X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fosd-ldiskfs%2Fosd_scrub.c;h=9a5572f81873a2b9a59acd0c2bd3da6375458674;hb=5afd3300d1ded80f79dbea92d47d287ad75f463a;hp=95b0caf14394a713e882eb209283826ef98f2864;hpb=0bc85700a2e41f369173e92bbe06b27ed1145618;p=fs%2Flustre-release.git diff --git a/lustre/osd-ldiskfs/osd_scrub.c b/lustre/osd-ldiskfs/osd_scrub.c index 95b0caf..9a5572f 100644 --- a/lustre/osd-ldiskfs/osd_scrub.c +++ b/lustre/osd-ldiskfs/osd_scrub.c @@ -40,12 +40,13 @@ #include #include #include +#include #include "osd_internal.h" #include "osd_oi.h" #include "osd_scrub.h" -#define HALF_SEC (CFS_HZ >> 1) +#define HALF_SEC (HZ >> 1) #define OSD_OTABLE_MAX_HASH 0x00000000ffffffffULL @@ -90,29 +91,29 @@ static int osd_scrub_refresh_mapping(struct osd_thread_info *info, struct osd_device *dev, const struct lu_fid *fid, const struct osd_inode_id *id, - int ops, enum oi_check_flags flags) + int ops, bool force, + enum oi_check_flags flags) { - const struct lu_env *env = info->oti_env; - struct thandle *th; - struct osd_thandle *oh; - int rc; + handle_t *th; + int rc; ENTRY; - th = dt_trans_create(env, &dev->od_dt_dev); - if (IS_ERR(th)) - RETURN(PTR_ERR(th)); + if (dev->od_scrub.os_file.sf_param & SP_DRYRUN && !force) + RETURN(0); - oh = container_of0(th, struct osd_thandle, ot_super); - LASSERT(oh->ot_handle == NULL); + /* DTO_INDEX_INSERT is enough for other two ops: + * delete/update, but save stack. */ + th = ldiskfs_journal_start_sb(osd_sb(dev), + osd_dto_credits_noquota[DTO_INDEX_INSERT]); + if (IS_ERR(th)) { + rc = PTR_ERR(th); + CERROR("%s: fail to start trans for scrub %d: rc = %d\n", + osd_name(dev), ops, rc); + RETURN(rc); + } switch (ops) { case DTO_INDEX_UPDATE: - osd_trans_declare_op(env, oh, OSD_OT_UPDATE, - osd_dto_credits_noquota[DTO_INDEX_UPDATE]); - rc = dt_trans_start_local(env, &dev->od_dt_dev, th); - if (rc != 0) - GOTO(stop, rc); - rc = osd_oi_update(info, dev, fid, id, th, flags); if (unlikely(rc == -ENOENT)) { /* Some unlink thread may removed the OI mapping. */ @@ -120,12 +121,6 @@ static int osd_scrub_refresh_mapping(struct osd_thread_info *info, } break; case DTO_INDEX_INSERT: - osd_trans_declare_op(env, oh, OSD_OT_INSERT, - osd_dto_credits_noquota[DTO_INDEX_INSERT]); - rc = dt_trans_start_local(env, &dev->od_dt_dev, th); - if (rc != 0) - GOTO(stop, rc); - rc = osd_oi_insert(info, dev, fid, id, th, flags); if (unlikely(rc == -EEXIST)) { rc = 1; @@ -163,12 +158,6 @@ static int osd_scrub_refresh_mapping(struct osd_thread_info *info, } break; case DTO_INDEX_DELETE: - osd_trans_declare_op(env, oh, OSD_OT_DELETE, - osd_dto_credits_noquota[DTO_INDEX_DELETE]); - rc = dt_trans_start_local(env, &dev->od_dt_dev, th); - if (rc != 0) - GOTO(stop, rc); - rc = osd_oi_delete(info, dev, fid, th, flags); if (rc == -ENOENT) { /* It is normal that the unlink thread has removed the @@ -181,10 +170,7 @@ static int osd_scrub_refresh_mapping(struct osd_thread_info *info, break; } - GOTO(stop, rc); - -stop: - dt_trans_stop(env, &dev->od_dt_dev, th); + ldiskfs_journal_stop(th); return rc; } @@ -221,6 +207,7 @@ static void osd_scrub_file_to_cpu(struct scrub_file *des, des->sf_run_time = le32_to_cpu(src->sf_run_time); des->sf_success_count = le32_to_cpu(src->sf_success_count); des->sf_oi_count = le16_to_cpu(src->sf_oi_count); + des->sf_internal_flags = le16_to_cpu(src->sf_internal_flags); memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE); } @@ -255,6 +242,7 @@ static void osd_scrub_file_to_le(struct scrub_file *des, des->sf_run_time = cpu_to_le32(src->sf_run_time); des->sf_success_count = cpu_to_le32(src->sf_success_count); des->sf_oi_count = cpu_to_le16(src->sf_oi_count); + des->sf_internal_flags = cpu_to_le16(src->sf_internal_flags); memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE); } @@ -276,7 +264,6 @@ void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags) memcpy(sf->sf_uuid, uuid, 16); sf->sf_status = SS_INIT; sf->sf_flags |= flags; - sf->sf_param = 0; sf->sf_run_time = 0; sf->sf_time_latest_start = 0; sf->sf_time_last_checkpoint = 0; @@ -336,7 +323,7 @@ int osd_scrub_file_store(struct osd_scrub *scrub) dev = container_of0(scrub, struct osd_device, od_scrub); credits = osd_dto_credits_noquota[DTO_WRITE_BASE] + osd_dto_credits_noquota[DTO_WRITE_BLOCK]; - jh = ldiskfs_journal_start_sb(osd_sb(dev), credits); + jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, credits); if (IS_ERR(jh)) { rc = PTR_ERR(jh); CERROR("%.16s: fail to start trans for scrub store, rc = %d\n", @@ -359,61 +346,81 @@ int osd_scrub_file_store(struct osd_scrub *scrub) return rc; } -/* OI scrub APIs */ - -static int osd_scrub_prep(struct osd_device *dev) +static int +osd_scrub_convert_ff(struct osd_thread_info *info, struct osd_device *dev, + struct inode *inode, const struct lu_fid *fid) { - struct osd_scrub *scrub = &dev->od_scrub; - struct ptlrpc_thread *thread = &scrub->os_thread; - struct scrub_file *sf = &scrub->os_file; - __u32 flags = scrub->os_start_flags; - int rc; + struct filter_fid_old *ff = &info->oti_ff; + struct dentry *dentry = &info->oti_obj_dentry; + handle_t *jh; + int size = 0; + int rc; + bool removed = false; + bool reset = true; ENTRY; - down_write(&scrub->os_rwsem); - if (flags & SS_SET_FAILOUT) - sf->sf_param |= SP_FAILOUT; + if (dev->od_scrub.os_file.sf_param & SP_DRYRUN) + RETURN(0); - if (flags & SS_CLEAR_FAILOUT) - sf->sf_param &= ~SP_FAILOUT; + /* We want the LMA to fit into the 256-byte OST inode, so operate + * as following: + * 1) read old XATTR_NAME_FID and save the parent FID; + * 2) delete the old XATTR_NAME_FID; + * 3) make new LMA and add it; + * 4) generate new XATTR_NAME_FID with the saved parent FID and add it. + * + * Making the LMA to fit into the 256-byte OST inode can save time for + * normal osd_check_lma() and for other OI scrub scanning in future. + * So it is worth to make some slow conversion here. */ + jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, + osd_dto_credits_noquota[DTO_XATTR_SET] * 3); + if (IS_ERR(jh)) { + rc = PTR_ERR(jh); + CERROR("%s: fail to start trans for convert ff: "DFID + ": rc = %d\n", + osd_name(dev), PFID(fid), rc); + RETURN(rc); + } - if (flags & SS_RESET) - osd_scrub_file_reset(scrub, - LDISKFS_SB(osd_sb(dev))->s_es->s_uuid, 0); + /* 1) read old XATTR_NAME_FID and save the parent FID */ + rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff)); + if (rc == sizeof(*ff)) { + /* 2) delete the old XATTR_NAME_FID */ + ll_vfs_dq_init(inode); + rc = inode->i_op->removexattr(dentry, XATTR_NAME_FID); + if (rc != 0) + GOTO(stop, rc); - if (flags & SS_AUTO) { - scrub->os_full_speed = 1; - sf->sf_flags |= SF_AUTO; - } else { - scrub->os_full_speed = 0; + removed = true; + } else if (unlikely(rc == -ENODATA)) { + reset = false; + } else if (rc != sizeof(struct filter_fid)) { + GOTO(stop, rc = -EINVAL); } - if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE)) - scrub->os_full_speed = 1; + /* 3) make new LMA and add it */ + rc = osd_ea_fid_set(info, inode, fid, LMAC_FID_ON_OST, 0); + if (rc == 0 && reset) + size = sizeof(struct filter_fid); + else if (rc != 0 && removed) + /* If failed, we should try to add the old back. */ + size = sizeof(struct filter_fid_old); - scrub->os_in_prior = 0; - scrub->os_waiting = 0; - scrub->os_paused = 0; - scrub->os_new_checked = 0; - if (sf->sf_pos_last_checkpoint != 0) - sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1; - else - sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1; + /* 4) generate new XATTR_NAME_FID with the saved parent FID and add it*/ + if (size > 0) { + int rc1; - 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_last_checkpoint = sf->sf_time_latest_start; - rc = osd_scrub_file_store(scrub); - if (rc == 0) { - spin_lock(&scrub->os_lock); - thread_set_flags(thread, SVC_RUNNING); - spin_unlock(&scrub->os_lock); - cfs_waitq_broadcast(&thread->t_ctl_waitq); + rc1 = __osd_xattr_set(info, inode, XATTR_NAME_FID, ff, size, + XATTR_CREATE); + if (rc1 != 0 && rc == 0) + rc = rc1; } - up_write(&scrub->os_rwsem); - RETURN(rc); + GOTO(stop, rc); + +stop: + ldiskfs_journal_stop(jh); + return rc; } static int @@ -430,6 +437,7 @@ osd_scrub_check_update(struct osd_thread_info *info, struct osd_device *dev, int ops = DTO_INDEX_UPDATE; int idx; int rc; + bool converted = false; ENTRY; down_write(&scrub->os_rwsem); @@ -458,14 +466,17 @@ osd_scrub_check_update(struct osd_thread_info *info, struct osd_device *dev, } sf->sf_flags |= SF_UPGRADE; - rc = osd_ea_fid_set(info, inode, fid, - LMAC_FID_ON_OST, 0); + sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID; + dev->od_check_ff = 1; + rc = osd_scrub_convert_ff(info, dev, inode, fid); if (rc != 0) GOTO(out, rc); + + converted = true; } if ((val == SCRUB_NEXT_NOLMA) && - (!dev->od_handle_nolma || OBD_FAIL_CHECK(OBD_FAIL_FID_NOLMA))) + (!scrub->os_convert_igif || OBD_FAIL_CHECK(OBD_FAIL_FID_NOLMA))) GOTO(out, rc = 0); if ((oii != NULL && oii->oii_insert) || (val == SCRUB_NEXT_NOLMA)) @@ -496,9 +507,11 @@ iget: switch (val) { case SCRUB_NEXT_NOLMA: sf->sf_flags |= SF_UPGRADE; - rc = osd_ea_fid_set(info, inode, fid, 0, 0); - if (rc != 0) - GOTO(out, rc); + if (!(sf->sf_param & SP_DRYRUN)) { + rc = osd_ea_fid_set(info, inode, fid, 0, 0); + if (rc != 0) + GOTO(out, rc); + } if (!(sf->sf_flags & SF_INCONSISTENT)) dev->od_igif_inoi = 0; @@ -514,6 +527,9 @@ iget: break; } } else if (osd_id_eq(lid, lid2)) { + if (converted) + sf->sf_items_updated++; + GOTO(out, rc = 0); } else { scrub->os_full_speed = 1; @@ -534,7 +550,7 @@ iget: dev->od_igif_inoi = 1; } - rc = osd_scrub_refresh_mapping(info, dev, fid, lid, ops, + rc = osd_scrub_refresh_mapping(info, dev, fid, lid, ops, false, (val == SCRUB_NEXT_OSTOBJ || val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0); if (rc == 0) { @@ -564,7 +580,7 @@ out: if (ops == DTO_INDEX_INSERT && inode != NULL && !IS_ERR(inode) && unlikely(inode->i_nlink == 0)) osd_scrub_refresh_mapping(info, dev, fid, lid, - DTO_INDEX_DELETE, + DTO_INDEX_DELETE, false, (val == SCRUB_NEXT_OSTOBJ || val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0); @@ -584,6 +600,78 @@ out: RETURN(sf->sf_param & SP_FAILOUT ? rc : 0); } +/* OI scrub APIs */ + +static int osd_scrub_prep(struct osd_device *dev) +{ + struct osd_scrub *scrub = &dev->od_scrub; + struct ptlrpc_thread *thread = &scrub->os_thread; + struct scrub_file *sf = &scrub->os_file; + __u32 flags = scrub->os_start_flags; + int rc; + bool drop_dryrun = false; + ENTRY; + + down_write(&scrub->os_rwsem); + if (flags & SS_SET_FAILOUT) + sf->sf_param |= SP_FAILOUT; + + if (flags & SS_CLEAR_FAILOUT) + sf->sf_param &= ~SP_FAILOUT; + + if (flags & SS_SET_DRYRUN) + sf->sf_param |= SP_DRYRUN; + + if (flags & SS_CLEAR_DRYRUN && sf->sf_param & SP_DRYRUN) { + sf->sf_param &= ~SP_DRYRUN; + drop_dryrun = true; + } + + if (flags & SS_RESET) + osd_scrub_file_reset(scrub, + LDISKFS_SB(osd_sb(dev))->s_es->s_uuid, 0); + + if (flags & SS_AUTO) { + scrub->os_full_speed = 1; + sf->sf_flags |= SF_AUTO; + /* For the case of OI scrub auto triggered, NOT dryrun. */ + sf->sf_param &= ~SP_FAILOUT; + } else { + scrub->os_full_speed = 0; + } + + if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE)) + scrub->os_full_speed = 1; + + scrub->os_in_prior = 0; + spin_lock(&scrub->os_lock); + scrub->os_waiting = 0; + scrub->os_paused = 0; + spin_unlock(&scrub->os_lock); + scrub->os_new_checked = 0; + if (drop_dryrun && sf->sf_pos_first_inconsistent != 0) + sf->sf_pos_latest_start = sf->sf_pos_first_inconsistent; + else if (sf->sf_pos_last_checkpoint != 0) + sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1; + else + sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1; + + 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_last_checkpoint = sf->sf_time_latest_start; + rc = osd_scrub_file_store(scrub); + if (rc == 0) { + spin_lock(&scrub->os_lock); + thread_set_flags(thread, SVC_RUNNING); + spin_unlock(&scrub->os_lock); + wake_up_all(&thread->t_ctl_waitq); + } + up_write(&scrub->os_rwsem); + + RETURN(rc); +} + static int osd_scrub_checkpoint(struct osd_scrub *scrub) { struct scrub_file *sf = &scrub->os_file; @@ -627,10 +715,13 @@ static void osd_scrub_post(struct osd_scrub *scrub, int result) container_of0(scrub, struct osd_device, od_scrub); dev->od_igif_inoi = 1; + dev->od_check_ff = 0; sf->sf_status = SS_COMPLETED; - memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE); - sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT | - SF_UPGRADE | SF_AUTO); + if (!(sf->sf_param & SP_DRYRUN)) { + memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE); + sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT | + SF_UPGRADE | SF_AUTO); + } sf->sf_time_last_complete = sf->sf_time_last_checkpoint; sf->sf_success_count++; } else if (result == 0) { @@ -700,81 +791,48 @@ static int osd_scrub_check_local_fldb(struct osd_thread_info *info, * a small local FLDB according to the . If the given FID * is in the local FLDB, then it is FID-on-OST; otherwise it's * quite possible for FID-on-MDT. */ - return 0; + if (dev->od_is_ost) + return SCRUB_NEXT_OSTOBJ_OLD; + else + return 0; } -static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev, - struct lu_fid *fid, struct osd_inode_id *lid, __u32 pos, - struct super_block *sb, bool scrub) +static int osd_scrub_get_fid(struct osd_thread_info *info, + struct osd_device *dev, struct inode *inode, + struct lu_fid *fid, bool scrub) { - struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs; - struct inode *inode; - int rc = 0; - bool has_lma = false; - ENTRY; - - osd_id_gen(lid, pos, OSD_OII_NOGEN); - inode = osd_iget(info, dev, lid); - if (IS_ERR(inode)) { - rc = PTR_ERR(inode); - /* The inode may be removed after bitmap searching, or the - * file is new created without inode initialized yet. */ - if (rc == -ENOENT || rc == -ESTALE) - RETURN(SCRUB_NEXT_CONTINUE); - - CERROR("%.16s: fail to read inode, ino# = %u, rc = %d\n", - LDISKFS_SB(sb)->s_es->s_volume_name, pos, rc); - RETURN(rc); - } - - /* If the inode has no OI mapping, then it is special locally used, - * should be invisible to OI scrub or up layer LFSCK. */ - if (ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NO_OI)) - GOTO(put, rc = SCRUB_NEXT_CONTINUE); - - if (scrub && - ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB)) { - /* Only skip it for the first OI scrub accessing. */ - ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB); - GOTO(put, rc = SCRUB_NEXT_NOSCRUB); - } + struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs; + int rc; + bool has_lma = false; rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma); if (rc == 0) { has_lma = true; - if (lma->lma_compat & LMAC_NOT_IN_OI) { - ldiskfs_set_inode_state(inode, - LDISKFS_STATE_LUSTRE_NO_OI); - GOTO(put, rc = SCRUB_NEXT_CONTINUE); - } - - if (fid_is_llog(&lma->lma_self_fid)) - GOTO(put, rc = SCRUB_NEXT_CONTINUE); + if (lma->lma_compat & LMAC_NOT_IN_OI || + lma->lma_incompat & LMAI_AGENT) + return SCRUB_NEXT_CONTINUE; *fid = lma->lma_self_fid; - if (fid_is_internal(&lma->lma_self_fid)) { - if (!scrub) - rc = SCRUB_NEXT_CONTINUE; - GOTO(put, rc); - } - if (!scrub) - GOTO(put, rc); + return 0; - if (fid_is_namespace_visible(fid) && !fid_is_norm(fid)) - GOTO(put, rc); - - if (lma->lma_compat & LMAC_FID_ON_OST || fid_is_last_id(fid)) - GOTO(put, rc = SCRUB_NEXT_OSTOBJ); + if (lma->lma_compat & LMAC_FID_ON_OST) + return SCRUB_NEXT_OSTOBJ; if (fid_is_idif(fid)) - GOTO(put, rc = SCRUB_NEXT_OSTOBJ_OLD); + return SCRUB_NEXT_OSTOBJ_OLD; - if (lma->lma_incompat & LMAI_AGENT) - GOTO(put, rc = SCRUB_NEXT_CONTINUE); + /* For local object. */ + if (fid_is_internal(fid)) + return 0; - /* Here, it may be MDT-object, or may be 2.4 OST-object. - * Fall through. */ + /* For external visible MDT-object with non-normal FID. */ + if (fid_is_namespace_visible(fid) && !fid_is_norm(fid)) + return 0; + + /* For the object with normal FID, it may be MDT-object, + * or may be 2.4 OST-object, need further distinguish. + * Fall through to next section. */ } if (rc == -ENODATA || rc == 0) { @@ -783,7 +841,7 @@ static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev, if (scrub) /* It is old 2.x (x <= 3) or 1.8 OST-object. */ rc = SCRUB_NEXT_OSTOBJ_OLD; - GOTO(put, rc); + return rc; } if (rc > 0) { @@ -794,14 +852,14 @@ static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev, else /* It is 2.4 OST-object. */ rc = SCRUB_NEXT_OSTOBJ_OLD; - GOTO(put, rc); + return rc; } if (rc != -ENODATA) - GOTO(put, rc); + return rc; if (!has_lma) { - if (dev->od_handle_nolma) { + if (dev->od_scrub.os_convert_igif) { lu_igif_build(fid, inode->i_ino, inode->i_generation); if (scrub) @@ -814,7 +872,7 @@ static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev, * generate its FID, ignore it directly. */ rc = SCRUB_NEXT_CONTINUE; } - GOTO(put, rc); + return rc; } /* For OI scrub case only: the object has LMA but has no ff @@ -823,6 +881,40 @@ static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev, rc = osd_scrub_check_local_fldb(info, dev, fid); } + return rc; +} + +static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev, + struct lu_fid *fid, struct osd_inode_id *lid, __u32 pos, + struct super_block *sb, bool scrub) +{ + struct inode *inode; + int rc; + ENTRY; + + osd_id_gen(lid, pos, OSD_OII_NOGEN); + inode = osd_iget(info, dev, lid); + if (IS_ERR(inode)) { + rc = PTR_ERR(inode); + /* The inode may be removed after bitmap searching, or the + * file is new created without inode initialized yet. */ + if (rc == -ENOENT || rc == -ESTALE) + RETURN(SCRUB_NEXT_CONTINUE); + + CERROR("%.16s: fail to read inode, ino# = %u, rc = %d\n", + LDISKFS_SB(sb)->s_es->s_volume_name, pos, rc); + RETURN(rc); + } + + if (scrub && + ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB)) { + /* Only skip it for the first OI scrub accessing. */ + ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB); + GOTO(put, rc = SCRUB_NEXT_NOSCRUB); + } + + rc = osd_scrub_get_fid(info, dev, inode, fid, scrub); + GOTO(put, rc); put: @@ -977,11 +1069,11 @@ next: scrub->os_pos_current = param->gbase + ++(param->offset); wait: - if (it != NULL && it->ooi_waiting && + if (it != NULL && it->ooi_waiting && ooc != NULL && ooc->ooc_pos_preload < scrub->os_pos_current) { spin_lock(&scrub->os_lock); it->ooi_waiting = 