X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Flfsck%2Flfsck_lib.c;h=178aa3ad98dfdb0ff44c26ac61f79cf5b7e315ba;hp=db73616915f0a9bd98d70b61660888c503422ac1;hb=8a11cb6282cfbdc8617b809344e6a11223e86a38;hpb=d46286bdfb704cc56972863cb067244b8bb6df43 diff --git a/lustre/lfsck/lfsck_lib.c b/lustre/lfsck/lfsck_lib.c index db73616..178aa3a 100644 --- a/lustre/lfsck/lfsck_lib.c +++ b/lustre/lfsck/lfsck_lib.c @@ -20,7 +20,7 @@ * GPL HEADER END */ /* - * Copyright (c) 2012, 2013, Intel Corporation. + * Copyright (c) 2013, 2014, Intel Corporation. */ /* * lustre/lfsck/lfsck_lib.c @@ -30,6 +30,7 @@ #define DEBUG_SUBSYSTEM S_LFSCK +#include #include #include #include @@ -42,6 +43,8 @@ #include "lfsck_internal.h" +#define LFSCK_CHECKPOINT_SKIP 1 + /* define lfsck thread key */ LU_KEY_INIT(lfsck, struct lfsck_thread_info); @@ -51,6 +54,7 @@ static void lfsck_key_fini(const struct lu_context *ctx, struct lfsck_thread_info *info = data; lu_buf_free(&info->lti_linkea_buf); + lu_buf_free(&info->lti_linkea_buf2); lu_buf_free(&info->lti_big_buf); OBD_FREE_PTR(info); } @@ -58,7 +62,7 @@ static void lfsck_key_fini(const struct lu_context *ctx, LU_CONTEXT_KEY_DEFINE(lfsck, LCT_MD_THREAD | LCT_DT_THREAD); LU_KEY_INIT_GENERIC(lfsck); -static CFS_LIST_HEAD(lfsck_instance_list); +static struct list_head lfsck_instance_list; static struct list_head lfsck_ost_orphan_list; static struct list_head lfsck_mdt_orphan_list; static DEFINE_SPINLOCK(lfsck_instance_lock); @@ -95,9 +99,15 @@ const char *lfsck_param_names[] = { "broadcast", "orphan", "create_ostobj", + "create_mdtobj", NULL }; +enum lfsck_verify_lpf_types { + LVLT_BY_BOOKMARK = 0, + LVLT_BY_NAMEENTRY = 1, +}; + const char *lfsck_status2names(enum lfsck_status status) { if (unlikely(status < 0 || status >= LS_MAX)) @@ -139,14 +149,16 @@ static void lfsck_tgt_descs_fini(struct lfsck_tgt_descs *ltds) } cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) { - ltd = LTD_TGT(ltds, idx); + ltd = lfsck_ltd2tgt(ltds, idx); if (likely(ltd != NULL)) { LASSERT(list_empty(<d->ltd_layout_list)); LASSERT(list_empty(<d->ltd_layout_phase_list)); + LASSERT(list_empty(<d->ltd_namespace_list)); + LASSERT(list_empty(<d->ltd_namespace_phase_list)); ltds->ltd_tgtnr--; cfs_bitmap_clear(ltds->ltd_tgts_bitmap, idx); - LTD_TGT(ltds, idx) = NULL; + lfsck_assign_tgt(ltds, NULL, idx); lfsck_tgt_put(ltd); } } @@ -218,7 +230,7 @@ static int __lfsck_add_target(const struct lu_env *env, GOTO(unlock, rc = -ENOMEM); } - LTD_TGT(ltds, index) = ltd; + lfsck_assign_tgt(ltds, ltd, index); cfs_bitmap_set(ltds->ltd_tgts_bitmap, index); ltds->ltd_tgtnr++; @@ -244,11 +256,9 @@ static int lfsck_add_target_from_orphan(const struct lu_env *env, again: spin_lock(&lfsck_instance_lock); list_for_each_entry_safe(ltd, next, head, ltd_orphan_list) { - if (ltd->ltd_key == lfsck->li_bottom) { - list_del_init(<d->ltd_orphan_list); - list_add_tail(<d->ltd_orphan_list, - <ds->ltd_orphan); - } + if (ltd->ltd_key == lfsck->li_bottom) + list_move_tail(<d->ltd_orphan_list, + <ds->ltd_orphan); } spin_unlock(&lfsck_instance_lock); @@ -279,11 +289,12 @@ again: } static inline struct lfsck_component * -__lfsck_component_find(struct lfsck_instance *lfsck, __u16 type, cfs_list_t *list) +__lfsck_component_find(struct lfsck_instance *lfsck, __u16 type, + struct list_head *list) { struct lfsck_component *com; - cfs_list_for_each_entry(com, list, lc_link) { + list_for_each_entry(com, list, lc_link) { if (com->lc_type == type) return com; } @@ -317,10 +328,10 @@ unlock: void lfsck_component_cleanup(const struct lu_env *env, struct lfsck_component *com) { - if (!cfs_list_empty(&com->lc_link)) - cfs_list_del_init(&com->lc_link); - if (!cfs_list_empty(&com->lc_link_dir)) - cfs_list_del_init(&com->lc_link_dir); + if (!list_empty(&com->lc_link)) + list_del_init(&com->lc_link); + if (!list_empty(&com->lc_link_dir)) + list_del_init(&com->lc_link_dir); lfsck_component_put(env, com); } @@ -351,27 +362,297 @@ int lfsck_fid_alloc(const struct lu_env *env, struct lfsck_instance *lfsck, RETURN(rc); } -static const char dot[] = "."; -static const char dotdot[] = ".."; +static int __lfsck_ibits_lock(const struct lu_env *env, + struct lfsck_instance *lfsck, + struct dt_object *obj, + struct ldlm_res_id *resid, + struct lustre_handle *lh, + __u64 bits, ldlm_mode_t mode) +{ + struct lfsck_thread_info *info = lfsck_env_info(env); + ldlm_policy_data_t *policy = &info->lti_policy; + __u64 flags = LDLM_FL_ATOMIC_CB; + int rc; + + LASSERT(lfsck->li_namespace != NULL); + + memset(policy, 0, sizeof(*policy)); + policy->l_inodebits.bits = bits; + if (dt_object_remote(obj)) { + struct ldlm_enqueue_info *einfo = &info->lti_einfo; + + memset(einfo, 0, sizeof(*einfo)); + einfo->ei_type = LDLM_IBITS; + einfo->ei_mode = mode; + einfo->ei_cb_bl = ldlm_blocking_ast; + einfo->ei_cb_cp = ldlm_completion_ast; + einfo->ei_res_id = resid; + + rc = dt_object_lock(env, obj, lh, einfo, policy); + } else { + rc = ldlm_cli_enqueue_local(lfsck->li_namespace, resid, + LDLM_IBITS, policy, mode, + &flags, ldlm_blocking_ast, + ldlm_completion_ast, NULL, NULL, + 0, LVB_T_NONE, NULL, lh); + } + + if (rc == ELDLM_OK) { + rc = 0; + } else { + memset(lh, 0, sizeof(*lh)); + rc = -EIO; + } + + return rc; +} + +/** + * Request the specified ibits lock for the given object. + * + * Before the LFSCK modifying on the namespace visible object, + * it needs to acquire related ibits ldlm lock. + * + * \param[in] env pointer to the thread context + * \param[in] lfsck pointer to the lfsck instance + * \param[in] obj pointer to the dt_object to be locked + * \param[out] lh pointer to the lock handle + * \param[in] bits the bits for the ldlm lock to be acquired + * \param[in] mode the mode for the ldlm lock to be acquired + * + * \retval 0 for success + * \retval negative error number on failure + */ +int lfsck_ibits_lock(const struct lu_env *env, struct lfsck_instance *lfsck, + struct dt_object *obj, struct lustre_handle *lh, + __u64 bits, ldlm_mode_t mode) +{ + struct ldlm_res_id *resid = &lfsck_env_info(env)->lti_resid; + + LASSERT(!lustre_handle_is_used(lh)); + + fid_build_reg_res_name(lfsck_dto2fid(obj), resid); + return __lfsck_ibits_lock(env, lfsck, obj, resid, lh, bits, mode); +} + +/** + * Release the the specified ibits lock. + * + * If the lock has been acquired before, release it + * and cleanup the handle. Otherwise, do nothing. + * + * \param[in] lh pointer to the lock handle + * \param[in] mode the mode for the ldlm lock to be released + */ +void lfsck_ibits_unlock(struct lustre_handle *lh, ldlm_mode_t mode) +{ + if (lustre_handle_is_used(lh)) { + ldlm_lock_decref(lh, mode); + memset(lh, 0, sizeof(*lh)); + } +} + +/** + * Request compound ibits locks for the given pairs. + * + * Before the LFSCK modifying on the namespace visible object, it needs to + * acquire related ibits ldlm lock. Usually, we can use lfsck_ibits_lock for + * the lock purpose. But the simple lfsck_ibits_lock for directory-based + * modificationis (such as insert name entry to the directory) may be too + * coarse-grained and not efficient. + * + * The lfsck_lock() will request compound ibits locks on the specified + * pairs: the PDO (Parallel Directory Operations) ibits (UPDATE) + * lock on the directory object, and the regular ibits lock on the name hash. + * + * \param[in] env pointer to the thread context + * \param[in] lfsck pointer to the lfsck instance + * \param[in] obj pointer to the dt_object to be locked + * \param[in] name used for building the PDO lock resource + * \param[out] llh pointer to the lfsck_lock_handle + * \param[in] bits the bits for the ldlm lock to be acquired + * \param[in] mode the mode for the ldlm lock to be acquired + * + * \retval 0 for success + * \retval negative error number on failure + */ +int lfsck_lock(const struct lu_env *env, struct lfsck_instance *lfsck, + struct dt_object *obj, const char *name, + struct lfsck_lock_handle *llh, __u64 bits, ldlm_mode_t mode) +{ + struct ldlm_res_id *resid = &lfsck_env_info(env)->lti_resid; + int rc; + + LASSERT(S_ISDIR(lfsck_object_type(obj))); + LASSERT(name != NULL); + LASSERT(!lustre_handle_is_used(&llh->llh_pdo_lh)); + LASSERT(!lustre_handle_is_used(&llh->llh_reg_lh)); + + switch (mode) { + case LCK_EX: + llh->llh_pdo_mode = LCK_EX; + break; + case LCK_PW: + llh->llh_pdo_mode = LCK_CW; + break; + case LCK_PR: + llh->llh_pdo_mode = LCK_CR; + break; + default: + CDEBUG(D_LFSCK, "%s: unexpected PDO lock mode %u on the obj " + DFID"\n", lfsck_lfsck2name(lfsck), mode, + PFID(lfsck_dto2fid(obj))); + LBUG(); + } + + fid_build_reg_res_name(lfsck_dto2fid(obj), resid); + rc = __lfsck_ibits_lock(env, lfsck, obj, resid, &llh->llh_pdo_lh, + MDS_INODELOCK_UPDATE, llh->llh_pdo_mode); + if (rc != 0) + return rc; + + llh->llh_reg_mode = mode; + resid->name[LUSTRE_RES_ID_HSH_OFF] = full_name_hash(name, strlen(name)); + rc = __lfsck_ibits_lock(env, lfsck, obj, resid, &llh->llh_reg_lh, + bits, llh->llh_reg_mode); + if (rc != 0) + lfsck_ibits_unlock(&llh->llh_pdo_lh, llh->llh_pdo_mode); + + return rc; +} + +/** + * Release the the compound ibits locks. + * + * \param[in] llh pointer to the lfsck_lock_handle to be released + */ +void lfsck_unlock(struct lfsck_lock_handle *llh) +{ + lfsck_ibits_unlock(&llh->llh_reg_lh, llh->llh_reg_mode); + lfsck_ibits_unlock(&llh->llh_pdo_lh, llh->llh_pdo_mode); +} + +int lfsck_find_mdt_idx_by_fid(const struct lu_env *env, + struct lfsck_instance *lfsck, + const struct lu_fid *fid) +{ + struct seq_server_site *ss = lfsck_dev_site(lfsck); + struct lu_seq_range *range = &lfsck_env_info(env)->lti_range; + int rc; + + fld_range_set_mdt(range); + rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range); + if (rc == 0) + rc = range->lsr_index; + + return rc; +} + +const char dot[] = "."; +const char dotdot[] = ".."; +static const char dotlustre[] = ".lustre"; +static const char lostfound[] = "lost+found"; + +/** + * Remove the name entry from the .lustre/lost+found 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 or will be handled by others. + * + * \param[in] env pointer to the thread context + * \param[in] lfsck pointer to the lfsck instance + * \param[in] name the name for the name entry to be removed + * + * \retval 0 for success + * \retval negative error number on failure + */ +static int lfsck_lpf_remove_name_entry(const struct lu_env *env, + struct lfsck_instance *lfsck, + const char *name) +{ + struct dt_object *parent = lfsck->li_lpf_root_obj; + struct dt_device *dev = lfsck_obj2dev(parent); + struct thandle *th; + struct lfsck_lock_handle *llh = &lfsck_env_info(env)->lti_llh; + int rc; + ENTRY; + + rc = lfsck_lock(env, lfsck, parent, name, llh, + MDS_INODELOCK_UPDATE, LCK_PW); + 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); + + rc = dt_declare_ref_del(env, parent, th); + if (rc != 0) + GOTO(stop, rc); + + 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); + if (rc != 0) + GOTO(stop, rc); + + 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_unlock(llh); + + CDEBUG(D_LFSCK, "%s: remove name entry "DFID"/%s: rc = %d\n", + lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(parent)), name, rc); + + return rc; +} static int lfsck_create_lpf_local(const struct lu_env *env, struct lfsck_instance *lfsck, - struct dt_object *parent, struct dt_object *child, struct lu_attr *la, struct dt_object_format *dof, const char *name) { - struct dt_device *dev = lfsck->li_bottom; + struct dt_insert_rec *rec = &lfsck_env_info(env)->lti_dt_rec; + struct dt_object *parent = lfsck->li_lpf_root_obj; + struct dt_device *dev = lfsck_obj2dev(child); struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram; struct dt_object *bk_obj = lfsck->li_bookmark_obj; - const struct lu_fid *cfid = lu_object_fid(&child->do_lu); + const struct lu_fid *cfid = lfsck_dto2fid(child); struct thandle *th = NULL; + struct linkea_data ldata = { NULL }; + struct lu_buf linkea_buf; + const struct lu_name *cname; loff_t pos = 0; int len = sizeof(struct lfsck_bookmark); - int rc = 0; + int rc; ENTRY; + rc = linkea_data_new(&ldata, + &lfsck_env_info(env)->lti_linkea_buf2); + if (rc != 0) + RETURN(rc); + + cname = lfsck_name_get_const(env, name, strlen(name)); + rc = linkea_add_buf(&ldata, cname, lfsck_dto2fid(parent)); + if (rc != 0) + RETURN(rc); + th = dt_trans_create(env, dev); if (IS_ERR(th)) RETURN(PTR_ERR(th)); @@ -386,18 +667,28 @@ static int lfsck_create_lpf_local(const struct lu_env *env, if (rc != 0) GOTO(stop, rc); - /* 3a. insert name into parent dir */ - rc = dt_declare_insert(env, parent, (const struct dt_rec *)cfid, + /* 3a. 