is upgraded from old Lustre-1.8 with + * IGIF assigned to such object. + * 2) The object is a backend local object and is + * invisible to client. + * 3) The object lost its LMV EA, and since there is + * no FID-in-dirent for the orphan in the backend + * /lost+found directory, then the low layer will + * return IGIF for such object. + * + * For MDTx (x != 0), it is either case 2) or case 3), + * but from the LFSCK view, they are indistinguishable. + * To be safe, the LFSCK will keep it there and report + * some message, then the adminstrator can handle that + * furtherly. + * + * For MDT0, it is more possible the case 1). The LFSCK + * will handle the orphan as an upgraded object. */ + CDEBUG(D_LFSCK, "%s: the orphan %.*s with IGIF "DFID + "in the backend /lost+found on the MDT %04x, " + "to be safe, skip it.\n", + lfsck_lfsck2name(lfsck), ent->lde_namelen, + ent->lde_name, PFID(&ent->lde_fid), idx); + goto skip; + } + + rc = lfsck_namespace_scan_local_lpf_one(env, com, parent, ent); + +skip: + down_write(&com->lc_sem); + com->lc_new_scanned++; + ns->ln_local_lpf_scanned++; + if (rc > 0) + ns->ln_local_lpf_moved++; + else if (rc == 0) + ns->ln_local_lpf_skipped++; + else + ns->ln_local_lpf_failed++; + up_write(&com->lc_sem); + + if (rc < 0 && bk->lb_param & LPF_FAILOUT) + break; + +next: + lfsck_control_speed_by_self(com); + if (unlikely(!thread_is_running(thread))) { + rc = 0; + break; + } + + rc = iops->next(env, di); + } + + iops->put(env, di); + iops->fini(env, di); + + EXIT; + +out: + CDEBUG(D_LFSCK, "%s: stop to scan backend /lost+found: rc = %d\n", + lfsck_lfsck2name(lfsck), rc); + + lfsck_object_put(env, parent); +} + +/** + * Rescan the striped directory after the master LMV EA reset. + * + * Sometimes, the master LMV EA of the striped directory maybe lost, so when + * the namespace LFSCK engine scan the striped directory for the first time, + * it will be regarded as a normal directory. As the LFSCK processing, some + * other LFSCK instance on other MDT will find the shard of this striped dir, + * and find that the master MDT-object of the striped directory lost its LMV + * EA, then such remote LFSCK instance will regenerate the master LMV EA and + * notify the LFSCK instance on this MDT to rescan the striped directory. + * + * \param[in] env pointer to the thread context + * \param[in] com pointer to the lfsck component + * \param[in] llu the lfsck_lmv_unit that contains the striped directory + * to be rescanned. + * + * \retval positive number for success + * \retval 0 for LFSCK stopped/paused + * \retval negative error number on failure + */ +static int lfsck_namespace_rescan_striped_dir(const struct lu_env *env, + struct lfsck_component *com, + struct lfsck_lmv_unit *llu) +{ + struct lfsck_thread_info *info = lfsck_env_info(env); + struct lfsck_instance *lfsck = com->lc_lfsck; + struct lfsck_assistant_data *lad = com->lc_data; + struct dt_object *dir; + const struct dt_it_ops *iops; + struct dt_it *di; + struct lu_dirent *ent = + (struct lu_dirent *)info->lti_key; + struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram; + struct ptlrpc_thread *thread = &lfsck->li_thread; + struct lfsck_assistant_object *lso = NULL; + struct lfsck_namespace_req *lnr; + struct lfsck_assistant_req *lar; + int rc; + __u16 type; + ENTRY; + + LASSERT(list_empty(&lad->lad_req_list)); + + lfsck->li_lmv = &llu->llu_lmv; + lfsck->li_obj_dir = lfsck_object_get(llu->llu_obj); + rc = lfsck_open_dir(env, lfsck, 0); + if (rc != 0) + RETURN(rc); + + dir = lfsck->li_obj_dir; + di = lfsck->li_di_dir; + iops = &dir->do_index_ops->dio_it; + do { + rc = iops->rec(env, di, (struct dt_rec *)ent, + lfsck->li_args_dir); + if (rc == 0) + rc = lfsck_unpack_ent(ent, &lfsck->li_cookie_dir, + &type); + + if (rc != 0) { + if (bk->lb_param & LPF_FAILOUT) + GOTO(out, rc); + + goto next; + } + + if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen)) + goto next; + + if (lso == NULL) { + lso = lfsck_assistant_object_init(env, + lfsck_dto2fid(dir), NULL, + lfsck->li_pos_current.lp_oit_cookie, true); + if (IS_ERR(lso)) { + if (bk->lb_param & LPF_FAILOUT) + GOTO(out, rc = PTR_ERR(lso)); + + lso = NULL; + goto next; + } + } + + lnr = lfsck_namespace_assistant_req_init(lfsck, lso, ent, type); + if (IS_ERR(lnr)) { + if (bk->lb_param & LPF_FAILOUT) + GOTO(out, rc = PTR_ERR(lnr)); + + goto next; + } + + lar = &lnr->lnr_lar; + rc = lfsck_namespace_assistant_handler_p1(env, com, lar); + lfsck_namespace_assistant_req_fini(env, lar); + if (rc != 0 && bk->lb_param & LPF_FAILOUT) + GOTO(out, rc); + + if (unlikely(!thread_is_running(thread))) + GOTO(out, rc = 0); + +next: + rc = iops->next(env, di); + } while (rc == 0); + +out: + if (lso != NULL && !IS_ERR(lso)) + lfsck_assistant_object_put(env, lso); + + lfsck_close_dir(env, lfsck, rc); + if (rc <= 0) + RETURN(rc); + + /* The close_dir() may insert a dummy lnr in the lad->lad_req_list. */ + if (list_empty(&lad->lad_req_list)) + RETURN(1); + + spin_lock(&lad->lad_lock); + lar = list_entry(lad->lad_req_list.next, struct lfsck_assistant_req, + lar_list); + list_del_init(&lar->lar_list); + spin_unlock(&lad->lad_lock); + + rc = lfsck_namespace_assistant_handler_p1(env, com, lar); + lfsck_namespace_assistant_req_fini(env, lar); + + RETURN(rc == 0 ? 1 : rc); +} + +static int +lfsck_namespace_double_scan_one_trace_file(const struct lu_env *env, + struct lfsck_component *com, + struct dt_object *obj, bool first) { struct lfsck_instance *lfsck = com->lc_lfsck; struct ptlrpc_thread *thread = &lfsck->li_thread; struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram; struct lfsck_namespace *ns = com->lc_file_ram; - struct dt_object *obj = com->lc_obj; const struct dt_it_ops *iops = &obj->do_index_ops->dio_it; struct dt_object *target; struct dt_it *di; @@ -4471,52 +6100,47 @@ static int lfsck_namespace_assistant_handler_p2(const struct lu_env *env, __u8 flags = 0; ENTRY; - CDEBUG(D_LFSCK, "%s: namespace LFSCK phase2 scan start\n", - lfsck_lfsck2name(lfsck)); - - com->lc_new_checked = 0; - com->lc_new_scanned = 0; - com->lc_time_last_checkpoint = cfs_time_current(); - com->lc_time_next_checkpoint = com->lc_time_last_checkpoint + - cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL); - - di = iops->init(env, obj, 0, BYPASS_CAPA); + di = iops->init(env, obj, 0); if (IS_ERR(di)) RETURN(PTR_ERR(di)); - fid_cpu_to_be(&fid, &ns->ln_fid_latest_scanned_phase2); + if (first) + fid_cpu_to_be(&fid, &ns->ln_fid_latest_scanned_phase2); + else + fid_zero(&fid); rc = iops->get(env, di, (const struct dt_key *)&fid); if (rc < 0) GOTO(fini, rc); - /* Skip the start one, which either has been processed or non-exist. */ - rc = iops->next(env, di); - if (rc != 0) - GOTO(put, rc); + if (first) { + /* The start one either has been processed