X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fosd-zfs%2Fosd_scrub.c;h=d3c3e900eef522ac24a2cf32c3f35abdb2ecd664;hb=c872afa36ff5de5910f0f524f7a487982fa0c776;hp=b2cda0a203e691fed665c295fadf016289e3562a;hpb=89ead218ebe99a955afc0bc7f6aba83ef35019fb;p=fs%2Flustre-release.git diff --git a/lustre/osd-zfs/osd_scrub.c b/lustre/osd-zfs/osd_scrub.c index b2cda0a..d3c3e90 100644 --- a/lustre/osd-zfs/osd_scrub.c +++ b/lustre/osd-zfs/osd_scrub.c @@ -46,6 +46,9 @@ #include #include #include +#include +#include +#include #include "osd_internal.h" @@ -329,7 +332,7 @@ static int osd_scrub_prep(const struct lu_env *env, struct osd_device *dev) scrub->os_pos_current = sf->sf_pos_latest_start; sf->sf_status = SS_SCANNING; - sf->sf_time_latest_start = cfs_time_current_sec(); + sf->sf_time_latest_start = ktime_get_real_seconds(); sf->sf_time_last_checkpoint = sf->sf_time_latest_start; sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1; rc = scrub_file_store(env, scrub); @@ -364,7 +367,7 @@ static int osd_scrub_post(const struct lu_env *env, struct osd_device *dev, scrub->os_new_checked = 0; sf->sf_pos_last_checkpoint = scrub->os_pos_current; } - sf->sf_time_last_checkpoint = cfs_time_current_sec(); + sf->sf_time_last_checkpoint = ktime_get_real_seconds(); if (result > 0) { sf->sf_status = SS_COMPLETED; if (!(sf->sf_param & SP_DRYRUN)) { @@ -382,8 +385,9 @@ static int osd_scrub_post(const struct lu_env *env, struct osd_device *dev, } else { sf->sf_status = SS_FAILED; } - sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC - - scrub->os_time_last_checkpoint); + sf->sf_run_time += ktime_get_seconds() - + scrub->os_time_last_checkpoint; + rc = scrub_file_store(env, scrub); up_write(&scrub->os_rwsem); @@ -697,7 +701,7 @@ static const struct osd_lf_map osd_lf_maps[] = { /* PENDING */ { - .olm_name = "PENDING", + .olm_name = MDT_ORPHAN_DIR, }, /* ROOT */ @@ -749,7 +753,7 @@ static const struct osd_lf_map osd_lf_maps[] = { /* LFSCK */ { .olm_name = LFSCK_DIR, - .olm_flags = OLF_SCAN_SUBITEMS, + .olm_flags = OLF_SCAN_SUBITEMS | OLF_NOT_BACKUP, .olm_scan_dir = osd_ios_general_sd, .olm_handle_dirent = osd_ios_varfid_hd, }, @@ -803,6 +807,18 @@ static const struct osd_lf_map osd_lf_maps[] = { .olm_name = LUSTRE_NODEMAP_NAME, }, + /* index_backup */ + { + .olm_name = INDEX_BACKUP_DIR, + .olm_fid = { + .f_seq = FID_SEQ_LOCAL_FILE, + .f_oid = INDEX_BACKUP_OID, + }, + .olm_flags = OLF_SCAN_SUBITEMS | OLF_NOT_BACKUP, + .olm_scan_dir = osd_ios_general_sd, + .olm_handle_dirent = osd_ios_varfid_hd, + }, + { .olm_name = NULL } @@ -864,6 +880,130 @@ static int osd_ios_new_item(struct osd_device *dev, uint64_t parent, return 0; } +static bool osd_index_need_recreate(const struct lu_env *env, + struct osd_device *dev, uint64_t oid) +{ + struct osd_thread_info *info = osd_oti_get(env); + zap_attribute_t *za = &info->oti_za2; + zap_cursor_t *zc = &info->oti_zc2; + int rc; + ENTRY; + + zap_cursor_init_serialized(zc, dev->od_os, oid, 0); + rc = -zap_cursor_retrieve(zc, za); + zap_cursor_fini(zc); + if (rc && rc != -ENOENT) + RETURN(true); + + RETURN(false); +} + +static void osd_ios_index_register(const struct lu_env *env, + struct osd_device *osd, + const struct lu_fid *fid, uint64_t oid) +{ + struct osd_thread_info *info = osd_oti_get(env); + zap_attribute_t *za = &info->oti_za2; + zap_cursor_t *zc = &info->oti_zc2; + struct zap_leaf_entry *le; + dnode_t *dn = NULL; + sa_handle_t *hdl; + __u64 mode = 0; + __u32 keysize = 0; + __u32 recsize = 0; + int rc; + ENTRY; + + rc = __osd_obj2dnode(osd->od_os, oid, &dn); + if (rc == -EEXIST || rc == -ENOENT) + RETURN_EXIT; + + if (rc < 0) + GOTO(log, rc); + + if (!osd_object_is_zap(dn)) + GOTO(log, rc = 1); + + rc = -sa_handle_get(osd->od_os, oid, NULL, SA_HDL_PRIVATE, &hdl); + if (rc) + GOTO(log, rc); + + rc = -sa_lookup(hdl, SA_ZPL_MODE(osd), &mode, sizeof(mode)); + sa_handle_destroy(hdl); + if (rc) + GOTO(log, rc); + + if (!S_ISREG(mode)) + GOTO(log, rc = 1); + + zap_cursor_init_serialized(zc, osd->od_os, oid, 0); + rc = -zap_cursor_retrieve(zc, za); + if (rc) + /* Skip empty index object */ + GOTO(fini, rc = (rc == -ENOENT ? 