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, + XATTR_NAME_LINK, 0, th); + if (rc != 0) + GOTO(stop, rc); + + /* 4a. insert name into parent dir */ + rec->rec_type = S_IFDIR; + rec->rec_fid = cfid; + rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec, (const struct dt_key *)name, th); if (rc != 0) GOTO(stop, rc); - /* 4a. increase parent nlink */ + /* 5a. increase parent nlink */ rc = dt_declare_ref_add(env, parent, th); if (rc != 0) GOTO(stop, rc); - /* 5a. update bookmark */ + /* 6a. update bookmark */ rc = dt_declare_record_write(env, bk_obj, lfsck_buf_get(env, bk, len), 0, th); if (rc != 0) @@ -408,7 +699,7 @@ static int lfsck_create_lpf_local(const struct lu_env *env, GOTO(stop, rc); dt_write_lock(env, child, 0); - /* 1b.1 create child */ + /* 1b.1. create child */ rc = dt_create(env, child, la, NULL, dof, th); if (rc != 0) GOTO(unlock, rc); @@ -416,32 +707,41 @@ static int lfsck_create_lpf_local(const struct lu_env *env, if (unlikely(!dt_try_as_dir(env, child))) GOTO(unlock, rc = -ENOTDIR); - /* 1b.2 insert dot into child dir */ - rc = dt_insert(env, child, (const struct dt_rec *)cfid, - (const struct dt_key *)dot, th, BYPASS_CAPA, 1); + /* 1b.2. insert dot into child dir */ + rec->rec_fid = cfid; + rc = dt_insert(env, child, (const struct dt_rec *)rec, + (const struct dt_key *)dot, th, 1); if (rc != 0) GOTO(unlock, rc); - /* 1b.3 insert dotdot into child dir */ - rc = dt_insert(env, child, (const struct dt_rec *)&LU_LPF_FID, - (const struct dt_key *)dotdot, th, BYPASS_CAPA, 1); + /* 1b.3. insert dotdot into child dir */ + rec->rec_fid = &LU_LPF_FID; + rc = dt_insert(env, child, (const struct dt_rec *)rec, + (const struct dt_key *)dotdot, th, 1); if (rc != 0) GOTO(unlock, rc); /* 2b. increase child nlink */ rc = dt_ref_add(env, child, th); + if (rc != 0) + GOTO(unlock, rc); + + /* 3b. insert linkEA for child. */ + rc = dt_xattr_set(env, child, &linkea_buf, + XATTR_NAME_LINK, 0, th); dt_write_unlock(env, child); if (rc != 0) GOTO(stop, rc); - /* 3b. insert name into parent dir */ - rc = dt_insert(env, parent, (const struct dt_rec *)cfid, - (const struct dt_key *)name, th, BYPASS_CAPA, 1); + /* 4b. insert name into parent dir */ + rec->rec_fid = cfid; + rc = dt_insert(env, parent, (const struct dt_rec *)rec, + (const struct dt_key *)name, th, 1); if (rc != 0) GOTO(stop, rc); dt_write_lock(env, parent, 0); - /* 4b. increase parent nlink */ + /* 5b. increase parent nlink */ rc = dt_ref_add(env, parent, th); dt_write_unlock(env, parent); if (rc != 0) @@ -450,7 +750,7 @@ static int lfsck_create_lpf_local(const struct lu_env *env, bk->lb_lpf_fid = *cfid; lfsck_bookmark_cpu_to_le(&lfsck->li_bookmark_disk, bk); - /* 5b. update bookmark */ + /* 6b. update bookmark */ rc = dt_record_write(env, bk_obj, lfsck_buf_get(env, bk, len), &pos, th); @@ -467,22 +767,36 @@ stop: static int lfsck_create_lpf_remote(const struct lu_env *env, struct lfsck_instance *lfsck, - struct dt_object *parent, struct dt_object *child, struct lu_attr *la, struct dt_object_format *dof, const char *name) { + struct dt_insert_rec *rec = &lfsck_env_info(env)->lti_dt_rec; + struct dt_object *parent = lfsck->li_lpf_root_obj; struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram; struct dt_object *bk_obj = lfsck->li_bookmark_obj; - const struct lu_fid *cfid = lu_object_fid(&child->do_lu); + const struct lu_fid *cfid = lfsck_dto2fid(child); struct thandle *th = NULL; + struct linkea_data ldata = { NULL }; + struct lu_buf linkea_buf; + const struct lu_name *cname; struct dt_device *dev; loff_t pos = 0; int len = sizeof(struct lfsck_bookmark); - int rc = 0; + int rc; ENTRY; + rc = linkea_data_new(&ldata, + &lfsck_env_info(env)->lti_linkea_buf2); + if (rc != 0) + RETURN(rc); + + cname = lfsck_name_get_const(env, name, strlen(name)); + rc = linkea_add_buf(&ldata, cname, lfsck_dto2fid(parent)); + if (rc != 0) + RETURN(rc); + /* Create .lustre/lost+found/MDTxxxx. */ /* XXX: Currently, cross-MDT create operation needs to create the child @@ -495,37 +809,51 @@ static int lfsck_create_lpf_remote(const struct lu_env *env, * To avoid more inconsistency, we split the create operation into * two transactions: * - * 1) create the child locally. + * 1) create the child and update the lfsck_bookmark::lb_lpf_fid + * locally. * 2) insert the name "MDTXXXX" in the parent ".lustre/lost+found" - * remotely and update the lfsck_bookmark::lb_lpf_fid locally. + * remotely. * - * If 1) done but 2) failed, then the worst case is that we lose - * one object locally, which is not a big issue. (can be repaird - * by LFSCK phase III) */ + * If 1) done, but 2) failed, then go ahead, the LFSCK will try to + * repair such inconsistency when LFSCK run next time. */ - /* Transaction I: */ + /* Transaction I: locally */ - dev = lfsck->li_bottom; + dev = lfsck_obj2dev(child); th = dt_trans_create(env, dev); if (IS_ERR(th)) RETURN(PTR_ERR(th)); - /* 1a. create child locally. */ + /* 1a. create child */ rc = dt_declare_create(env, child, la, NULL, dof, th); if (rc != 0) GOTO(stop, rc); - /* 2a. increase child nlink locally. */ + /* 2a. increase child nlink */ rc = dt_declare_ref_add(env, child, th); if (rc != 0) GOTO(stop, rc); + /* 3a. 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, + XATTR_NAME_LINK, 0, th); + if (rc != 0) + GOTO(stop, rc); + + /* 4a. update bookmark */ + rc = dt_declare_record_write(env, bk_obj, + lfsck_buf_get(env, bk, len), 0, 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, child, 0); - /* 1b. create child locally. */ + /* 1b.1. create child */ rc = dt_create(env, child, la, NULL, dof, th); if (rc != 0) GOTO(unlock, rc); @@ -533,95 +861,78 @@ static int lfsck_create_lpf_remote(const struct lu_env *env, if (unlikely(!dt_try_as_dir(env, child))) GOTO(unlock, rc = -ENOTDIR); - /* 2b.1 insert dot into child dir locally. */ - rc = dt_insert(env, child, (const struct dt_rec *)cfid, - (const struct dt_key *)dot, th, BYPASS_CAPA, 1); + /* 1b.2. insert dot into child dir */ + 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, 1); if (rc != 0) GOTO(unlock, rc); - /* 2b.2 insert dotdot into child dir locally. */ - rc = dt_insert(env, child, (const struct dt_rec *)&LU_LPF_FID, - (const struct dt_key *)dotdot, th, BYPASS_CAPA, 1); + /* 1b.3. insert dotdot into child dir */ + rec->rec_fid = &LU_LPF_FID; + rc = dt_insert(env, child, (const struct dt_rec *)rec, + (const struct dt_key *)dotdot, th, 1); if (rc != 0) GOTO(unlock, rc); - /* 2b.3 increase child nlink locally. */ + /* 2b. increase child nlink */ rc = dt_ref_add(env, child, th); + if (rc != 0) + GOTO(unlock, rc); + + /* 3b. insert linkEA for child */ + rc = dt_xattr_set(env, child, &linkea_buf, + XATTR_NAME_LINK, 0, th); + if (rc != 0) + GOTO(unlock, rc); + + bk->lb_lpf_fid = *cfid; + lfsck_bookmark_cpu_to_le(&lfsck->li_bookmark_disk, bk); + + /* 4b. update bookmark */ + rc = dt_record_write(env, bk_obj, + lfsck_buf_get(env, bk, len), &pos, th); + dt_write_unlock(env, child); dt_trans_stop(env, dev, th); if (rc != 0) RETURN(rc); - /* Transaction II: */ + /* Transaction II: remotely */ - dev = lfsck->li_next; + dev = lfsck_obj2dev(parent); th = dt_trans_create(env, dev); if (IS_ERR(th)) RETURN(PTR_ERR(th)); - /* 3a. insert name into parent dir remotely. */ - rc = dt_declare_insert(env, parent, (const struct dt_rec *)cfid, + th->th_sync = 1; + /* 5a. insert name into parent dir */ + rec->rec_fid = cfid; + rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec, (const struct dt_key *)name, th); if (rc != 0) GOTO(stop, rc); - /* 4a. increase parent nlink remotely. */ + /* 6a. increase parent nlink */ rc = dt_declare_ref_add(env, parent, th); if (rc != 0) GOTO(stop, rc); - /* 5a. decrease child nlink for dotdot locally if former remote - * update failed. */ - rc = dt_declare_ref_del(env, child, th); - if (rc != 0) - GOTO(stop, rc); - - /* 6a. decrease child nlink for dot locally if former remote - * update failed. */ - rc = dt_declare_ref_del(env, child, th); - if (rc != 0) - GOTO(stop, rc); - - /* 7a. destroy child locally if former remote update failed. */ - rc = dt_declare_destroy(env, child, th); - if (rc != 0) - GOTO(stop, rc); - - /* 8a. update bookmark locally. */ - rc = dt_declare_record_write(env, bk_obj, - lfsck_buf_get(env, bk, len), 0, th); + rc = dt_trans_start_local(env, dev, th); if (rc != 0) GOTO(stop, rc); - rc = dt_trans_start(env, dev, th); + /* 5b. insert name into parent dir */ + rc = dt_insert(env, parent, (const struct dt_rec *)rec, + (const struct dt_key *)name, th, 1); if (rc != 0) GOTO(stop, rc); - /* 3b. insert name into parent dir remotely. */ - rc = dt_insert(env, parent, (const struct dt_rec *)cfid, - (const struct dt_key *)name, th, BYPASS_CAPA, 1); - if (rc == 0) { - dt_write_lock(env, parent, 0); - /* 4b. increase parent nlink remotely. */ - rc = dt_ref_add(env, parent, th); - dt_write_unlock(env, parent); - } - if (rc != 0) { - /* 5b. decrease child nlink for dotdot locally. */ - dt_ref_del(env, child, th); - /* 6b. decrease child nlink for dot locally. */ - dt_ref_del(env, child, th); - /* 7b. destroy child locally. */ - dt_destroy(env, child, th); - GOTO(stop, rc); - } - - bk->lb_lpf_fid = *cfid; - lfsck_bookmark_cpu_to_le(&lfsck->li_bookmark_disk, bk); - - /* 8b. update bookmark locally. */ - rc = dt_record_write(env, bk_obj, - lfsck_buf_get(env, bk, len), &pos, th); + dt_write_lock(env, parent, 0); + /* 6b. increase parent nlink */ + rc = dt_ref_add(env, parent, th); + dt_write_unlock(env, parent); GOTO(stop, rc); @@ -630,60 +941,63 @@ unlock: stop: dt_trans_stop(env, dev, th); + if (rc != 0 && dev == lfsck_obj2dev(parent)) + CDEBUG(D_LFSCK, "%s: partially created the object "DFID + "for orphans, but failed to insert the name %s " + "to the .lustre/lost+found/. Such inconsistency " + "will be repaired when LFSCK run next time: rc = %d\n", + lfsck_lfsck2name(lfsck), PFID(cfid), name, rc); + return rc; } -/* Do NOT create .lustre/lost+found/MDTxxxx when register the lfsck instance, - * because