0; - cfs_waitq_broadcast(&thread->t_ctl_waitq); + wake_up_all(&thread->t_ctl_waitq); spin_unlock(&scrub->os_lock); } @@ -1119,8 +1211,10 @@ static int osd_otable_it_preload(const struct lu_env *env, ooc->ooc_pos_preload, rc); if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) { + spin_lock(&scrub->os_lock); scrub->os_waiting = 0; - cfs_waitq_broadcast(&scrub->os_thread.t_ctl_waitq); + wake_up_all(&scrub->os_thread.t_ctl_waitq); + spin_unlock(&scrub->os_lock); } RETURN(rc < 0 ? rc : ooc->ooc_cached_items); @@ -1191,7 +1285,7 @@ out: noenv: spin_lock(&scrub->os_lock); thread_set_flags(thread, SVC_STOPPED); - cfs_waitq_broadcast(&thread->t_ctl_waitq); + wake_up_all(&thread->t_ctl_waitq); spin_unlock(&scrub->os_lock); return rc; } @@ -1203,6 +1297,10 @@ typedef int (*scandir_t)(struct osd_thread_info *, struct osd_device *, static int osd_ios_varfid_fill(void *buf, const char *name, int namelen, loff_t offset, __u64 ino, unsigned d_type); +static int osd_ios_lf_fill(void *buf, const char *name, int namelen, + loff_t offset, __u64 ino, unsigned d_type); +static int osd_ios_dl_fill(void *buf, const char *name, int namelen, + loff_t offset, __u64 ino, unsigned d_type); static int osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev, @@ -1219,6 +1317,7 @@ enum osd_lf_flags { OLF_SCAN_SUBITEMS = 0x0001, OLF_HIDE_FID = 0x0002, OLF_SHOW_NAME = 0x0004, + OLF_NO_OI = 0x0008, }; struct osd_lf_map { @@ -1312,6 +1411,19 @@ static const struct osd_lf_map osd_lf_maps[] = { { "LAST_GROUP", { FID_SEQ_LOCAL_FILE, OFD_LAST_GROUP_OID, 0 }, OLF_SHOW_NAME, NULL, NULL }, + /* lost+found */ + { "lost+found", { 0, 0, 0 }, OLF_SCAN_SUBITEMS | OLF_NO_OI, + osd_ios_general_scan, osd_ios_lf_fill }, + + { NULL, { 0, 0, 0 }, 0, NULL, NULL } +}; + +/* Add the new introduced files under .lustre/ in the list in the future. */ +static const struct osd_lf_map osd_dl_maps[] = { + /* .lustre/fid */ + { "fid", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_OBF, 0 }, 0, + NULL, NULL }, + { NULL, { 0, 0, 0 }, 0, NULL, NULL } }; @@ -1333,6 +1445,9 @@ osd_ios_lookup_one_len(const char *name, struct dentry *parent, int namelen) { struct dentry *dentry; + CDEBUG(D_LFSCK, "init lookup one: parent = %.*s, name = %.*s\n", + parent->d_name.len, parent->d_name.name, namelen, name); + dentry = ll_lookup_one_len(name, parent, namelen); if (!IS_ERR(dentry) && dentry->d_inode == NULL) { dput(dentry); @@ -1342,48 +1457,24 @@ osd_ios_lookup_one_len(const char *name, struct dentry *parent, int namelen) return dentry; } -static inline void -osd_ios_llogname2fid(struct lu_fid *fid, const char *name, int namelen) -{ - obd_id id = 0; - int i = 0; - - fid->f_seq = FID_SEQ_LLOG; - while (i < namelen) - id = id * 10 + name[i++] - '0'; - - fid->f_oid = id & 0x00000000ffffffffULL; - fid->f_ver = id >> 32; -} - -static inline void -osd_ios_Oname2fid(struct lu_fid *fid, const char *name, int namelen) -{ - __u64 seq = 0; - int i = 0; - - while (i < namelen) - seq = seq * 10 + name[i++] - '0'; - - lu_last_id_fid(fid, seq); -} - static int osd_ios_new_item(struct osd_device *dev, struct dentry *dentry, scandir_t scandir, filldir_t filldir) { struct osd_ios_item *item; + ENTRY; OBD_ALLOC_PTR(item); if (item == NULL) - return -ENOMEM; + RETURN(-ENOMEM); CFS_INIT_LIST_HEAD(&item->oii_list); item->oii_dentry = dget(dentry); item->oii_scandir = scandir; item->oii_filldir = filldir; cfs_list_add_tail(&item->oii_list, &dev->od_ios_list); - return 0; + + RETURN(0); } /** @@ -1406,6 +1497,8 @@ osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev, int rc; ENTRY; + CDEBUG(D_LFSCK, "init scan one: ino = %ld\n", inode->i_ino); + rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma); if (rc != 0 && rc != -ENODATA) RETURN(rc); @@ -1432,7 +1525,7 @@ osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev, RETURN(rc); rc = osd_scrub_refresh_mapping(info, dev, &tfid, id, - DTO_INDEX_INSERT, 0); + DTO_INDEX_INSERT, true, 0); if (rc > 0) rc = 0; @@ -1452,13 +1545,84 @@ osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev, } rc = osd_scrub_refresh_mapping(info, dev, &tfid, id, - DTO_INDEX_UPDATE, 0); + DTO_INDEX_UPDATE, true, 0); if (rc > 0) rc = 0; RETURN(rc); } +/** + * It scans the /lost+found, and for the OST-object (with filter_fid + * or filter_fid_old), move them back to its proper /O//d. + */ +static int osd_ios_lf_fill(void *buf, const char *name, int namelen, + loff_t offset, __u64 ino, unsigned d_type) +{ + struct osd_ios_filldir_buf *fill_buf = buf; + struct osd_thread_info *info = fill_buf->oifb_info; + struct osd_device *dev = fill_buf->oifb_dev; + struct lu_fid *fid = &info->oti_fid; + struct osd_scrub *scrub = &dev->od_scrub; + struct dentry *parent = fill_buf->oifb_dentry; + struct dentry *child; + struct inode *dir = parent->d_inode; + struct inode *inode; + int rc; + ENTRY; + + /* skip any '.' started names */ + if (name[0] == '.') + RETURN(0); + + scrub->os_lf_scanned++; + child = osd_ios_lookup_one_len(name, parent, namelen); + if (IS_ERR(child)) { + CWARN("%s: cannot lookup child '%.*s': rc = %d\n", + osd_name(dev), namelen, name, (int)PTR_ERR(child)); + RETURN(0); + } + + inode = child->d_inode; + if (S_ISDIR(inode->i_mode)) { + rc = osd_ios_new_item(dev, child, osd_ios_general_scan, + osd_ios_lf_fill); + if (rc != 0) + CWARN("%s: cannot add child '%.*s': rc = %d\n", + osd_name(dev), namelen, name, rc); + GOTO(put, rc); + } + + if (!S_ISREG(inode->i_mode)) + GOTO(put, rc = 0); + + rc = osd_scrub_get_fid(info, dev, inode, fid, true); + if (rc == SCRUB_NEXT_OSTOBJ || rc == SCRUB_NEXT_OSTOBJ_OLD) { + rc = osd_obj_map_recover(info, dev, dir, child, fid); + if (rc == 0) { + CDEBUG(D_LFSCK, "recovered '%.*s' ["DFID"] from " + "/lost+found.\n", namelen, name, PFID(fid)); + scrub->os_lf_repaired++; + } else { + CWARN("%s: cannot rename for '%.*s' "DFID": rc = %d\n", + osd_name(dev), namelen, name, PFID(fid), rc); + } + } + + /* XXX: For MDT-objects, we can move them from /lost+found to namespace + * visible place, such as the /ROOT/.lustre/lost+found, then LFSCK + * can process them in furtuer. */ + + GOTO(put, rc); + +put: + if (rc < 0) + scrub->os_lf_failed++; + dput(child); + /* skip the failure to make the scanning to continue. */ + return 0; +} + static int osd_ios_varfid_fill(void *buf, const char *name, int namelen, loff_t offset, __u64 ino, unsigned d_type) { @@ -1486,6 +1650,42 @@ static int osd_ios_varfid_fill(void *buf, const char *name, int namelen, RETURN(rc); } +static int osd_ios_dl_fill(void *buf, const char *name, int namelen, + loff_t offset, __u64 ino, unsigned d_type) +{ + struct osd_ios_filldir_buf *fill_buf = buf; + struct osd_device *dev = fill_buf->oifb_dev; + const struct osd_lf_map *map; + struct dentry *child; + int rc = 0; + ENTRY; + + /* skip any '.' started names */ + if (name[0] == '.') + RETURN(0); + + for (map = osd_dl_maps; map->olm_name != NULL; map++) { + if (strlen(map->olm_name) != namelen) + continue; + + if (strncmp(map->olm_name, name, namelen) == 0) + break; + } + + if (map->olm_name == NULL) + RETURN(0); + + child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen); + if (IS_ERR(child)) + RETURN(PTR_ERR(child)); + + rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode, + &map->olm_fid, map->olm_flags); + dput(child); + + RETURN(rc); +} + static int osd_ios_root_fill(void *buf, const char *name, int namelen, loff_t offset, __u64 ino, unsigned d_type) { @@ -1515,8 +1715,9 @@ static int osd_ios_root_fill(void *buf, const char *name, int namelen, if (IS_ERR(child)) RETURN(PTR_ERR(child)); - rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode, - &map->olm_fid, map->olm_flags); + if (!