or does not exist, + * skip it. */ + rc = iops->next(env, di); + if (rc != 0) + GOTO(put, rc); + } do { - if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY3) && - cfs_fail_val > 0) { - struct l_wait_info lwi; - - lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val), - NULL, NULL); - l_wait_event(thread->t_ctl_waitq, - !thread_is_running(thread), - &lwi); - - if (unlikely(!thread_is_running(thread))) - GOTO(put, rc = 0); - } + if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) && + unlikely(!thread_is_running(thread))) + GOTO(put, rc = 0); key = iops->key(env, di); + if (IS_ERR(key)) { + rc = PTR_ERR(key); + if (rc == -ENOENT) + GOTO(put, rc = 1); + + goto checkpoint; + } + fid_be_to_cpu(&fid, (const struct lu_fid *)key); if (!fid_is_sane(&fid)) { rc = 0; goto checkpoint; } - target = lfsck_object_find(env, lfsck, &fid); + target = lfsck_object_find_bottom(env, lfsck, &fid); if (IS_ERR(target)) { rc = PTR_ERR(target); goto checkpoint; @@ -4538,7 +6162,8 @@ checkpoint: down_write(&com->lc_sem); com->lc_new_checked++; com->lc_new_scanned++; - ns->ln_fid_latest_scanned_phase2 = fid; + if (rc >= 0 && fid_is_sane(&fid)) + ns->ln_fid_latest_scanned_phase2 = fid; if (rc > 0) ns->ln_objs_repaired_phase2++; else if (rc < 0) @@ -4558,7 +6183,7 @@ checkpoint: ns->ln_time_last_checkpoint = cfs_time_current_sec(); ns->ln_objs_checked_phase2 += com->lc_new_checked; com->lc_new_checked = 0; - rc = lfsck_namespace_store(env, com, false); + rc = lfsck_namespace_store(env, com); up_write(&com->lc_sem); if (rc != 0) GOTO(put, rc); @@ -4584,12 +6209,56 @@ put: fini: iops->fini(env, di); - CDEBUG(D_LFSCK, "%s: namespace LFSCK phase2 scan stop: rc = %d\n", - lfsck_lfsck2name(lfsck), rc); - return rc; } +static int lfsck_namespace_assistant_handler_p2(const struct lu_env *env, + struct lfsck_component *com) +{ + struct lfsck_instance *lfsck = com->lc_lfsck; + struct lfsck_namespace *ns = com->lc_file_ram; + int rc; + int i; + ENTRY; + + while (!list_empty(&lfsck->li_list_lmv)) { + struct lfsck_lmv_unit *llu; + + spin_lock(&lfsck->li_lock); + llu = list_entry(lfsck->li_list_lmv.next, + struct lfsck_lmv_unit, llu_link); + list_del_init(&llu->llu_link); + spin_unlock(&lfsck->li_lock); + + rc = lfsck_namespace_rescan_striped_dir(env, com, llu); + if (rc <= 0) + RETURN(rc); + } + + CDEBUG(D_LFSCK, "%s: namespace LFSCK phase2 scan start\n", + lfsck_lfsck2name(lfsck)); + + lfsck_namespace_scan_local_lpf(env, com); + + com->lc_new_checked = 0; + com->lc_new_scanned = 0; + com->lc_time_last_checkpoint = cfs_time_current(); + com->lc_time_next_checkpoint = com->lc_time_last_checkpoint + + cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL); + + i = lfsck_sub_trace_file_fid2idx(&ns->ln_fid_latest_scanned_phase2); + rc = lfsck_namespace_double_scan_one_trace_file(env, com, + com->lc_sub_trace_objs[i].lsto_obj, true); + while (rc > 0 && ++i < LFSCK_STF_COUNT) + rc = lfsck_namespace_double_scan_one_trace_file(env, com, + com->lc_sub_trace_objs[i].lsto_obj, false); + + CDEBUG(D_LFSCK, "%s: namespace LFSCK phase2 scan stop at the No. %d " + "trace file: rc = %d\n", lfsck_lfsck2name(lfsck), i, rc); + + RETURN(rc); +} + static void lfsck_namespace_assistant_fill_pos(const struct lu_env *env, struct lfsck_component *com, struct lfsck_position *pos) @@ -4597,15 +6266,19 @@ static void lfsck_namespace_assistant_fill_pos(const struct