the MDT0 maybe not reaady for sequence allocation yet. We do that - * only when it is required, such as orphan OST-objects repairing. */ -int lfsck_create_lpf(const struct lu_env *env, struct lfsck_instance *lfsck) +/** + * Create the MDTxxxx directory under /ROOT/.lustre/lost+found/ + * + * The /ROOT/.lustre/lost+found/MDTxxxx/ directory is used for holding + * orphans and other uncertain inconsistent objects found during the + * LFSCK. Such directory will be created by the LFSCK engine on the + * local MDT before the LFSCK scanning. + * + * \param[in] env pointer to the thread context + * \param[in] lfsck pointer to the lfsck instance + * + * \retval 0 for success + * \retval negative error number on failure + */ +static int lfsck_create_lpf(const struct lu_env *env, + struct lfsck_instance *lfsck) { struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram; struct lfsck_thread_info *info = lfsck_env_info(env); struct lu_fid *cfid = &info->lti_fid2; struct lu_attr *la = &info->lti_la; struct dt_object_format *dof = &info->lti_dof; - struct dt_object *parent = NULL; + struct dt_object *parent = lfsck->li_lpf_root_obj; struct dt_object *child = NULL; + struct lfsck_lock_handle *llh = &info->lti_llh; char name[8]; - int node = lfsck_dev_idx(lfsck->li_bottom); + int node = lfsck_dev_idx(lfsck); int rc = 0; ENTRY; LASSERT(lfsck->li_master); + LASSERT(parent != NULL); + LASSERT(lfsck->li_lpf_obj == NULL); - sprintf(name, "MDT%04x", node); - if (node == 0) { - parent = lfsck_object_find_by_dev(env, lfsck->li_bottom, - &LU_LPF_FID); - } else { - struct lfsck_tgt_desc *ltd; - - ltd = lfsck_tgt_get(&lfsck->li_mdt_descs, 0); - if (unlikely(ltd == NULL)) - RETURN(-ENXIO); - - parent = lfsck_object_find_by_dev(env, ltd->ltd_tgt, - &LU_LPF_FID); - lfsck_tgt_put(ltd); - } - if (IS_ERR(parent)) - RETURN(PTR_ERR(parent)); - - if (unlikely(!dt_try_as_dir(env, parent))) - GOTO(out, rc = -ENOTDIR); - - mutex_lock(&lfsck->li_mutex); - if (lfsck->li_lpf_obj != NULL) - GOTO(unlock, rc = 0); + rc = lfsck_lock(env, lfsck, parent, name, llh, + MDS_INODELOCK_UPDATE, LCK_PW); + if (rc != 0) + RETURN(rc); + snprintf(name, 8, "MDT%04x", node); if (fid_is_zero(&bk->lb_lpf_fid)) { /* There is corner case that: in former LFSCK scanning we have * created the .lustre/lost+found/MDTxxxx but failed to update * the lfsck_bookmark::lb_lpf_fid successfully. So need lookup * it from MDT0 firstly. */ rc = dt_lookup(env, parent, (struct dt_rec *)cfid, - (const struct dt_key *)name, BYPASS_CAPA); + (const struct dt_key *)name); if (rc != 0 && rc != -ENOENT) GOTO(unlock, rc); @@ -699,16 +1013,17 @@ int lfsck_create_lpf(const struct lu_env *env, struct lfsck_instance *lfsck) *cfid = bk->lb_lpf_fid; } - child = lfsck_object_find_by_dev(env, lfsck->li_bottom, cfid); + child = lfsck_object_find_bottom(env, lfsck, cfid); if (IS_ERR(child)) GOTO(unlock, rc = PTR_ERR(child)); if (dt_object_exists(child) != 0) { if (unlikely(!dt_try_as_dir(env, child))) - GOTO(unlock, rc = -ENOTDIR); + rc = -ENOTDIR; + else + lfsck->li_lpf_obj = child; - lfsck->li_lpf_obj = child; - GOTO(unlock, rc = 0); + GOTO(unlock, rc); } memset(la, 0, sizeof(*la)); @@ -720,68 +1035,549 @@ int lfsck_create_lpf(const struct lu_env *env, struct lfsck_instance *lfsck) dof->dof_type = dt_mode_to_dft(S_IFDIR); if (node == 0) - rc = lfsck_create_lpf_local(env, lfsck, parent, child, la, - dof, name); + rc = lfsck_create_lpf_local(env, lfsck, child, la, dof, name); else - rc = lfsck_create_lpf_remote(env, lfsck, parent, child, la, - dof, name); + rc = lfsck_create_lpf_remote(env, lfsck, child, la, dof, name); if (rc == 0) lfsck->li_lpf_obj = child; GOTO(unlock, rc); unlock: - mutex_unlock(&lfsck->li_mutex); + lfsck_unlock(llh); if (rc != 0 && child != NULL && !IS_ERR(child)) - lu_object_put(env, &child->do_lu); -out: - if (parent != NULL && !IS_ERR(parent)) - lu_object_put(env, &parent->do_lu); + lfsck_object_put(env, child); return rc; } -static int lfsck_fid_init(struct lfsck_instance *lfsck) +/** + * Scan .lustre/lost+found for bad name entries and remove them. + * + * The valid name entry should be "MDTxxxx", the "xxxx" is the MDT device + * index in the system. Any other formatted name is invalid and should be + * removed. + * + * \param[in] env pointer to the thread context + * \param[in] lfsck pointer to the lfsck instance + * + * \retval 0 for success + * \retval negative error number on failure + */ +static int lfsck_scan_lpf_bad_entries(const struct lu_env *env, + struct lfsck_instance *lfsck) { - struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram; - struct seq_server_site *ss; - char *prefix; - int rc = 0; + struct dt_object *parent = lfsck->li_lpf_root_obj; + struct lu_dirent *ent = + (struct lu_dirent *)lfsck_env_info(env)->lti_key; + const struct dt_it_ops *iops = &parent->do_index_ops->dio_it; + struct dt_it *it; + int rc; ENTRY; - ss = lu_site2seq(lfsck->li_bottom->dd_lu_dev.ld_site); - if (unlikely(ss == NULL)) - RETURN(-ENXIO); + it = iops->init(env, parent, LUDA_64BITHASH); + if (IS_ERR(it)) + RETURN(PTR_ERR(it)); - OBD_ALLOC_PTR(lfsck->li_seq); - if (lfsck->li_seq == NULL) - RETURN(-ENOMEM); + rc = iops->load(env, it, 0); + if (rc == 0) + rc = iops->next(env, it); + else if (rc > 0) + rc = 0; - OBD_ALLOC(prefix, MAX_OBD_NAME + 7); - if (prefix == NULL) - GOTO(out, rc = -ENOMEM); + while (rc == 0) { + int off = 3; - snprintf(prefix, MAX_OBD_NAME + 7, "lfsck-%s", lfsck_lfsck2name(lfsck)); - rc = seq_client_init(lfsck->li_seq, NULL, LUSTRE_SEQ_METADATA, prefix, - ss->ss_server_seq); - OBD_FREE(prefix, MAX_OBD_NAME + 7); - if (rc != 0) - GOTO(out, rc); + rc = iops->rec(env, it, (struct dt_rec *)ent, LUDA_64BITHASH); + if (rc != 0) + break; - if (fid_is_sane(&bk->lb_last_fid)) - lfsck->li_seq->lcs_fid = bk->lb_last_fid; + ent->lde_namelen = le16_to_cpu(ent->lde_namelen); + if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen)) + goto next; - RETURN(0); + /* name length must be strlen("MDTxxxx") */ + if (ent->lde_namelen != 7) + goto remove; -out: - OBD_FREE_PTR(lfsck->li_seq); - lfsck->li_seq = NULL; + if (memcmp(ent->lde_name, "MDT", off) != 0) + goto remove; - return rc; -} + while (off < 7 && isxdigit(ent->lde_name[off])) + off++; -static void lfsck_fid_fini(struct lfsck_instance *lfsck) -{ + if (off != 7) { + +remove: + rc = lfsck_lpf_remove_name_entry(env, lfsck, + ent->lde_name); + if (rc != 0) + break; + } + +next: + rc = iops->next(env, it); + } + + iops->put(env, it); + iops->fini(env, it); + + RETURN(rc > 0 ? 0 : rc); +} + +static int lfsck_update_lpf_entry(const struct lu_env *env, + struct lfsck_instance *lfsck, + struct dt_object *parent, + struct dt_object *child, + const char *name, + enum lfsck_verify_lpf_types type) +{ + int rc; + + if (type == LVLT_BY_BOOKMARK) { + rc = lfsck_update_name_entry(env, lfsck, parent, name, + lfsck_dto2fid(child), S_IFDIR); + } else /* if (type == LVLT_BY_NAMEENTRY) */ { + lfsck->li_bookmark_ram.lb_lpf_fid = *lfsck_dto2fid(child); + rc = lfsck_bookmark_store(env, lfsck); + + CDEBUG(D_LFSCK, "%s: update LPF fid "DFID + " in the bookmark file: rc = %d\n", + lfsck_lfsck2name(lfsck), + PFID(lfsck_dto2fid(child)), rc); + } + + return rc; +} + +/** + * Check whether the @child back references the @parent. + * + * Two cases: + * 1) The child's FID is stored in the bookmark file. If the child back + * references the parent (LU_LPF_FID object) via its ".." entry, then + * insert the name (MDTxxxx) to the .lustre/lost+found; otherwise, if + * the child back references another parent2, then: + * 1.1) If the parent2 recognizes the child, then update the bookmark file; + * 1.2) Otherwise, the LFSCK cannot know whether there will be parent3 that + * references the child. So keep them there. As the LFSCK processing, + * the parent3 may be found, then when the LFSCK run next time, the + * inconsistency can be repaired. + * + * 2) The child's FID is stored in the .lustre/lost+found/ sub-directory name + * entry (MDTxxxx). If the child back references the parent (LU_LPF_FID obj) + * via its ".." entry, then update the bookmark file, otherwise, if the child + * back references another parent2, then: + * 2.1) If the parent2 recognizes the child, then remove the sub-directory + * from .lustre/lost+found/; + * 2.2) Otherwise, if the parent2 does not recognizes the child, trust the + * sub-directory name entry and update the child; + * 2.3) Otherwise, if we do not know whether the parent2 recognizes the child + * or not, then keep them there. + * + * \param[in] env pointer to the thread context + * \param[in] lfsck pointer to the lfsck instance + * \param[in] child pointer to the lost+found sub-directory object + * \param[in] name the name for lost+found sub-directory object + * \param[out] fid pointer to the buffer to hold the FID of the object + * (called it as parent2) that is referenced via the + * child's dotdot entry; it also can be the FID that + * is referenced by the name entry under the parent2. + * \param[in] type to indicate where the child's FID is stored in + * + * \retval positive number for uncertain inconsistency + * \retval 0 for success + * \retval negative error number on failure + */ +static int lfsck_verify_lpf_pairs(const struct lu_env *env, + struct lfsck_instance *lfsck, + struct dt_object *child, const char *name, + struct lu_fid *fid, + enum lfsck_verify_lpf_types type) +{ + struct dt_object *parent = lfsck->li_lpf_root_obj; + struct lfsck_thread_info *info = lfsck_env_info(env); + char *name2 = info->lti_key; + struct lu_fid *fid2 = &info->lti_fid3; + struct dt_object *parent2 = NULL; + struct lustre_handle lh = { 0 }; + int rc; + ENTRY; + + fid_zero(fid); + rc = dt_lookup(env, child, (struct dt_rec *)fid, + (const struct dt_key *)dotdot); + if (rc != 0) + GOTO(linkea, rc); + + if (!fid_is_sane(fid)) + GOTO(linkea, rc = -EINVAL); + + if (lu_fid_eq(fid, &LU_LPF_FID)) { + const struct lu_name *cname; + + if (lfsck->li_lpf_obj == NULL) { + lu_object_get(&child->do_lu); + lfsck->li_lpf_obj = child; + } + + cname = lfsck_name_get_const(env, name, strlen(name)); + rc = lfsck_verify_linkea(env, child, cname, &LU_LPF_FID); + if (rc == 0) + rc = lfsck_update_lpf_entry(env, lfsck, parent, child, + name, type); + + GOTO(out_done, rc); + } + + parent2 = lfsck_object_find_bottom(env, lfsck, fid); + if (IS_ERR(parent2)) + GOTO(linkea, parent2); + + if (!dt_object_exists(parent2)) { + lfsck_object_put(env, parent2); + + GOTO(linkea, parent2 = ERR_PTR(-ENOENT)); + } + + if (!dt_try_as_dir(env, parent2)) { + lfsck_object_put(env, parent2); + + GOTO(linkea, parent2 = ERR_PTR(-ENOTDIR)); + } + +linkea: + /* To prevent rename/unlink race */ + rc = lfsck_ibits_lock(env, lfsck, child, &lh, + MDS_INODELOCK_UPDATE, LCK_PR); + if (rc != 0) + GOTO(out_put, rc); + + dt_read_lock(env, child, 0); + rc = lfsck_links_get_first(env, child, name2, fid2); + if (rc != 0) { + dt_read_unlock(env, child); + lfsck_ibits_unlock(&lh, LCK_PR); + + GOTO(out_put, rc = 1); + } + + /* It is almost impossible that the bookmark file (or the name entry) + * and the linkEA hit the same data corruption. Trust the linkEA. */ + if (lu_fid_eq(fid2, &LU_LPF_FID) && strcmp(name, name2) == 0) { + dt_read_unlock(env, child); + lfsck_ibits_unlock(&lh, LCK_PR); + + *fid = *fid2; + if (lfsck->li_lpf_obj == NULL) { + lu_object_get(&child->do_lu); + lfsck->li_lpf_obj = child; + } + + /* Update the child's dotdot entry */ + rc = lfsck_update_name_entry(env, lfsck, child, dotdot, + &LU_LPF_FID, S_IFDIR); + if (rc == 0) + rc = lfsck_update_lpf_entry(env, lfsck, parent, child, + name, type); + + GOTO(out_put, rc); + } + + if (parent2 == NULL || IS_ERR(parent2)) { + dt_read_unlock(env, child); + lfsck_ibits_unlock(&lh, LCK_PR); + + GOTO(out_done, rc = 1); + } + + rc = dt_lookup(env, parent2, (struct dt_rec *)fid, + (const struct dt_key *)name2); + dt_read_unlock(env, child); + lfsck_ibits_unlock(&lh, LCK_PR); + if (rc != 0 && rc != -ENOENT) + GOTO(out_put, rc); + + if (rc == -ENOENT || !lu_fid_eq(fid, lfsck_dto2fid(child))) { + if (type == LVLT_BY_BOOKMARK) + GOTO(out_put, rc = 1); + + /* Trust the name entry, update the child's dotdot entry. */ + rc = lfsck_update_name_entry(env, lfsck, child, dotdot, + &LU_LPF_FID, S_IFDIR); + + GOTO(out_put, rc); + } + + if (type == LVLT_BY_BOOKMARK) { + /* Invalid FID record in the bookmark file, reset it. */ + fid_zero(&lfsck->li_bookmark_ram.lb_lpf_fid); + rc = lfsck_bookmark_store(env, lfsck); + + CDEBUG(D_LFSCK, "%s: reset invalid LPF fid "DFID + " in the bookmark file: rc = %d\n", + lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(child)), rc); + } else /* if (type == LVLT_BY_NAMEENTRY) */ { + /* The name entry is wrong, remove it. */ + rc = lfsck_lpf_remove_name_entry(env, lfsck, name); + } + + GOTO(out_put, rc); + +out_put: + if (parent2 != NULL && !IS_ERR(parent2)) + lfsck_object_put(env, parent2); + +out_done: + return rc; +} + +/** + * Verify the /ROOT/.lustre/lost+found/ directory. + * + * /ROOT/.lustre/lost+found/ is a special directory to hold the objects that + * the LFSCK does not exactly know how to handle, such as orphans. So before + * the LFSCK scanning the system, the consistency of such directory needs to + * be verified firstly to allow the users to use it during the LFSCK. + * + * \param[in] env pointer to the thread context + * \param[in] lfsck pointer to the lfsck instance + * + * \retval positive number for uncertain inconsistency + * \retval 0 for success + * \retval negative error number on failure + */ +int lfsck_verify_lpf(const struct lu_env *env, struct lfsck_instance *lfsck) +{ + struct lfsck_thread_info *info = lfsck_env_info(env); + struct lu_fid *pfid = &info->lti_fid; + struct lu_fid *cfid = &info->lti_fid2; + struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram; + struct dt_object *parent; + /* child1's FID is in the bookmark file. */ + struct dt_object *child1 = NULL; + /* child2's FID is in the name entry MDTxxxx. */ + struct dt_object *child2 = NULL; + const struct lu_name *cname; + char name[8]; + int node = lfsck_dev_idx(lfsck); + int rc = 0; + ENTRY; + + LASSERT(lfsck->li_master); + + if (lfsck->li_lpf_root_obj != NULL) + RETURN(0); + + if (node == 0) { + parent = lfsck_object_find_by_dev(env, lfsck->li_bottom, + &LU_LPF_FID); + } else { + struct lfsck_tgt_desc *ltd; + + ltd = lfsck_tgt_get(&lfsck->li_mdt_descs, 0); + if (unlikely(ltd == NULL)) + RETURN(-ENXIO); + + parent = lfsck_object_find_by_dev(env, ltd->ltd_tgt, + &LU_LPF_FID); + lfsck_tgt_put(ltd); + } + + if (IS_ERR(parent)) + RETURN(PTR_ERR(parent)); + + LASSERT(dt_object_exists(parent)); + + if (unlikely(!dt_try_as_dir(env, parent))) { + lfsck_object_put(env, parent); + + GOTO(put, rc = -ENOTDIR); + } + + lfsck->li_lpf_root_obj = parent; + if (node == 0) { + rc = lfsck_scan_lpf_bad_entries(env, lfsck); + if (rc != 0) + CDEBUG(D_LFSCK, "%s: scan .lustre/lost+found/ " + "for bad sub-directories: rc = %d\n", + lfsck_lfsck2name(lfsck), rc); + } + + if (!fid_is_zero(&bk->lb_lpf_fid)) { + if (unlikely(!fid_is_norm(&bk->lb_lpf_fid))) { + struct lu_fid tfid = bk->lb_lpf_fid; + + /* Invalid FID record in the bookmark file, reset it. */ + fid_zero(&bk->lb_lpf_fid); + rc = lfsck_bookmark_store(env, lfsck); + + CDEBUG(D_LFSCK, "%s: reset invalid LPF fid "DFID + " in the bookmark file: rc = %d\n", + lfsck_lfsck2name(lfsck), PFID(&tfid), rc); + + if (rc != 0) + GOTO(put, rc); + } else { + child1 = lfsck_object_find_bottom(env, lfsck, + &bk->lb_lpf_fid); + if (IS_ERR(child1)) { + child1 = NULL; + goto find_child2; + } + + if (unlikely(!dt_object_exists(child1) || + dt_object_remote(child1)) || + !S_ISDIR(lfsck_object_type(child1))) { + /* Invalid FID record in the bookmark file, + * reset it. */ + fid_zero(&bk->lb_lpf_fid); + rc = lfsck_bookmark_store(env, lfsck); + + CDEBUG(D_LFSCK, "%s: reset invalid LPF fid "DFID + " in the bookmark file: rc = %d\n", + lfsck_lfsck2name(lfsck), + PFID(lfsck_dto2fid(child1)), rc); + + if (rc != 0) + GOTO(put, rc); + + lfsck_object_put(env, child1); + child1 = NULL; + } else if (unlikely(!dt_try_as_dir(env, child1))) { + GOTO(put, rc = -ENOTDIR); + } + } + } + +find_child2: + snprintf(name, 8, "MDT%04x", node); + rc = dt_lookup(env, parent, (struct dt_rec *)cfid, + (const struct dt_key *)name); + if (rc == -ENOENT) { + if (!fid_is_zero(&bk->lb_lpf_fid)) + goto check_child1; + + GOTO(put, rc = 0); + } + + if (rc != 0) + GOTO(put, rc); + + /* Invalid FID in the name entry, remove the name entry. */ + if (!fid_is_norm(cfid)) { + rc = lfsck_lpf_remove_name_entry(env, lfsck, name); + if (rc != 0) + GOTO(put, rc); + + goto check_child1; + } + + child2 = lfsck_object_find_bottom(env, lfsck, cfid); + if (IS_ERR(child2)) + GOTO(put, rc = PTR_ERR(child2)); + + if (unlikely(!dt_object_exists(child2) || + dt_object_remote(child2)) || + !S_ISDIR(lfsck_object_type(child2))) { + rc = lfsck_lpf_remove_name_entry(env, lfsck, name); + if (rc != 0) + GOTO(put, rc); + + goto check_child1; + } + + if (unlikely(!dt_try_as_dir(env, child2))) + GOTO(put, rc = -ENOTDIR); + + if (child1 == NULL) { + rc = lfsck_verify_lpf_pairs(env, lfsck, child2, name, + pfid, LVLT_BY_NAMEENTRY); + } else if (!lu_fid_eq(cfid, &bk->lb_lpf_fid)) { + rc = lfsck_verify_lpf_pairs(env, lfsck, child1, name, + pfid, LVLT_BY_BOOKMARK); + if (!lu_fid_eq(pfid, &LU_LPF_FID)) + rc = lfsck_verify_lpf_pairs(env, lfsck, child2, + name, pfid, + LVLT_BY_NAMEENTRY); + } else { + if (lfsck->li_lpf_obj == NULL) { + lu_object_get(&child2->do_lu); + lfsck->li_lpf_obj = child2; + } + + cname = lfsck_name_get_const(env, name, strlen(name)); + rc = lfsck_verify_linkea(env, child2, cname, &LU_LPF_FID); + } + + GOTO(put, rc); + +check_child1: + if (child1 != NULL) + rc = lfsck_verify_lpf_pairs(env, lfsck, child1, name, + pfid, LVLT_BY_BOOKMARK); + + GOTO(put, rc); + +put: + if (lfsck->li_lpf_obj != NULL) { + if (unlikely(!dt_try_as_dir(env, lfsck->li_lpf_obj))) { + lfsck_object_put(env, lfsck->li_lpf_obj); + lfsck->li_lpf_obj = NULL; + rc = -ENOTDIR; + } + } else if (rc == 0) { + rc = lfsck_create_lpf(env, lfsck); + } + + if (child2 != NULL && !IS_ERR(child2)) + lfsck_object_put(env, child2); + if (child1 != NULL && !IS_ERR(child1)) + lfsck_object_put(env, child1); + + return rc; +} + +static int lfsck_fid_init(struct lfsck_instance *lfsck) +{ + struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram; + struct seq_server_site *ss = lfsck_dev_site(lfsck); + char *prefix; + int rc = 0; + ENTRY; + + if (unlikely(ss == NULL)) + RETURN(-ENXIO); + + OBD_ALLOC_PTR(lfsck->li_seq); + if (lfsck->li_seq == NULL) + RETURN(-ENOMEM); + + OBD_ALLOC(prefix, MAX_OBD_NAME + 7); + if (prefix == NULL) + GOTO(out, rc = -ENOMEM); + + snprintf(prefix, MAX_OBD_NAME + 7, "lfsck-%s", lfsck_lfsck2name(lfsck)); + rc = seq_client_init(lfsck->li_seq, NULL, LUSTRE_SEQ_METADATA, prefix, + ss->ss_server_seq); + OBD_FREE(prefix, MAX_OBD_NAME + 7); + if (rc != 0) + GOTO(out, rc); + + if (fid_is_sane(&bk->lb_last_fid)) + lfsck->li_seq->lcs_fid = bk->lb_last_fid; + + RETURN(0); + +out: + OBD_FREE_PTR(lfsck->li_seq); + lfsck->li_seq = NULL; + + return rc; +} + +static void lfsck_fid_fini(struct lfsck_instance *lfsck) +{ if (lfsck->li_seq != NULL) { seq_client_fini(lfsck->li_seq); OBD_FREE_PTR(lfsck->li_seq); @@ -794,54 +1590,71 @@ void lfsck_instance_cleanup(const struct lu_env *env, { struct ptlrpc_thread *thread = &lfsck->li_thread; struct lfsck_component *com; + struct lfsck_component *next; + struct lfsck_lmv_unit *llu; + struct lfsck_lmv_unit *llu_next; + struct lfsck_lmv *llmv; ENTRY; LASSERT(list_empty(&lfsck->li_link)); LASSERT(thread_is_init(thread) || thread_is_stopped(thread)); if (lfsck->li_obj_oit != NULL) { - lu_object_put_nocache(env, &lfsck->li_obj_oit->do_lu); + lfsck_object_put(env, lfsck->li_obj_oit); lfsck->li_obj_oit = NULL; } LASSERT(lfsck->li_obj_dir == NULL); + LASSERT(lfsck->li_lmv == NULL); - while (!cfs_list_empty(&lfsck->li_list_scan)) { - com = cfs_list_entry(lfsck->li_list_scan.next, - struct lfsck_component, - lc_link); + list_for_each_entry_safe(llu, llu_next, &lfsck->li_list_lmv, llu_link) { + llmv = &llu->llu_lmv; + + LASSERTF(atomic_read(&llmv->ll_ref) == 1, + "still in using: %u\n", + atomic_read(&llmv->ll_ref)); + + lfsck_lmv_put(env, llmv); + } + + list_for_each_entry_safe(com, next, &lfsck->li_list_scan, lc_link) { lfsck_component_cleanup(env, com); } - LASSERT(cfs_list_empty(&lfsck->li_list_dir)); + LASSERT(list_empty(&lfsck->li_list_dir)); - while (!cfs_list_empty(&lfsck->li_list_double_scan)) { - com = cfs_list_entry(lfsck->li_list_double_scan.next, - struct lfsck_component, - lc_link); + list_for_each_entry_safe(com, next, &lfsck->li_list_double_scan, + lc_link) { lfsck_component_cleanup(env, com); } - while (!cfs_list_empty(&lfsck->li_list_idle)) { - com = cfs_list_entry(lfsck->li_list_idle.next, - struct lfsck_component, - lc_link); + list_for_each_entry_safe(com, next, &lfsck->li_list_idle, lc_link) { lfsck_component_cleanup(env, com); } lfsck_tgt_descs_fini(&lfsck->li_ost_descs); lfsck_tgt_descs_fini(&lfsck->li_mdt_descs); + if (lfsck->li_lfsck_dir != NULL) { + lfsck_object_put(env, lfsck->li_lfsck_dir); + lfsck->li_lfsck_dir = NULL; + } + if (lfsck->li_bookmark_obj != NULL) { - lu_object_put_nocache(env, &lfsck->li_bookmark_obj->do_lu); + lfsck_object_put(env, lfsck->li_bookmark_obj); lfsck->li_bookmark_obj = NULL; } if (lfsck->li_lpf_obj != NULL) { - lu_object_put(env, &lfsck->li_lpf_obj->do_lu); + lfsck_object_put(env, lfsck->li_lpf_obj); lfsck->li_lpf_obj = NULL; } + if (lfsck->li_lpf_root_obj != NULL) { + lfsck_object_put(env, lfsck->li_lpf_root_obj); + lfsck->li_lpf_root_obj = NULL; + } + if (lfsck->li_los != NULL) { local_oid_storage_fini(env, lfsck->li_los); lfsck->li_los = NULL; @@ -857,7 +1670,7 @@ __lfsck_instance_find(struct dt_device *key, bool ref, bool unlink) { struct lfsck_instance *lfsck; - cfs_list_for_each_entry(lfsck, &lfsck_instance_list, li_link) { + list_for_each_entry(lfsck, &lfsck_instance_list, li_link) { if (lfsck->li_bottom == key) { if (ref) lfsck_instance_get(lfsck); @@ -888,33 +1701,30 @@ static inline int lfsck_instance_add(struct lfsck_instance *lfsck) struct lfsck_instance *tmp; spin_lock(&lfsck_instance_lock); - cfs_list_for_each_entry(tmp, &lfsck_instance_list, li_link) { + list_for_each_entry(tmp, &lfsck_instance_list, li_link) { if (lfsck->li_bottom == tmp->li_bottom) { spin_unlock(&lfsck_instance_lock); return -EEXIST; } } - cfs_list_add_tail(&lfsck->li_link, &lfsck_instance_list); + list_add_tail(&lfsck->li_link, &lfsck_instance_list); spin_unlock(&lfsck_instance_lock); return 