1 : rc)); + + if (zc->zc_zap->zap_ismicro || + !(zap_f_phys(zc->zc_zap)->zap_flags & ZAP_FLAG_UINT64_KEY)) + GOTO(fini, rc = 1); + + le = ZAP_LEAF_ENTRY(zc->zc_leaf, 0); + keysize = le->le_name_numints * 8; + recsize = za->za_integer_length * za->za_num_integers; + if (likely(keysize && recsize)) + rc = osd_index_register(osd, fid, keysize, recsize); + + GOTO(fini, rc); + +fini: + zap_cursor_fini(zc); + +log: + if (dn) + osd_dnode_rele(dn); + if (rc < 0) + CWARN("%s: failed to register index "DFID" (%u/%u): rc = %d\n", + osd_name(osd), PFID(fid), keysize, recsize, rc); + else if (!rc) + CDEBUG(D_LFSCK, "%s: registered index "DFID" (%u/%u)\n", + osd_name(osd), PFID(fid), keysize, recsize); +} + +static void osd_index_restore(const struct lu_env *env, struct osd_device *dev, + struct lustre_index_restore_unit *liru, void *buf, + int bufsize) +{ + struct luz_direntry *zde = &osd_oti_get(env)->oti_zde; + struct lu_fid *tgt_fid = &liru->liru_cfid; + struct lu_fid bak_fid; + int rc; + ENTRY; + + lustre_fid2lbx(buf, tgt_fid, bufsize); + rc = -zap_lookup(dev->od_os, dev->od_index_backup_id, buf, 8, + sizeof(*zde) / 8, (void *)zde); + if (rc) + GOTO(log, rc); + + rc = osd_get_fid_by_oid(env, dev, zde->lzd_reg.zde_dnode, &bak_fid); + if (rc) + GOTO(log, rc); + + /* The OI mapping for index may be invalid, since it will be + * re-created, not update the OI mapping, just cache it in RAM. */ + rc = osd_idc_find_and_init_with_oid(env, dev, tgt_fid, + liru->liru_clid); + if (!rc) + rc = lustre_index_restore(env, &dev->od_dt_dev, + &liru->liru_pfid, tgt_fid, &bak_fid, + liru->liru_name, &dev->od_index_backup_list, + &dev->od_lock, buf, bufsize); + GOTO(log, rc); + +log: + CDEBUG(D_WARNING, "%s: restore index '%s' with "DFID": rc = %d\n", + osd_name(dev), liru->liru_name, PFID(tgt_fid), rc); +} + /** * verify FID-in-LMA and OI entry for one object * @@ -912,7 +1052,31 @@ static int osd_ios_scan_one(const struct lu_env *env, struct osd_device *dev, RETURN(0); } + if (lma->lma_compat & LMAC_IDX_BACKUP && + osd_index_need_recreate(env, dev, oid)) { + if (parent == dev->od_root) { + lu_local_obj_fid(&tfid, + OSD_FS_ROOT_OID); + } else { + rc = osd_get_fid_by_oid(env, dev, + parent, &tfid); + if (rc) { + nvlist_free(nvbuf); + RETURN(rc); + } + } + + rc = lustre_liru_new( + &dev->od_index_restore_list, + &tfid, &lma->lma_self_fid, oid, + name, strlen(name)); + nvlist_free(nvbuf); + RETURN(rc); + } + tfid = lma->lma_self_fid; + if (!(flags & OLF_NOT_BACKUP)) + osd_ios_index_register(env, dev, &tfid, oid); } nvlist_free(nvbuf); } @@ -1176,6 +1340,31 @@ static void osd_initial_OI_scrub(const struct lu_env *env, OBD_FREE_PTR(item); } + if (!list_empty(&dev->od_index_restore_list)) { + char *buf; + + OBD_ALLOC_LARGE(buf, INDEX_BACKUP_BUFSIZE); + if (!buf) + CERROR("%s: not enough RAM for rebuild index\n", + osd_name(dev)); + + while (!list_empty(&dev->od_index_restore_list)) { + struct lustre_index_restore_unit *liru; + + liru = list_entry(dev->od_index_restore_list.next, + struct lustre_index_restore_unit, + liru_link); + list_del(&liru->liru_link); + if (buf) + osd_index_restore(env, dev, liru, buf, + INDEX_BACKUP_BUFSIZE); + OBD_FREE(liru, liru->liru_len); + } + + if (buf) + OBD_FREE_LARGE(buf, INDEX_BACKUP_BUFSIZE); + } + EXIT; } @@ -1198,7 +1387,7 @@ int osd_scrub_start(const struct lu_env *env, struct osd_device *dev, RETURN(rc == -EALREADY ? 0 : rc); } -static void osd_scrub_stop(struct osd_device *dev) +void osd_scrub_stop(struct osd_device *dev) { struct lustre_scrub *scrub = &dev->od_scrub; ENTRY; @@ -1259,6 +1448,7 @@ int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev) if (IS_ERR_OR_NULL(obj)) RETURN(obj ? PTR_ERR(obj) : -ENOENT); + obj->do_body_ops = &osd_body_scrub_ops; scrub->os_obj = obj; rc = scrub_file_load(env, scrub); if (rc == -ENOENT || rc == -EFAULT) {