lu_env *env, struct lfsck_assistant_data *lad = com->lc_data; struct lfsck_namespace_req *lnr; + if (((struct lfsck_namespace *)(com->lc_file_ram))->ln_status != + LS_SCANNING_PHASE1) + return; + if (list_empty(&lad->lad_req_list)) return; lnr = list_entry(lad->lad_req_list.next, struct lfsck_namespace_req, lnr_lar.lar_list); - pos->lp_oit_cookie = lnr->lnr_oit_cookie; + pos->lp_oit_cookie = lnr->lnr_lar.lar_parent->lso_oit_cookie; pos->lp_dir_cookie = lnr->lnr_dir_cookie - 1; - pos->lp_dir_parent = *lfsck_dto2fid(lnr->lnr_obj); + pos->lp_dir_parent = lnr->lnr_lar.lar_parent->lso_fid; } static int lfsck_namespace_double_scan_result(const struct lu_env *env, @@ -4617,7 +6290,7 @@ static int lfsck_namespace_double_scan_result(const struct lu_env *env, down_write(&com->lc_sem); ns->ln_run_time_phase2 += cfs_duration_sec(cfs_time_current() + - HALF_SEC - lfsck->li_time_last_checkpoint); + HALF_SEC - com->lc_time_last_checkpoint); ns->ln_time_last_checkpoint = cfs_time_current_sec(); ns->ln_objs_checked_phase2 += com->lc_new_checked; com->lc_new_checked = 0; @@ -4632,24 +6305,112 @@ static int lfsck_namespace_double_scan_result(const struct lu_env *env, ns->ln_time_last_complete = ns->ln_time_last_checkpoint; ns->ln_success_count++; } else if (rc == 0) { - ns->ln_status = lfsck->li_status; - if (ns->ln_status == 0) + if (lfsck->li_status != 0) + ns->ln_status = lfsck->li_status; + else ns->ln_status = LS_STOPPED; } else { ns->ln_status = LS_FAILED; } - rc = lfsck_namespace_store(env, com, false); + rc = lfsck_namespace_store(env, com); up_write(&com->lc_sem); return rc; } +static int +lfsck_namespace_assistant_sync_failures_interpret(const struct lu_env *env, + struct ptlrpc_request *req, + void *args, int rc) +{ + if (rc == 0) { + struct lfsck_async_interpret_args *laia = args; + struct lfsck_tgt_desc *ltd = laia->laia_ltd; + + ltd->ltd_synced_failures = 1; + } + + return 0; +} + +/** + * Notify remote LFSCK instances about former failures. + * + * The local LFSCK instance has recorded which MDTs have ever failed to respond + * some LFSCK verification requests (maybe because of network issues or the MDT + * itself trouble). During the respond gap the MDT may missed some name entries + * verification, then the MDT cannot know whether related MDT-objects have been + * referenced by related name entries or not, then in the second-stage scanning, + * these MDT-objects will be regarded as orphan, if the MDT-object contains bad + * linkEA for back reference, then it will misguide the LFSCK to generate wrong + * name entry for repairing the orphan. + * + * To avoid above trouble, when layout LFSCK finishes the first-stage scanning, + * it will scan the bitmap for the ever failed MDTs, and notify them that they + * have ever missed some name entries verification and should skip the handling + * for orphan MDT-objects. + * + * \param[in] env pointer to the thread context + * \param[in] com pointer to the lfsck component + * \param[in] lr pointer to the lfsck request + */ static void lfsck_namespace_assistant_sync_failures(const struct lu_env *env, struct lfsck_component *com, struct lfsck_request *lr) { - /* XXX: TBD */ + struct lfsck_async_interpret_args *laia = + &lfsck_env_info(env)->lti_laia2; + struct lfsck_assistant_data *lad = com->lc_data; + struct lfsck_namespace *ns = com->lc_file_ram; + struct lfsck_instance *lfsck = com->lc_lfsck; + struct lfsck_tgt_descs *ltds = &lfsck->li_mdt_descs; + struct lfsck_tgt_desc *ltd; + struct ptlrpc_request_set *set; + __u32 idx; + int rc = 0; + ENTRY; + + if (!lad->lad_incomplete) + RETURN_EXIT; + + set = ptlrpc_prep_set(); + if (set == NULL) + GOTO(out, rc = -ENOMEM); + + lr->lr_flags2 = ns->ln_flags | LF_INCOMPLETE; + memset(laia, 0, sizeof(*laia)); + lad->lad_touch_gen++; + + down_read(<ds->ltd_rw_sem); + cfs_foreach_bit(lad->lad_bitmap, idx) { + ltd = lfsck_ltd2tgt(ltds, idx); + LASSERT(ltd != NULL); + + laia->laia_ltd = ltd; + rc = lfsck_async_request(env, ltd->ltd_exp, lr, set, + lfsck_namespace_assistant_sync_failures_interpret, + laia, LFSCK_NOTIFY); + if (rc != 0) + CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail " + "to sync failure with MDT %x: rc = %d\n", + lfsck_lfsck2name(lfsck), ltd->ltd_index, rc); + } + up_read(<ds->ltd_rw_sem); + + rc = ptlrpc_set_wait(set); + ptlrpc_set_destroy(set); + + GOTO(out, rc); + +out: + if (rc != 0) + CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail " + "to sync failure with MDTs, and related MDTs " + "may handle orphan improperly: rc = %d\n", + lfsck_lfsck2name(lfsck), rc); + + EXIT; } struct lfsck_assistant_operations lfsck_namespace_assistant_ops = { @@ -4667,7 +6428,6 @@ struct lfsck_assistant_operations lfsck_namespace_assistant_ops = { * entries, then re-generate the linkEA with the given information. * * \param[in] env pointer to the thread context - * \param[in] dev pointer to the dt_device * \param[in] obj pointer to the dt_object to be handled * \param[in] cname the name for the child in the parent directory * \param[in] pfid the parent directory's FID for the linkEA @@ -4675,11 +6435,11 @@ struct lfsck_assistant_operations lfsck_namespace_assistant_ops = { * \retval 0 for success * \retval negative error number on failure */ -int lfsck_verify_linkea(const struct lu_env *env, struct dt_device *dev, - struct dt_object *obj, const struct lu_name *cname, - const struct lu_fid *pfid) +int lfsck_verify_linkea(const struct lu_env *env, struct dt_object *obj, + const struct lu_name *cname, const struct lu_fid *pfid) { - struct linkea_data ldata = { 0 }; + struct dt_device *dev = lfsck_obj2dev(obj); + struct linkea_data ldata = { NULL }; struct lu_buf linkea_buf; struct thandle *th; int rc; @@ -4731,7 +6491,7 @@ int lfsck_verify_linkea(const struct lu_env *env, struct dt_device *dev, dt_write_lock(env, obj, 0); rc = dt_xattr_set(env, obj, &linkea_buf, - XATTR_NAME_LINK, fl, th, BYPASS_CAPA); + XATTR_NAME_LINK, fl, th); dt_write_unlock(env, obj); GOTO(stop, rc); @@ -4758,7 +6518,7 @@ int lfsck_links_get_first(const struct lu_env *env, struct dt_object *obj, char *name, struct lu_fid *pfid) { struct lu_name *cname = &lfsck_env_info(env)->lti_name; - struct linkea_data ldata = { 0 }; + struct linkea_data ldata = { NULL }; int rc; rc = lfsck_links_read(env, obj, &ldata); @@ -4778,90 +6538,14 @@ int lfsck_links_get_first(const struct lu_env *env, struct dt_object *obj, } /** - * Remove the name entry from the parent directory. - * - * No need to care about the object referenced by the name entry, - * either the name entry is invalid or redundant, or the referenced - * object has been processed has been or will be handled by others. - * - * \param[in] env pointer to the thread context - * \param[in] lfsck pointer to the lfsck instance - * \param[in] parent pointer to the lost+found object - * \param[in] name the name for the name entry to be removed - * \param[in] type the type for the name entry to be removed - * - * \retval 0 for success - * \retval negative error number on failure - */ -int lfsck_remove_name_entry(const struct lu_env *env, - struct lfsck_instance *lfsck, - struct dt_object *parent, - const char *name, __u32 type) -{ - struct dt_device *dev = lfsck->li_next; - struct thandle *th; - struct lustre_handle lh = { 0 }; - int rc; - ENTRY; - - rc = lfsck_ibits_lock(env, lfsck, parent, &lh, - MDS_INODELOCK_UPDATE, LCK_EX); - if (rc != 0) - RETURN(rc); - - th = dt_trans_create(env, dev); - if (IS_ERR(th)) - GOTO(unlock, rc = PTR_ERR(th)); - - rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th); - if (rc != 0) - GOTO(stop, rc); - - if (S_ISDIR(type)) { - rc = dt_declare_ref_del(env, parent, th); - if (rc != 0) - GOTO(stop, rc); - } - - rc = dt_trans_start(env, dev, th); - if (rc != 0) - GOTO(stop, rc); - - rc = dt_delete(env, parent, (const struct dt_key *)name, th, - BYPASS_CAPA); - if (rc != 0) - GOTO(stop, rc); - - if (S_ISDIR(type)) { - dt_write_lock(env, parent, 0); - rc = dt_ref_del(env, parent, th); - dt_write_unlock(env, parent); - } - - GOTO(stop, rc); - -stop: - dt_trans_stop(env, dev, th); - -unlock: - lfsck_ibits_unlock(&lh, LCK_EX); - - CDEBUG(D_LFSCK, "%s: remove name entry "DFID"/%s " - "with type %o: rc = %d\n", lfsck_lfsck2name(lfsck), - PFID(lfsck_dto2fid(parent)), name, type, rc); - - return rc; -} - -/** * Update the object's name entry with the given FID. * * \param[in] env pointer to the thread context * \param[in] lfsck pointer to the lfsck instance - * \param[in] parent pointer to the parent directory that holds + * \param[in] dir pointer to the directory that holds * the name entry * \param[in] name the name for the entry to be updated - * \param[in] pfid the new PFID for the name entry + * \param[in] fid the new FID for the name entry referenced * \param[in] type the type for the name entry to be updated * * \retval 0 for success @@ -4869,19 +6553,20 @@ unlock: */ int lfsck_update_name_entry(const struct lu_env *env, struct lfsck_instance *lfsck, - struct dt_object *parent, const char *name, - const struct lu_fid *pfid, __u32 type) + struct dt_object *dir, const char *name, + const struct lu_fid *fid, __u32 type) { - struct dt_insert_rec *rec = &lfsck_env_info(env)->lti_dt_rec; - struct dt_device *dev = lfsck->li_next; - struct lustre_handle lh = { 0 }; - struct thandle *th; - int rc; - bool exists = true; + struct lfsck_thread_info *info = lfsck_env_info(env); + struct dt_insert_rec *rec = &info->lti_dt_rec; + struct lfsck_lock_handle *llh = &info->lti_llh; + struct dt_device *dev = lfsck_obj2dev(dir); + struct thandle *th; + int rc; + bool exists = true; ENTRY; - rc = lfsck_ibits_lock(env, lfsck, parent, &lh, - MDS_INODELOCK_UPDATE, LCK_EX); + rc = lfsck_lock(env, lfsck, dir, name, llh, + MDS_INODELOCK_UPDATE, LCK_PW); if (rc != 0) RETURN(rc); @@ -4889,27 +6574,26 @@ int lfsck_update_name_entry(const struct lu_env *env, if (IS_ERR(th)) GOTO(unlock, rc = PTR_ERR(th)); - rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th); + rc = dt_declare_delete(env, dir, (const struct dt_key *)name, th); if (rc != 0) GOTO(stop, rc); rec->rec_type = type; - rec->rec_fid = pfid; - rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec, + rec->rec_fid = fid; + rc = dt_declare_insert(env, dir, (const struct dt_rec *)rec, (const struct dt_key *)name, th); if (rc != 0) GOTO(stop, rc); - rc = dt_declare_ref_add(env, parent, th); + rc = dt_declare_ref_add(env, dir, th); if (rc != 0) GOTO(stop, rc); - rc = dt_trans_start(env, dev, th); + rc = dt_trans_start_local(env, dev, th); if (rc != 0) GOTO(stop, rc); - rc = dt_delete(env, parent, (const struct dt_key *)name, th, - BYPASS_CAPA); + rc = dt_delete(env, dir, (const struct dt_key *)name, th); if (rc == -ENOENT) { exists = false; rc = 0; @@ -4918,12 +6602,12 @@ int lfsck_update_name_entry(const struct lu_env *env, if (rc != 0) GOTO(stop, rc); - rc = dt_insert(env, parent, (const struct dt_rec *)rec, - (const struct dt_key *)name, th, BYPASS_CAPA, 1); + rc = dt_insert(env, dir, (const struct dt_rec *)rec, + (const struct dt_key *)name, th, 1); if (rc == 0 && S_ISDIR(type) && !exists) { - dt_write_lock(env, parent, 0); - rc = dt_ref_add(env, parent, th); - dt_write_unlock(env, parent); + dt_write_lock(env, dir, 0); + rc = dt_ref_add(env, dir, th); + dt_write_unlock(env, dir); } GOTO(stop, rc); @@ -4932,11 +6616,10 @@ stop: dt_trans_stop(env, dev, th); unlock: - lfsck_ibits_unlock(&lh, LCK_EX); - + lfsck_unlock(llh); CDEBUG(D_LFSCK, "%s: update name entry "DFID"/%s with the FID "DFID " and the type %o: rc = %d\n", lfsck_lfsck2name(lfsck), - PFID(lfsck_dto2fid(parent)), name, PFID(pfid), type, rc); + PFID(lfsck_dto2fid(dir)), name, PFID(fid), type, rc); return rc; } @@ -4948,6 +6631,7 @@ int lfsck_namespace_setup(const struct lu_env *env, struct lfsck_namespace *ns; struct dt_object *root = NULL; struct dt_object *obj; + int i; int rc; ENTRY; @@ -4966,7 +6650,7 @@ int lfsck_namespace_setup(const struct lu_env *env, com->lc_ops = &lfsck_namespace_ops; com->lc_data = lfsck_assistant_data_init( &lfsck_namespace_assistant_ops, - "lfsck_namespace"); + LFSCK_NAMESPACE); if (com->lc_data == NULL) GOTO(out, rc = -ENOMEM); @@ -4979,6 +6663,9 @@ int lfsck_namespace_setup(const struct lu_env *env, if (com->lc_file_disk == NULL) GOTO(out, rc = -ENOMEM); + for (i = 0; i < LFSCK_STF_COUNT; i++) + mutex_init(&com->lc_sub_trace_objs[i].lsto_mutex); + root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid); if (IS_ERR(root)) GOTO(out, rc = PTR_ERR(root)); @@ -4987,22 +6674,20 @@ int lfsck_namespace_setup(const struct lu_env *env, GOTO(out, rc = -ENOTDIR); obj = local_index_find_or_create(env, lfsck->li_los, root, - lfsck_namespace_name, + LFSCK_NAMESPACE, S_IFREG | S_IRUGO | S_IWUSR, &dt_lfsck_features); if (IS_ERR(obj)) GOTO(out, rc = PTR_ERR(obj)); com->lc_obj = obj; - rc = obj->do_ops->do_index_try(env, obj, &dt_lfsck_features); - if (rc != 0) - GOTO(out, rc); - rc = lfsck_namespace_load(env, com); - if (rc > 0) - rc = lfsck_namespace_reset(env, com, true); - else if (rc == -ENODATA) + if (rc == -ENODATA) rc = lfsck_namespace_init(env, com); + else if (rc < 0) + rc = lfsck_namespace_reset(env, com, true); + else + rc = lfsck_namespace_load_sub_trace_files(env, com, false); if (rc != 0) GOTO(out, rc); @@ -5040,7 +6725,7 @@ int lfsck_namespace_setup(const struct lu_env *env, out: if (root != NULL && !IS_ERR(root)) - lu_object_put(env, &root->do_lu); + lfsck_object_put(env, root); if (rc != 0) { lfsck_component_cleanup(env, com); CERROR("%s: fail to init namespace LFSCK component: rc = %d\n",