0; } -int lfsck_bits_dump(char **buf, int *len, int bits, const char *names[], +int lfsck_bits_dump(struct seq_file *m, int bits, const char *names[], const char *prefix) { - int save = *len; int flag; - int rc; int i; bool newline = (bits != 0 ? false : true); + int rc; - rc = snprintf(*buf, *len, "%s:%c", prefix, newline ? '\n' : ' '); - if (rc <= 0) - return -ENOSPC; + rc = seq_printf(m, "%s:%c", prefix, bits != 0 ? ' ' : '\n'); + if (rc < 0) + return rc; - *buf += rc; - *len -= rc; for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) { if (flag & bits) { bits &= ~flag; @@ -922,69 +1732,54 @@ int lfsck_bits_dump(char **buf, int *len, int bits, const char *names[], if (bits == 0) newline = true; - rc = snprintf(*buf, *len, "%s%c", names[i], - newline ? '\n' : ','); - if (rc <= 0) - return -ENOSPC; - - *buf += rc; - *len -= rc; + rc = seq_printf(m, "%s%c", names[i], + newline ? '\n' : ','); + if (rc < 0) + return rc; } } } - if (!newline) { - rc = snprintf(*buf, *len, "\n"); - if (rc <= 0) - return -ENOSPC; - - *buf += rc; - *len -= rc; - } + if (!newline) + rc = seq_printf(m, "\n"); - return save - *len; + return rc; } -int lfsck_time_dump(char **buf, int *len, __u64 time, const char *prefix) +int lfsck_time_dump(struct seq_file *m, __u64 time, const char *name) { int rc; - if (time != 0) - rc = snprintf(*buf, *len, "%s: "LPU64" seconds\n", prefix, - cfs_time_current_sec() - time); - else - rc = snprintf(*buf, *len, "%s: N/A\n", prefix); - if (rc <= 0) - return -ENOSPC; + if (time == 0) { + rc = seq_printf(m, "%s_time: N/A\n", name); + if (rc == 0) + rc = seq_printf(m, "time_since_%s: N/A\n", name); + + return rc; + } + + rc = seq_printf(m, "%s_time: "LPU64"\n", name, time); + if (rc == 0) + rc = seq_printf(m, "time_since_%s: "LPU64" seconds\n", + name, cfs_time_current_sec() - time); - *buf += rc; - *len -= rc; return rc; } -int lfsck_pos_dump(char **buf, int *len, struct lfsck_position *pos, +int lfsck_pos_dump(struct seq_file *m, struct lfsck_position *pos, const char *prefix) { - int rc; - if (fid_is_zero(&pos->lp_dir_parent)) { if (pos->lp_oit_cookie == 0) - rc = snprintf(*buf, *len, "%s: N/A, N/A, N/A\n", - prefix); - else - rc = snprintf(*buf, *len, "%s: "LPU64", N/A, N/A\n", - prefix, pos->lp_oit_cookie); - } else { - rc = snprintf(*buf, *len, "%s: "LPU64", "DFID", "LPU64"\n", - prefix, pos->lp_oit_cookie, - PFID(&pos->lp_dir_parent), pos->lp_dir_cookie); + return seq_printf(m, "%s: N/A, N/A, N/A\n", prefix); + + return seq_printf(m, "%s: "LPU64", N/A, N/A\n", + prefix, pos->lp_oit_cookie); } - if (rc <= 0) - return -ENOSPC; - *buf += rc; - *len -= rc; - return rc; + return seq_printf(m, "%s: "LPU64", "DFID", "LPX64"\n", + prefix, pos->lp_oit_cookie, + PFID(&pos->lp_dir_parent), pos->lp_dir_cookie); } void lfsck_pos_fill(const struct lu_env *env, struct lfsck_instance *lfsck, @@ -1026,12 +1821,14 @@ bool __lfsck_set_speed(struct lfsck_instance *lfsck, __u32 limit) bool dirty = false; if (limit != LFSCK_SPEED_NO_LIMIT) { - if (limit > HZ) { - lfsck->li_sleep_rate = limit / HZ; + if (limit > msecs_to_jiffies(MSEC_PER_SEC)) { + lfsck->li_sleep_rate = limit / + msecs_to_jiffies(MSEC_PER_SEC); lfsck->li_sleep_jif = 1; } else { lfsck->li_sleep_rate = 1; - lfsck->li_sleep_jif = HZ / limit; + lfsck->li_sleep_jif = msecs_to_jiffies(MSEC_PER_SEC) / + limit; } } else { lfsck->li_sleep_jif = 0; @@ -1081,112 +1878,10 @@ void lfsck_control_speed_by_self(struct lfsck_component *com) } } -static int lfsck_parent_fid(const struct lu_env *env, struct dt_object *obj, - struct lu_fid *fid) -{ - if (unlikely(!S_ISDIR(lfsck_object_type(obj)) || - !dt_try_as_dir(env, obj))) - return -ENOTDIR; - - return dt_lookup(env, obj, (struct dt_rec *)fid, - (const struct dt_key *)"..", BYPASS_CAPA); -} - -static int lfsck_needs_scan_dir(const struct lu_env *env, - struct lfsck_instance *lfsck, - struct dt_object *obj) -{ - struct lu_fid *fid = &lfsck_env_info(env)->lti_fid; - int depth = 0; - int rc; - - if (!lfsck->li_master || !S_ISDIR(lfsck_object_type(obj)) || - cfs_list_empty(&lfsck->li_list_dir)) - RETURN(0); - - while (1) { - /* XXX: Currently, we do not scan the "/REMOTE_PARENT_DIR", - * which is the agent directory to manage the objects - * which name entries reside on remote MDTs. Related - * consistency verification will be processed in LFSCK - * phase III. */ - if (lu_fid_eq(lfsck_dto2fid(obj), &lfsck->li_global_root_fid)) { - if (depth > 0) - lfsck_object_put(env, obj); - return 1; - } - - /* No need to check .lustre and its children. */ - if (fid_seq_is_dot_lustre(fid_seq(lfsck_dto2fid(obj)))) { - if (depth > 0) - lfsck_object_put(env, obj); - return 0; - } - - dt_read_lock(env, obj, MOR_TGT_CHILD); - if (unlikely(lfsck_is_dead_obj(obj))) { - dt_read_unlock(env, obj); - if (depth > 0) - lfsck_object_put(env, obj); - return 0; - } - - rc = dt_xattr_get(env, obj, - lfsck_buf_get(env, NULL, 0), XATTR_NAME_LINK, - BYPASS_CAPA); - dt_read_unlock(env, obj); - if (rc >= 0) { - if (depth > 0) - lfsck_object_put(env, obj); - return 1; - } - - if (rc < 0 && rc != -ENODATA) { - if (depth > 0) - lfsck_object_put(env, obj); - return rc; - } - - rc = lfsck_parent_fid(env, obj, fid); - if (depth > 0) - lfsck_object_put(env, obj); - if (rc != 0) - return rc; - - if (unlikely(lu_fid_eq(fid, &lfsck->li_local_root_fid))) - return 0; - - obj = lfsck_object_find(env, lfsck, fid); - if (obj == NULL) - return 0; - else if (IS_ERR(obj)) - return PTR_ERR(obj); - - if (!dt_object_exists(obj)) { - lfsck_object_put(env, obj); - return 0; - } - - if (dt_object_remote(obj)) { - /* .lustre/lost+found/MDTxxx can be remote directory. */ - if (fid_seq_is_dot_lustre(fid_seq(lfsck_dto2fid(obj)))) - rc = 0; - else - /* Other remote directory should be client - * visible and need to be checked. */ - rc = 1; - lfsck_object_put(env, obj); - return rc; - } - - depth++; - } - return 0; -} - -struct lfsck_thread_args *lfsck_thread_args_init(struct lfsck_instance *lfsck, - struct lfsck_component *com, - struct lfsck_start_param *lsp) +static struct lfsck_thread_args * +lfsck_thread_args_init(struct lfsck_instance *lfsck, + struct lfsck_component *com, + struct lfsck_start_param *lsp) { struct lfsck_thread_args *lta; int rc; @@ -1219,253 +1914,289 @@ void lfsck_thread_args_fini(struct lfsck_thread_args *lta) OBD_FREE_PTR(lta); } -/* LFSCK wrap functions */ - -void lfsck_fail(const struct lu_env *env, struct lfsck_instance *lfsck, - bool new_checked) +struct lfsck_assistant_data * +lfsck_assistant_data_init(struct lfsck_assistant_operations *lao, + const char *name) { - struct lfsck_component *com; - - cfs_list_for_each_entry(com, &lfsck->li_list_scan, lc_link) { - com->lc_ops->lfsck_fail(env, com, new_checked); - } -} - -int lfsck_checkpoint(const struct lu_env *env, struct lfsck_instance *lfsck) -{ - struct lfsck_component *com; - int rc = 0; - int rc1 = 0; - - if (likely(cfs_time_beforeq(cfs_time_current(), - lfsck->li_time_next_checkpoint))) - return 0; + struct lfsck_assistant_data *lad; + + OBD_ALLOC_PTR(lad); + if (lad != NULL) { + lad->lad_bitmap = CFS_ALLOCATE_BITMAP(BITS_PER_LONG); + if (lad->lad_bitmap == NULL) { + OBD_FREE_PTR(lad); + return NULL; + } - lfsck_pos_fill(env, lfsck, &lfsck->li_pos_current, false); - cfs_list_for_each_entry(com, &lfsck->li_list_scan, lc_link) { - rc = com->lc_ops->lfsck_checkpoint(env, com, false); - if (rc != 0) - rc1 = rc; + INIT_LIST_HEAD(&lad->lad_req_list); + spin_lock_init(&lad->lad_lock); + INIT_LIST_HEAD(&lad->lad_ost_list); + INIT_LIST_HEAD(&lad->lad_ost_phase1_list); + INIT_LIST_HEAD(&lad->lad_ost_phase2_list); + INIT_LIST_HEAD(&lad->lad_mdt_list); + INIT_LIST_HEAD(&lad->lad_mdt_phase1_list); + INIT_LIST_HEAD(&lad->lad_mdt_phase2_list); + init_waitqueue_head(&lad->lad_thread.t_ctl_waitq); + lad->lad_ops = lao; + lad->lad_name = name; } - lfsck->li_time_last_checkpoint = cfs_time_current(); - lfsck->li_time_next_checkpoint = lfsck->li_time_last_checkpoint + - cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL); - return rc1 != 0 ? rc1 : rc; + return lad; } -int lfsck_prep(const struct lu_env *env, struct lfsck_instance *lfsck, - struct lfsck_start_param *lsp) +struct lfsck_assistant_object * +lfsck_assistant_object_init(const struct lu_env *env, const struct lu_fid *fid, + const struct lu_attr *attr, __u64 cookie, + bool is_dir) { - struct dt_object *obj = NULL; - struct lfsck_component *com; - struct lfsck_component *next; - struct lfsck_position *pos = NULL; - const struct dt_it_ops *iops = - &lfsck->li_obj_oit->do_index_ops->dio_it; - struct dt_it *di; - int rc; - ENTRY; - - LASSERT(lfsck->li_obj_dir == NULL); - LASSERT(lfsck->li_di_dir == NULL); + struct lfsck_assistant_object *lso; - lfsck->li_current_oit_processed = 0; - cfs_list_for_each_entry_safe(com, next, &lfsck->li_list_scan, lc_link) { - com->lc_new_checked = 0; - if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN) - com->lc_journal = 0; - - rc = com->lc_ops->lfsck_prep(env, com, lsp); - if (rc != 0) - GOTO(out, rc); - - if ((pos == NULL) || - (!lfsck_pos_is_zero(&com->lc_pos_start) && - lfsck_pos_is_eq(pos, &com->lc_pos_start) > 0)) - pos = &com->lc_pos_start; - } - - /* Init otable-based iterator. */ - if (pos == NULL) { - rc = iops->load(env, lfsck->li_di_oit, 0); - if (rc > 0) { - lfsck->li_oit_over = 1; - rc = 0; - } - - GOTO(out, rc); - } - - rc = iops->load(env, lfsck->li_di_oit, pos->lp_oit_cookie); - if (rc < 0) - GOTO(out, rc); - else if (rc > 0) - lfsck->li_oit_over = 1; - - if (!lfsck->li_master || fid_is_zero(&pos->lp_dir_parent)) - GOTO(out, rc = 0); - - /* Find the directory for namespace-based traverse. */ - obj = lfsck_object_find(env, lfsck, &pos->lp_dir_parent); - if (obj == NULL) - GOTO(out, rc = 0); - else if (IS_ERR(obj)) - RETURN(PTR_ERR(obj)); - - /* XXX: Currently, skip remote object, the consistency for - * remote object will be processed in LFSCK phase III. */ - if (!dt_object_exists(obj) || dt_object_remote(obj) || - unlikely(!S_ISDIR(lfsck_object_type(obj)))) - GOTO(out, rc = 0); - - if (unlikely(!dt_try_as_dir(env, obj))) - GOTO(out, rc = -ENOTDIR); + OBD_ALLOC_PTR(lso); + if (lso == NULL) + return ERR_PTR(-ENOMEM); - /* Init the namespace-based directory traverse. */ - iops = &obj->do_index_ops->dio_it; - di = iops->init(env, obj, lfsck->li_args_dir, BYPASS_CAPA); - if (IS_ERR(di)) - GOTO(out, rc = PTR_ERR(di)); + lso->lso_fid = *fid; + if (attr != NULL) + lso->lso_attr = *attr; - LASSERT(pos->lp_dir_cookie < MDS_DIR_END_OFF); + atomic_set(&lso->lso_ref, 1); + lso->lso_oit_cookie = cookie; + if (is_dir) + lso->lso_is_dir = 1; - rc = iops->load(env, di, pos->lp_dir_cookie); - if ((rc == 0) || (rc > 0 && pos->lp_dir_cookie > 0)) - rc = iops->next(env, di); - else if (rc > 0) - rc = 0; - - if (rc != 0) { - iops->put(env, di); - iops->fini(env, di); - GOTO(out, rc); - } + return lso; +} - lfsck->li_obj_dir = lfsck_object_get(obj); - lfsck->li_cookie_dir = iops->store(env, di); - spin_lock(&lfsck->li_lock); - lfsck->li_di_dir = di; - spin_unlock(&lfsck->li_lock); +struct dt_object * +lfsck_assistant_object_load(const struct lu_env *env, + struct lfsck_instance *lfsck, + struct lfsck_assistant_object *lso) +{ + struct dt_object *obj; - GOTO(out, rc = 0); + obj = lfsck_object_find_bottom(env, lfsck, &lso->lso_fid); + if (IS_ERR(obj)) + return obj; -out: - if (obj != NULL) + if (unlikely(!dt_object_exists(obj) || lfsck_is_dead_obj(obj))) { + lso->lso_dead = 1; lfsck_object_put(env, obj); - if (rc < 0) { - cfs_list_for_each_entry_safe(com, next, &lfsck->li_list_scan, - lc_link) - com->lc_ops->lfsck_post(env, com, rc, true); - - return rc; + return ERR_PTR(-ENOENT); } - rc = 0; - lfsck_pos_fill(env, lfsck, &lfsck->li_pos_current, true); - cfs_list_for_each_entry(com, &lfsck->li_list_scan, lc_link) { - rc = com->lc_ops->lfsck_checkpoint(env, com, true); - if (rc != 0) - break; + if (lso->lso_is_dir && unlikely(!dt_try_as_dir(env, obj))) { + lfsck_object_put(env, obj); + + return ERR_PTR(-ENOTDIR); } - lfsck->li_time_last_checkpoint = cfs_time_current(); - lfsck->li_time_next_checkpoint = lfsck->li_time_last_checkpoint + - cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL); - return rc; + return obj; } -int lfsck_exec_oit(const struct lu_env *env, struct lfsck_instance *lfsck, - struct dt_object *obj) +/** + * Generic LFSCK asynchronous communication interpretor function. + * The LFSCK RPC reply for both the event notification and status + * querying will be handled here. + * + * \param[in] env pointer to the thread context + * \param[in] req pointer to the LFSCK request + * \param[in] args pointer to the lfsck_async_interpret_args + * \param[in] rc the result for handling the LFSCK request + * + * \retval 0 for success + * \retval negative error number on failure + */ +int lfsck_async_interpret_common(const struct lu_env *env, + struct ptlrpc_request *req, + void *args, int rc) { - struct lfsck_component *com; - const struct dt_it_ops *iops; - struct dt_it *di; - int rc; - ENTRY; + struct lfsck_async_interpret_args *laia = args; + struct lfsck_component *com = laia->laia_com; + struct lfsck_assistant_data *lad = com->lc_data; + struct lfsck_tgt_descs *ltds = laia->laia_ltds; + struct lfsck_tgt_desc *ltd = laia->laia_ltd; + struct lfsck_request *lr = laia->laia_lr; - LASSERT(lfsck->li_obj_dir == NULL); + LASSERT(com->lc_lfsck->li_master); - cfs_list_for_each_entry(com, &lfsck->li_list_scan, lc_link) { - rc = com->lc_ops->lfsck_exec_oit(env, com, obj); - if (rc != 0) - RETURN(rc); - } + switch (lr->lr_event) { + case LE_START: + if (rc != 0) { + CDEBUG(D_LFSCK, "%s: fail to notify %s %x for %s " + "start: rc = %d\n", + lfsck_lfsck2name(com->lc_lfsck), + (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT", + ltd->ltd_index, lad->lad_name, rc); + + if (com->lc_type == LFSCK_TYPE_LAYOUT) { + struct lfsck_layout *lo = com->lc_file_ram; + + if (lr->lr_flags & LEF_TO_OST) + lfsck_lad_set_bitmap(env, com, + ltd->ltd_index); + else + lo->ll_flags |= LF_INCOMPLETE; + } else { + struct lfsck_namespace *ns = com->lc_file_ram; + + /* If some MDT does not join the namespace + * LFSCK, then we cannot know whether there + * is some name entry on such MDT that with + * the referenced MDT-object on this MDT or + * not. So the namespace LFSCK on this MDT + * cannot handle orphan MDT-objects properly. + * So we mark the LFSCK as LF_INCOMPLETE and + * skip orphan MDT-objects handling. */ + ns->ln_flags |= LF_INCOMPLETE; + } + break; + } - rc = lfsck_needs_scan_dir(env, lfsck, obj); - if (rc <= 0) - GOTO(out, rc); + spin_lock(<ds->ltd_lock); + if (ltd->ltd_dead) { + spin_unlock(<ds->ltd_lock); + break; + } - if (unlikely(!dt_try_as_dir(env, obj))) - GOTO(out, rc = -ENOTDIR); + if (com->lc_type == LFSCK_TYPE_LAYOUT) { + struct list_head *list; + struct list_head *phase_list; - iops = &obj->do_index_ops->dio_it; - di = iops->init(env, obj, lfsck->li_args_dir, BYPASS_CAPA); - if (IS_ERR(di)) - GOTO(out, rc = PTR_ERR(di)); + if (ltd->ltd_layout_done) { + spin_unlock(<ds->ltd_lock); + break; + } - rc = iops->load(env, di, 0); - if (rc == 0) - rc = iops->next(env, di); - else if (rc > 0) - rc = 0; + if (lr->lr_flags & LEF_TO_OST) { + list = &lad->lad_ost_list; + phase_list = &lad->lad_ost_phase1_list; + } else { + list = &lad->lad_mdt_list; + phase_list = &lad->lad_mdt_phase1_list; + } - if (rc != 0) { - iops->put(env, di); - iops->fini(env, di); - GOTO(out, rc); - } + if (list_empty(<d->ltd_layout_list)) + list_add_tail(<d->ltd_layout_list, list); + if (list_empty(<d->ltd_layout_phase_list)) + list_add_tail(<d->ltd_layout_phase_list, + phase_list); + } else { + if (ltd->ltd_namespace_done) { + spin_unlock(<ds->ltd_lock); + break; + } - lfsck->li_obj_dir = lfsck_object_get(obj); - lfsck->li_cookie_dir = iops->store(env, di); - spin_lock(&lfsck->li_lock); - lfsck->li_di_dir = di; - spin_unlock(&lfsck->li_lock); + if (list_empty(<d->ltd_namespace_list)) + list_add_tail(<d->ltd_namespace_list, + &lad->lad_mdt_list); + if (list_empty(<d->ltd_namespace_phase_list)) + list_add_tail(<d->ltd_namespace_phase_list, + &lad->lad_mdt_phase1_list); + } + spin_unlock(<ds->ltd_lock); + break; + case LE_STOP: + case LE_PHASE1_DONE: + case LE_PHASE2_DONE: + case LE_PEER_EXIT: + if (rc != 0 && rc != -EALREADY) + CDEBUG(D_LFSCK, "%s: fail to notify %s %x for %s: " + "event = %d, rc = %d\n", + lfsck_lfsck2name(com->lc_lfsck), + (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT", + ltd->ltd_index, lad->lad_name, lr->lr_event, rc); + break; + case LE_QUERY: { + struct lfsck_reply *reply; + struct list_head *list; + struct list_head *phase_list; + + if (com->lc_type == LFSCK_TYPE_LAYOUT) { + list = <d->ltd_layout_list; + phase_list = <d->ltd_layout_phase_list; + } else { + list = <d->ltd_namespace_list; + phase_list = <d->ltd_namespace_phase_list; + } - GOTO(out, rc = 0); + if (rc != 0) { + spin_lock(<ds->ltd_lock); + list_del_init(phase_list); + list_del_init(list); + spin_unlock(<ds->ltd_lock); + break; + } -out: - if (rc < 0) - lfsck_fail(env, lfsck, false); - return (rc > 0 ? 0 : rc); -} + reply = req_capsule_server_get(&req->rq_pill, + &RMF_LFSCK_REPLY); + if (reply == NULL) { + rc = -EPROTO; + CDEBUG(D_LFSCK, "%s: invalid query reply for %s: " + "rc = %d\n", lfsck_lfsck2name(com->lc_lfsck), + lad->lad_name, rc); + spin_lock(<ds->ltd_lock); + list_del_init(phase_list); + list_del_init(list); + spin_unlock(<ds->ltd_lock); + break; + } -int lfsck_exec_dir(const struct lu_env *env, struct lfsck_instance *lfsck, - struct dt_object *obj, struct lu_dirent *ent) -{ - struct lfsck_component *com; - int rc; + switch (reply->lr_status) { + case LS_SCANNING_PHASE1: + break; + case LS_SCANNING_PHASE2: + spin_lock(<ds->ltd_lock); + list_del_init(phase_list); + if (ltd->ltd_dead) { + spin_unlock(<ds->ltd_lock); + break; + } - cfs_list_for_each_entry(com, &lfsck->li_list_scan, lc_link) { - rc = com->lc_ops->lfsck_exec_dir(env, com, obj, ent); - if (rc != 0) - return rc; - } - return 0; -} + if (com->lc_type == LFSCK_TYPE_LAYOUT) { + if (ltd->ltd_layout_done) { + spin_unlock(<ds->ltd_lock); + break; + } -int lfsck_post(const struct lu_env *env, struct lfsck_instance *lfsck, - int result) -{ - struct lfsck_component *com; - struct lfsck_component *next; - int rc = 0; - int rc1 = 0; + if (lr->lr_flags & LEF_TO_OST) + list_add_tail(phase_list, + &lad->lad_ost_phase2_list); + else + list_add_tail(phase_list, + &lad->lad_mdt_phase2_list); + } else { + if (ltd->ltd_namespace_done) { + spin_unlock(<ds->ltd_lock); + break; + } - lfsck_pos_fill(env, lfsck, &lfsck->li_pos_current, false); - cfs_list_for_each_entry_safe(com, next, &lfsck->li_list_scan, lc_link) { - rc = com->lc_ops->lfsck_post(env, com, result, false); - if (rc != 0) - rc1 = rc; + list_add_tail(phase_list, + &lad->lad_mdt_phase2_list); + } + spin_unlock(<ds->ltd_lock); + break; + default: + spin_lock(<ds->ltd_lock); + list_del_init(phase_list); + list_del_init(list); + spin_unlock(<ds->ltd_lock); + break; + } + break; + } + default: + CDEBUG(D_LFSCK, "%s: unexpected event: rc = %d\n", + lfsck_lfsck2name(com->lc_lfsck), lr->lr_event); + break; } - lfsck->li_time_last_checkpoint = cfs_time_current(); - lfsck->li_time_next_checkpoint = lfsck->li_time_last_checkpoint + - cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL); + if (!laia->laia_shared) { + lfsck_tgt_put(ltd); + lfsck_component_put(env, com); + } - /* Ignore some component post failure to make other can go ahead. */ - return result; + return 0; } static void lfsck_interpret(const struct lu_env *env, @@ -1480,66 +2211,28 @@ static void lfsck_interpret(const struct lu_env *env, spin_lock(&lfsck->li_lock); list_for_each_entry(com, &lfsck->li_list_scan, lc_link) { - if (com->lc_ops->lfsck_interpret != NULL) { - laia->laia_com = com; - com->lc_ops->lfsck_interpret(env, req, laia, result); - } + laia->laia_com = com; + lfsck_async_interpret_common(env, req, laia, result); } list_for_each_entry(com, &lfsck->li_list_double_scan, lc_link) { - if (com->lc_ops->lfsck_interpret != NULL) { - laia->laia_com = com; - com->lc_ops->lfsck_interpret(env, req, laia, result); - } + laia->laia_com = com; + lfsck_async_interpret_common(env, req, laia, result); } spin_unlock(&lfsck->li_lock); } -int lfsck_double_scan(const struct lu_env *env, struct lfsck_instance *lfsck) -{ - struct lfsck_component *com; - struct lfsck_component *next; - struct l_wait_info lwi = { 0 }; - int rc = 0; - int rc1 = 0; - - list_for_each_entry(com, &lfsck->li_list_double_scan, lc_link) { - if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN) - com->lc_journal = 0; - - rc = com->lc_ops->lfsck_double_scan(env, com); - if (rc != 0) - rc1 = rc; - } - - l_wait_event(lfsck->li_thread.t_ctl_waitq, - atomic_read(&lfsck->li_double_scan_count) == 0, - &lwi); - - if (lfsck->li_status != LS_PAUSED && - lfsck->li_status != LS_CO_PAUSED) { - list_for_each_entry_safe(com, next, &lfsck->li_list_double_scan, - lc_link) { - spin_lock(&lfsck->li_lock); - list_del_init(&com->lc_link); - list_add_tail(&com->lc_link, &lfsck->li_list_idle); - spin_unlock(&lfsck->li_lock); - } - } - - return rc1 != 0 ? rc1 : rc; -} - static int lfsck_stop_notify(const struct lu_env *env, struct lfsck_instance *lfsck, struct lfsck_tgt_descs *ltds, struct lfsck_tgt_desc *ltd, __u16 type) { - struct ptlrpc_request_set *set; - struct lfsck_component *com; - int rc = 0; + struct lfsck_component *com; + int rc = 0; ENTRY; + LASSERT(lfsck->li_master); + spin_lock(&lfsck->li_lock); com = __lfsck_component_find(lfsck, type, &lfsck->li_list_scan); if (com == NULL) @@ -1547,58 +2240,79 @@ static int lfsck_stop_notify(const struct lu_env *env, &lfsck->li_list_double_scan); if (com != NULL) lfsck_component_get(com); - spin_lock(&lfsck->li_lock); + spin_unlock(&lfsck->li_lock); if (com != NULL) { - if (com->lc_ops->lfsck_stop_notify != NULL) { - set = ptlrpc_prep_set(); - if (set == NULL) { - lfsck_component_put(env, com); + struct lfsck_thread_info *info = lfsck_env_info(env); + struct lfsck_async_interpret_args *laia = &info->lti_laia; + struct lfsck_request *lr = &info->lti_lr; + struct lfsck_assistant_data *lad = com->lc_data; + struct list_head *list; + struct list_head *phase_list; + struct ptlrpc_request_set *set; + + set = ptlrpc_prep_set(); + if (set == NULL) { + lfsck_component_put(env, com); - RETURN(-ENOMEM); - } + RETURN(-ENOMEM); + } - rc = com->lc_ops->lfsck_stop_notify(env, com, ltds, - ltd, set); - if (rc == 0) - rc = ptlrpc_set_wait(set); + if (type == LFSCK_TYPE_LAYOUT) { + list = <d->ltd_layout_list; + phase_list = <d->ltd_layout_phase_list; + } else { + list = <d->ltd_namespace_list; + phase_list = <d->ltd_namespace_phase_list; + } + spin_lock(<ds->ltd_lock); + if (list_empty(list)) { + LASSERT(list_empty(phase_list)); + spin_unlock(<ds->ltd_lock); ptlrpc_set_destroy(set); - } - lfsck_component_put(env, com); - } + RETURN(0); + } - RETURN(rc); -} + list_del_init(phase_list); + list_del_init(list); + spin_unlock(<ds->ltd_lock); -void lfsck_quit(const struct lu_env *env, struct lfsck_instance *lfsck) -{ - struct lfsck_component *com; - struct lfsck_component *next; + memset(lr, 0, sizeof(*lr)); + lr->lr_index = lfsck_dev_idx(lfsck); + lr->lr_event = LE_PEER_EXIT; + lr->lr_active = type; + lr->lr_status = LS_CO_PAUSED; + if (ltds == &lfsck->li_ost_descs) + lr->lr_flags = LEF_TO_OST; + + laia->laia_com = com; + laia->laia_ltds = ltds; + atomic_inc(<d->ltd_ref); + laia->laia_ltd = ltd; + laia->laia_lr = lr; + laia->laia_shared = 0; - list_for_each_entry_safe(com, next, &lfsck->li_list_scan, - lc_link) { - if (com->lc_ops->lfsck_quit != NULL) - com->lc_ops->lfsck_quit(env, com); + rc = lfsck_async_request(env, ltd->ltd_exp, lr, set, + lfsck_async_interpret_common, + laia, LFSCK_NOTIFY); + if (rc != 0) { + CDEBUG(D_LFSCK, "%s: fail to notify %s %x for " + "co-stop for %s: rc = %d\n", + lfsck_lfsck2name(lfsck), + (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT", + ltd->ltd_index, lad->lad_name, rc); + lfsck_tgt_put(ltd); + } else { + rc = ptlrpc_set_wait(set); + } - spin_lock(&lfsck->li_lock); - list_del_init(&com->lc_link); - list_del_init(&com->lc_link_dir); - list_add_tail(&com->lc_link, &lfsck->li_list_idle); - spin_unlock(&lfsck->li_lock); + ptlrpc_set_destroy(set); + lfsck_component_put(env, com); } - list_for_each_entry_safe(com, next, &lfsck->li_list_double_scan, - lc_link) { - if (com->lc_ops->lfsck_quit != NULL) - com->lc_ops->lfsck_quit(env, com); - - spin_lock(&lfsck->li_lock); - list_del_init(&com->lc_link); - list_add_tail(&com->lc_link, &lfsck->li_list_idle); - spin_unlock(&lfsck->li_lock); - } + RETURN(rc); } static int lfsck_async_interpret(const struct lu_env *env, @@ -1638,8 +2352,8 @@ int lfsck_async_request(const struct lu_env *env, struct obd_export *exp, format = &RQF_LFSCK_QUERY; break; default: - CERROR("%s: unknown async request: opc = %d\n", - exp->exp_obd->obd_name, request); + CDEBUG(D_LFSCK, "%s: unknown async request %d: rc = %d\n", + exp->exp_obd->obd_name, request, -EINVAL); return -EINVAL; } @@ -1668,9 +2382,140 @@ int lfsck_async_request(const struct lu_env *env, struct obd_export *exp, return 0; } +int lfsck_start_assistant(const struct lu_env *env, struct lfsck_component *com, + struct lfsck_start_param *lsp) +{ + struct lfsck_instance *lfsck = com->lc_lfsck; + struct lfsck_assistant_data *lad = com->lc_data; + struct ptlrpc_thread *mthread = &lfsck->li_thread; + struct ptlrpc_thread *athread = &lad->lad_thread; + struct lfsck_thread_args *lta; + struct task_struct *task; + int rc; + ENTRY; + + lad->lad_assistant_status = 0; + lad->lad_post_result = 0; + lad->lad_to_post = 0; + lad->lad_to_double_scan = 0; + lad->lad_in_double_scan = 0; + lad->lad_exit = 0; + lad->lad_advance_lock = false; + thread_set_flags(athread, 0); + + lta = lfsck_thread_args_init(lfsck, com, lsp); + if (IS_ERR(lta)) + RETURN(PTR_ERR(lta)); + + task = kthread_run(lfsck_assistant_engine, lta, lad->lad_name); + if (IS_ERR(task)) { + rc = PTR_ERR(task); + CERROR("%s: cannot start LFSCK assistant thread for %s: " + "rc = %d\n", lfsck_lfsck2name(lfsck), lad->lad_name, rc); + lfsck_thread_args_fini(lta); + } else { + struct l_wait_info lwi = { 0 }; + + l_wait_event(mthread->t_ctl_waitq, + thread_is_running(athread) || + thread_is_stopped(athread), + &lwi); + if (unlikely(!thread_is_running(athread))) + rc = lad->lad_assistant_status; + else + rc = 0; + } + + RETURN(rc); +} + +int lfsck_checkpoint_generic(const struct lu_env *env, + struct lfsck_component *com) +{ + struct lfsck_assistant_data *lad = com->lc_data; + struct ptlrpc_thread *mthread = &com->lc_lfsck->li_thread; + struct ptlrpc_thread *athread = &lad->lad_thread; + struct l_wait_info lwi = { 0 }; + + l_wait_event(mthread->t_ctl_waitq, + list_empty(&lad->lad_req_list) || + !thread_is_running(mthread) || + thread_is_stopped(athread), + &lwi); + + if (!thread_is_running(mthread) || thread_is_stopped(athread)) + return LFSCK_CHECKPOINT_SKIP; + + return 0; +} + +void lfsck_post_generic(const struct lu_env *env, + struct lfsck_component *com, int *result) +{ + struct lfsck_assistant_data *lad = com->lc_data; + struct ptlrpc_thread *athread = &lad->lad_thread; + struct ptlrpc_thread *mthread = &com->lc_lfsck->li_thread; + struct l_wait_info lwi = { 0 }; + + lad->lad_post_result = *result; + if (*result <= 0) + lad->lad_exit = 1; + lad->lad_to_post = 1; + + wake_up_all(&athread->t_ctl_waitq); + l_wait_event(mthread->t_ctl_waitq, + (*result > 0 && list_empty(&lad->lad_req_list)) || + thread_is_stopped(athread), + &lwi); + + if (lad->lad_assistant_status < 0) + *result = lad->lad_assistant_status; +} + +int lfsck_double_scan_generic(const struct lu_env *env, + struct lfsck_component *com, int status) +{ + struct lfsck_assistant_data *lad = com->lc_data; + struct ptlrpc_thread *mthread = &com->lc_lfsck->li_thread; + struct ptlrpc_thread *athread = &lad->lad_thread; + struct l_wait_info lwi = { 0 }; + + if (status != LS_SCANNING_PHASE2) + lad->lad_exit = 1; + else + lad->lad_to_double_scan = 1; + + wake_up_all(&athread->t_ctl_waitq); + l_wait_event(mthread->t_ctl_waitq, + lad->lad_in_double_scan || + thread_is_stopped(athread), + &lwi); + + if (lad->lad_assistant_status < 0) + return lad->lad_assistant_status; + + return 0; +} + +void lfsck_quit_generic(const struct lu_env *env, + struct lfsck_component *com) +{ + struct lfsck_assistant_data *lad = com->lc_data; + struct ptlrpc_thread *mthread = &com->lc_lfsck->li_thread; + struct ptlrpc_thread *athread = &lad->lad_thread; + struct l_wait_info lwi = { 0 }; + + lad->lad_exit = 1; + wake_up_all(&athread->t_ctl_waitq); + l_wait_event(mthread->t_ctl_waitq, + thread_is_init(athread) || + thread_is_stopped(athread), + &lwi); +} + /* external interfaces */ -int lfsck_get_speed(struct dt_device *key, void *buf, int len) +int lfsck_get_speed(struct seq_file *m, struct dt_device *key) { struct lu_env env; struct lfsck_instance *lfsck; @@ -1683,8 +2528,7 @@ int lfsck_get_speed(struct dt_device *key, void *buf, int len) lfsck = lfsck_instance_find(key, true, false); if (likely(lfsck != NULL)) { - rc = snprintf(buf, len, "%u\n", - lfsck->li_bookmark_ram.lb_speed_limit); + seq_printf(m, "%u\n", lfsck->li_bookmark_ram.lb_speed_limit); lfsck_instance_put(&env, lfsck); } else { rc = -ENXIO; @@ -1724,7 +2568,7 @@ int lfsck_set_speed(struct dt_device *key, int val) } EXPORT_SYMBOL(lfsck_set_speed); -int lfsck_get_windows(struct dt_device *key, void *buf, int len) +int lfsck_get_windows(struct seq_file *m, struct dt_device *key) { struct lu_env env; struct lfsck_instance *lfsck; @@ -1737,8 +2581,7 @@ int lfsck_get_windows(struct dt_device *key, void *buf, int len) lfsck = lfsck_instance_find(key, true, false); if (likely(lfsck != NULL)) { - rc = snprintf(buf, len, "%u\n", - lfsck->li_bookmark_ram.lb_async_windows); + seq_printf(m, "%u\n", lfsck->li_bookmark_ram.lb_async_windows); lfsck_instance_put(&env, lfsck); } else { rc = -ENXIO; @@ -1763,12 +2606,10 @@ int lfsck_set_windows(struct dt_device *key, int val) lfsck = lfsck_instance_find(key, true, false); if (likely(lfsck != NULL)) { - if (val > LFSCK_ASYNC_WIN_MAX) { - CWARN("%s: Too large async window size, which " - "may cause memory issues. The valid range " - "is [0 - %u]. If you do not want to restrict " - "the window size for async requests pipeline, " - "just set it as 0.\n", + if (val < 1 || val > LFSCK_ASYNC_WIN_MAX) { + CWARN("%s: invalid async windows size that may " + "cause memory issues. The valid range is " + "[1 - %u].\n", lfsck_lfsck2name(lfsck), LFSCK_ASYNC_WIN_MAX); rc = -EINVAL; } else if (lfsck->li_bookmark_ram.lb_async_windows != val) { @@ -1788,7 +2629,7 @@ int lfsck_set_windows(struct dt_device *key, int val) } EXPORT_SYMBOL(lfsck_set_windows); -int lfsck_dump(struct dt_device *key, void *buf, int len, enum lfsck_type type) +int lfsck_dump(struct seq_file *m, struct dt_device *key, enum lfsck_type type) { struct lu_env env; struct lfsck_instance *lfsck; @@ -1804,7 +2645,7 @@ int lfsck_dump(struct dt_device *key, void *buf, int len, enum lfsck_type type) if (likely(lfsck != NULL)) { com = lfsck_component_find(lfsck, type); if (likely(com != NULL)) { - rc = com->lc_ops->lfsck_dump(&env, com, buf, len); + rc = com->lc_ops->lfsck_dump(&env, com, m); lfsck_component_put(&env, com); } else { rc = -ENOTSUPP; @@ -1840,16 +2681,12 @@ static int lfsck_stop_all(const struct lu_env *env, LASSERT(stop->ls_flags & LPF_BROADCAST); set = ptlrpc_prep_set(); - if (unlikely(set == NULL)) { - CERROR("%s: cannot allocate memory for stop LFSCK on " - "all targets\n", lfsck_lfsck2name(lfsck)); - + if (unlikely(set == NULL)) RETURN(-ENOMEM); - } memset(lr, 0, sizeof(*lr)); lr->lr_event = LE_STOP; - lr->lr_index = lfsck_dev_idx(lfsck->li_bottom); + lr->lr_index = lfsck_dev_idx(lfsck); lr->lr_status = stop->ls_status; lr->lr_version = bk->lb_version; lr->lr_active = LFSCK_TYPES_ALL; @@ -1873,8 +2710,8 @@ static int lfsck_stop_all(const struct lu_env *env, if (rc != 0) { lfsck_interpret(env, lfsck, NULL, laia, rc); lfsck_tgt_put(ltd); - CWARN("%s: cannot notify MDT %x for LFSCK stop: " - "rc = %d\n", lfsck_lfsck2name(lfsck), idx, rc); + CERROR("%s: cannot notify MDT %x for LFSCK stop: " + "rc = %d\n", lfsck_lfsck2name(lfsck), idx, rc); rc1 = rc; } } @@ -1890,8 +2727,8 @@ static int lfsck_stop_all(const struct lu_env *env, rc = 0; if (rc != 0) - CWARN("%s: fail to stop LFSCK on some MDTs: rc = %d\n", - lfsck_lfsck2name(lfsck), rc); + CERROR("%s: fail to stop LFSCK on some MDTs: rc = %d\n", + lfsck_lfsck2name(lfsck), rc); RETURN(rc != 0 ? rc : rc1); } @@ -1914,32 +2751,20 @@ static int lfsck_start_all(const struct lu_env *env, LASSERT(start->ls_flags & LPF_BROADCAST); set = ptlrpc_prep_set(); - if (unlikely(set == NULL)) { - if (bk->lb_param & LPF_FAILOUT) { - CERROR("%s: cannot allocate memory for start LFSCK on " - "all targets, failout.\n", - lfsck_lfsck2name(lfsck)); - - RETURN(-ENOMEM); - } else { - CWARN("%s: cannot allocate memory for start LFSCK on " - "all targets, partly scan.\n", - lfsck_lfsck2name(lfsck)); - - RETURN(0); - } - } + if (unlikely(set == NULL)) + RETURN(-ENOMEM); memset(lr, 0, sizeof(*lr)); lr->lr_event = LE_START; - lr->lr_index = lfsck_dev_idx(lfsck->li_bottom); + lr->lr_index = lfsck_dev_idx(lfsck); lr->lr_speed = bk->lb_speed_limit; lr->lr_version = bk->lb_version; lr->lr_active = start->ls_active; lr->lr_param = start->ls_flags; lr->lr_async_windows = bk->lb_async_windows; lr->lr_valid = LSV_SPEED_LIMIT | LSV_ERROR_HANDLE | LSV_DRYRUN | - LSV_ASYNC_WINDOWS; + LSV_ASYNC_WINDOWS | LSV_CREATE_OSTOBJ | + LSV_CREATE_MDTOBJ; laia->laia_com = NULL; laia->laia_ltds = ltds; @@ -1954,23 +2779,18 @@ static int lfsck_start_all(const struct lu_env *env, laia->laia_ltd = ltd; ltd->ltd_layout_done = 0; + ltd->ltd_namespace_done = 0; + ltd->ltd_synced_failures = 0; rc = lfsck_async_request(env, ltd->ltd_exp, lr, set, lfsck_async_interpret, laia, LFSCK_NOTIFY); if (rc != 0) { lfsck_interpret(env, lfsck, NULL, laia, rc); lfsck_tgt_put(ltd); - if (bk->lb_param & LPF_FAILOUT) { - CERROR("%s: cannot notify MDT %x for LFSCK " - "start, failout: rc = %d\n", - lfsck_lfsck2name(lfsck), idx, rc); - break; - } else { - CWARN("%s: cannot notify MDT %x for LFSCK " - "start, partly scan: rc = %d\n", - lfsck_lfsck2name(lfsck), idx, rc); - rc = 0; - } + CERROR("%s: cannot notify MDT %x for LFSCK " + "start, failout: rc = %d\n", + lfsck_lfsck2name(lfsck), idx, rc); + break; } } up_read(<ds->ltd_rw_sem); @@ -1988,22 +2808,15 @@ static int lfsck_start_all(const struct lu_env *env, rc = laia->laia_result; if (rc != 0) { - if (bk->lb_param & LPF_FAILOUT) { - struct lfsck_stop *stop = &info->lti_stop; + struct lfsck_stop *stop = &info->lti_stop; - CERROR("%s: cannot start LFSCK on some MDTs, " - "stop all: rc = %d\n", - lfsck_lfsck2name(lfsck), rc); - if (rc != -EALREADY) { - stop->ls_status = LS_FAILED; - stop->ls_flags = LPF_ALL_TGT | LPF_BROADCAST; - lfsck_stop_all(env, lfsck, stop); - } - } else { - CWARN("%s: cannot start LFSCK on some MDTs, " - "partly scan: rc = %d\n", - lfsck_lfsck2name(lfsck), rc); - rc = 0; + CERROR("%s: cannot start LFSCK on some MDTs, " + "stop all: rc = %d\n", + lfsck_lfsck2name(lfsck), rc); + if (rc != -EALREADY) { + stop->ls_status = LS_FAILED; + stop->ls_flags = LPF_ALL_TGT | LPF_BROADCAST; + lfsck_stop_all(env, lfsck, stop); } } @@ -2037,7 +2850,7 @@ int lfsck_start(const struct lu_env *env, struct dt_device *key, /* start == NULL means auto trigger paused LFSCK. */ if ((start == NULL) && - (cfs_list_empty(&lfsck->li_list_scan) || + (list_empty(&lfsck->li_list_scan) || OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_AUTO))) GOTO(put, rc = 0); @@ -2125,14 +2938,12 @@ int lfsck_start(const struct lu_env *env, struct dt_device *key, if (type & start->ls_active) { com = __lfsck_component_find(lfsck, type, &lfsck->li_list_idle); - if (com != NULL) { + if (com != NULL) /* The component status will be updated * when its prep() is called later by * the LFSCK main engine. */ - list_del_init(&com->lc_link); - list_add_tail(&com->lc_link, - &lfsck->li_list_scan); - } + list_move_tail(&com->lc_link, + &lfsck->li_list_scan); start->ls_active &= ~type; } type <<= 1; @@ -2170,7 +2981,7 @@ int lfsck_start(const struct lu_env *env, struct dt_device *key, } trigger: - lfsck->li_args_dir = LUDA_64BITHASH | LUDA_VERIFY; + lfsck->li_args_dir = LUDA_64BITHASH | LUDA_VERIFY | LUDA_TYPE; if (bk->lb_param & LPF_DRYRUN) lfsck->li_args_dir |= LUDA_VERIFY_DRYRUN; @@ -2282,9 +3093,10 @@ int lfsck_stop(const struct lu_env *env, struct dt_device *key, mutex_lock(&lfsck->li_mutex); spin_lock(&lfsck->li_lock); + /* no error if LFSCK is already stopped, or was never started */ if (thread_is_init(thread) || thread_is_stopped(thread)) { spin_unlock(&lfsck->li_lock); - GOTO(out, rc = -EALREADY); + GOTO(out, rc = 0); } if (stop != NULL) { @@ -2314,7 +3126,7 @@ out: EXPORT_SYMBOL(lfsck_stop); int lfsck_in_notify(const struct lu_env *env, struct dt_device *key, - struct lfsck_request *lr) + struct lfsck_request *lr, struct thandle *th) { int rc = -EOPNOTSUPP; ENTRY; @@ -2352,6 +3164,10 @@ int lfsck_in_notify(const struct lu_env *env, struct dt_device *key, case LE_FID_ACCESSED: case LE_PEER_EXIT: case LE_CONDITIONAL_DESTROY: + case LE_SKIP_NLINK_DECLARE: + case LE_SKIP_NLINK: + case LE_SET_LMV_MASTER: + case LE_SET_LMV_SLAVE: case LE_PAIRS_VERIFY: { struct lfsck_instance *lfsck; struct lfsck_component *com; @@ -2362,7 +3178,7 @@ int lfsck_in_notify(const struct lu_env *env, struct dt_device *key, com = lfsck_component_find(lfsck, lr->lr_active); if (likely(com != NULL)) { - rc = com->lc_ops->lfsck_in_notify(env, com, lr); + rc = com->lc_ops->lfsck_in_notify(env, com, lr, th); lfsck_component_put(env, com); } @@ -2401,7 +3217,6 @@ int lfsck_query(const struct lu_env *env, struct dt_device *key, RETURN(rc); } -EXPORT_SYMBOL(lfsck_query); int lfsck_register_namespace(const struct lu_env *env, struct dt_device *key, struct ldlm_namespace *ns) @@ -2426,7 +3241,7 @@ int lfsck_register(const struct lu_env *env, struct dt_device *key, { struct lfsck_instance *lfsck; struct dt_object *root = NULL; - struct dt_object *obj; + struct dt_object *obj = NULL; struct lu_fid *fid = &lfsck_env_info(env)->lti_fid; int rc; ENTRY; @@ -2441,11 +3256,12 @@ int lfsck_register(const struct lu_env *env, struct dt_device *key, mutex_init(&lfsck->li_mutex); spin_lock_init(&lfsck->li_lock); - CFS_INIT_LIST_HEAD(&lfsck->li_link); - CFS_INIT_LIST_HEAD(&lfsck->li_list_scan); - CFS_INIT_LIST_HEAD(&lfsck->li_list_dir); - CFS_INIT_LIST_HEAD(&lfsck->li_list_double_scan); - CFS_INIT_LIST_HEAD(&lfsck->li_list_idle); + INIT_LIST_HEAD(&lfsck->li_link); + INIT_LIST_HEAD(&lfsck->li_list_scan); + INIT_LIST_HEAD(&lfsck->li_list_dir); + INIT_LIST_HEAD(&lfsck->li_list_double_scan); + INIT_LIST_HEAD(&lfsck->li_list_idle); + INIT_LIST_HEAD(&lfsck->li_list_lmv); atomic_set(&lfsck->li_ref, 1); atomic_set(&lfsck->li_double_scan_count, 0); init_waitqueue_head(&lfsck->li_thread.t_ctl_waitq); @@ -2466,7 +3282,7 @@ int lfsck_register(const struct lu_env *env, struct dt_device *key, fid->f_seq = FID_SEQ_LOCAL_NAME; fid->f_oid = 1; fid->f_ver = 0; - rc = local_oid_storage_init(env, lfsck->li_bottom, fid, &lfsck->li_los); + rc = local_oid_storage_init(env, key, fid, &lfsck->li_los); if (rc != 0) GOTO(out, rc); @@ -2474,7 +3290,7 @@ int lfsck_register(const struct lu_env *env, struct dt_device *key, if (rc != 0) GOTO(out, rc); - root = dt_locate(env, lfsck->li_bottom, fid); + root = dt_locate(env, key, fid); if (IS_ERR(root)) GOTO(out, rc = PTR_ERR(root)); @@ -2484,31 +3300,84 @@ int lfsck_register(const struct lu_env *env, struct dt_device *key, lfsck->li_local_root_fid = *fid; if (master) { lfsck->li_master = 1; - if (lfsck_dev_idx(lfsck->li_bottom) == 0) { + if (lfsck_dev_idx(lfsck) == 0) { + struct lu_fid *pfid = &lfsck_env_info(env)->lti_fid2; + const struct lu_name *cname; + rc = dt_lookup(env, root, (struct dt_rec *)(&lfsck->li_global_root_fid), - (const struct dt_key *)"ROOT", BYPASS_CAPA); + (const struct dt_key *)"ROOT"); if (rc != 0) GOTO(out, rc); + + obj = dt_locate(env, key, &lfsck->li_global_root_fid); + if (IS_ERR(obj)) + GOTO(out, rc = PTR_ERR(obj)); + + if (unlikely(!dt_try_as_dir(env, obj))) + GOTO(out, rc = -ENOTDIR); + + rc = dt_lookup(env, obj, (struct dt_rec *)fid, + (const struct dt_key *)dotlustre); + if (rc != 0) + GOTO(out, rc); + + lfsck_object_put(env, obj); + obj = dt_locate(env, key, fid); + if (IS_ERR(obj)) + GOTO(out, rc = PTR_ERR(obj)); + + cname = lfsck_name_get_const(env, dotlustre, + strlen(dotlustre)); + rc = lfsck_verify_linkea(env, obj, cname, + &lfsck->li_global_root_fid); + if (rc != 0) + GOTO(out, rc); + + if (unlikely(!dt_try_as_dir(env, obj))) + GOTO(out, rc = -ENOTDIR); + + *pfid = *fid; + rc = dt_lookup(env, obj, (struct dt_rec *)fid, + (const struct dt_key *)lostfound); + if (rc != 0) + GOTO(out, rc); + + lfsck_object_put(env, obj); + obj = dt_locate(env, key, fid); + if (IS_ERR(obj)) + GOTO(out, rc = PTR_ERR(obj)); + + cname = lfsck_name_get_const(env, lostfound, + strlen(lostfound)); + rc = lfsck_verify_linkea(env, obj, cname, pfid); + if (rc != 0) + GOTO(out, rc); + + lfsck_object_put(env, obj); + obj = NULL; } } fid->f_seq = FID_SEQ_LOCAL_FILE; fid->f_oid = OTABLE_IT_OID; fid->f_ver = 0; - obj = dt_locate(env, lfsck->li_bottom, fid); + obj = dt_locate(env, key, fid); if (IS_ERR(obj)) GOTO(out, rc = PTR_ERR(obj)); - lfsck->li_obj_oit = obj; rc = obj->do_ops->do_index_try(env, obj, &dt_otable_features); - if (rc != 0) { - if (rc == -ENOTSUPP) - GOTO(add, rc = 0); - + if (rc != 0) GOTO(out, rc); - } + lfsck->li_obj_oit = obj; + obj = local_file_find_or_create(env, lfsck->li_los, root, LFSCK_DIR, + S_IFDIR | S_IRUGO | S_IWUSR); + if (IS_ERR(obj)) + GOTO(out, rc = PTR_ERR(obj)); + + lu_object_get(&obj->do_lu); + lfsck->li_lfsck_dir = obj; rc = lfsck_bookmark_setup(env, lfsck); if (rc != 0) GOTO(out, rc); @@ -2529,13 +3398,14 @@ int lfsck_register(const struct lu_env *env, struct dt_device *key, /* XXX: more LFSCK components initialization to be added here. */ -add: rc = lfsck_instance_add(lfsck); if (rc == 0) rc = lfsck_add_target_from_orphan(env, lfsck); out: + if (obj != NULL && !IS_ERR(obj)) + lfsck_object_put(env, obj); if (root != NULL && !IS_ERR(root)) - lu_object_put(env, &root->do_lu); + lfsck_object_put(env, root); if (rc != 0) lfsck_instance_cleanup(env, lfsck); return rc; @@ -2571,6 +3441,8 @@ int lfsck_add_target(const struct lu_env *env, struct dt_device *key, INIT_LIST_HEAD(<d->ltd_orphan_list); INIT_LIST_HEAD(<d->ltd_layout_list); INIT_LIST_HEAD(<d->ltd_layout_phase_list); + INIT_LIST_HEAD(<d->ltd_namespace_list); + INIT_LIST_HEAD(<d->ltd_namespace_phase_list); atomic_set(<d->ltd_ref, 1); ltd->ltd_index = index; @@ -2604,7 +3476,7 @@ void lfsck_del_target(const struct lu_env *env, struct dt_device *key, { struct lfsck_instance *lfsck; struct lfsck_tgt_descs *ltds; - struct lfsck_tgt_desc *ltd = NULL; + struct lfsck_tgt_desc *ltd; struct list_head *head; if (for_ost) @@ -2623,6 +3495,7 @@ void lfsck_del_target(const struct lu_env *env, struct dt_device *key, } } + ltd = NULL; lfsck = __lfsck_instance_find(key, true, false); spin_unlock(&lfsck_instance_lock); if (unlikely(lfsck == NULL)) @@ -2639,7 +3512,7 @@ void lfsck_del_target(const struct lu_env *env, struct dt_device *key, if (unlikely(index >= ltds->ltd_tgts_bitmap->size)) goto unlock; - ltd = LTD_TGT(ltds, index); + ltd = lfsck_ltd2tgt(ltds, index); if (unlikely(ltd == NULL)) goto unlock; @@ -2647,14 +3520,14 @@ void lfsck_del_target(const struct lu_env *env, struct dt_device *key, ltds->ltd_tgtnr--; cfs_bitmap_clear(ltds->ltd_tgts_bitmap, index); - LTD_TGT(ltds, index) = NULL; + lfsck_assign_tgt(ltds, NULL, index); unlock: if (ltd == NULL) { if (for_ost) head = &lfsck->li_ost_descs.ltd_orphan; else - head = &lfsck->li_ost_descs.ltd_orphan; + head = &lfsck->li_mdt_descs.ltd_orphan; list_for_each_entry(ltd, head, ltd_orphan_list) { if (ltd->ltd_tgt == tgt) { @@ -2669,7 +3542,8 @@ unlock: spin_lock(<ds->ltd_lock); ltd->ltd_dead = 1; spin_unlock(<ds->ltd_lock); - lfsck_stop_notify(env, lfsck, ltds, ltd, LT_LAYOUT); + lfsck_stop_notify(env, lfsck, ltds, ltd, LFSCK_TYPE_NAMESPACE); + lfsck_stop_notify(env, lfsck, ltds, ltd, LFSCK_TYPE_LAYOUT); lfsck_tgt_put(ltd); } @@ -2681,6 +3555,7 @@ static int __init lfsck_init(void) { int rc; + INIT_LIST_HEAD(&lfsck_instance_list); INIT_LIST_HEAD(&lfsck_ost_orphan_list); INIT_LIST_HEAD(&lfsck_mdt_orphan_list); lfsck_key_init_generic(&lfsck_thread_key, NULL); @@ -2698,7 +3573,7 @@ static void __exit lfsck_exit(void) struct lfsck_tgt_desc *ltd; struct lfsck_tgt_desc *next; - LASSERT(cfs_list_empty(&lfsck_instance_list)); + LASSERT(list_empty(&lfsck_instance_list)); list_for_each_entry_safe(ltd, next, &lfsck_ost_orphan_list, ltd_orphan_list) { @@ -2717,6 +3592,8 @@ static void __exit lfsck_exit(void) MODULE_AUTHOR("Intel Corporation "); MODULE_DESCRIPTION("LFSCK"); +MODULE_VERSION(LUSTRE_VERSION_STRING); MODULE_LICENSE("GPL"); -cfs_module(lfsck, LUSTRE_VERSION_STRING, lfsck_init, lfsck_exit); +module_init(lfsck_init); +module_exit(lfsck_exit);