(map->olm_flags & OLF_NO_OI)) + rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode, + &map->olm_fid, map->olm_flags); if (rc == 0 && map->olm_flags & OLF_SCAN_SUBITEMS) rc = osd_ios_new_item(dev, child, map->olm_scandir, map->olm_filldir); @@ -1544,6 +1745,7 @@ osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev, filp->f_mapping = inode->i_mapping; filp->f_op = fops; filp->private_data = NULL; + set_file_inode(filp, inode); rc = fops->readdir(filp, &buf, filldir); fops->release(inode, filp); @@ -1574,7 +1776,7 @@ osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev, * and try to re-generate the LMA from the OI mapping. But if the * OI mapping crashed or lost also, then we have to give up under * double failure cases. */ - dev->od_handle_nolma = 1; + scrub->os_convert_igif = 1; child = osd_ios_lookup_one_len(dot_lustre_name, dentry, strlen(dot_lustre_name)); if (IS_ERR(child)) { @@ -1585,6 +1787,7 @@ osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev, osd_scrub_file_reset(scrub, LDISKFS_SB(osd_sb(dev))->s_es->s_uuid, SF_UPGRADE); + sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID; rc = osd_scrub_file_store(scrub); } else { rc = 0; @@ -1616,6 +1819,9 @@ osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev, * to the solution 2). */ rc = osd_ios_scan_one(info, dev, child->d_inode, &LU_DOT_LUSTRE_FID, 0); + if (rc == 0) + rc = osd_ios_new_item(dev, child, osd_ios_general_scan, + osd_ios_dl_fill); dput(child); } @@ -1626,10 +1832,19 @@ static int osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev, struct dentry *dentry, filldir_t filldir) { - struct dentry *child; - int rc; + struct osd_scrub *scrub = &dev->od_scrub; + struct scrub_file *sf = &scrub->os_file; + struct dentry *child; + int rc; ENTRY; + if (unlikely(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID)) { + sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID; + rc = osd_scrub_file_store(scrub); + if (rc != 0) + RETURN(rc); + } + child = osd_ios_lookup_one_len(ADMIN_USR, dentry, strlen(ADMIN_USR)); if (!IS_ERR(child)) { rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0); @@ -1666,6 +1881,9 @@ static int osd_initial_OI_scrub(struct osd_thread_info *info, int rc; ENTRY; + /* Lookup IGIF in OI by force for initial OI scrub. */ + dev->od_igif_inoi = 1; + while (1) { rc = scandir(info, dev, dentry, filldir); if (item != NULL) { @@ -1718,7 +1936,8 @@ static int osd_initial_OI_scrub(struct osd_thread_info *info, dput(child); else if (PTR_ERR(child) == -ENOENT) osd_scrub_refresh_mapping(info, dev, &map->olm_fid, - NULL, DTO_INDEX_DELETE, 0); + NULL, DTO_INDEX_DELETE, + true, 0); map++; } @@ -1811,7 +2030,7 @@ static void do_osd_scrub_stop(struct osd_scrub *scrub) if (!thread_is_init(thread) && !thread_is_stopped(thread)) { thread_set_flags(thread, SVC_STOPPING); spin_unlock(&scrub->os_lock); - cfs_waitq_broadcast(&thread->t_ctl_waitq); + wake_up_all(&thread->t_ctl_waitq); l_wait_event(thread->t_ctl_waitq, thread_is_stopped(thread), &lwi); @@ -1857,37 +2076,50 @@ int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev) ctxt->pwd = dev->od_mnt->mnt_root; ctxt->fs = get_ds(); - cfs_waitq_init(&scrub->os_thread.t_ctl_waitq); + init_waitqueue_head(&scrub->os_thread.t_ctl_waitq); init_rwsem(&scrub->os_rwsem); spin_lock_init(&scrub->os_lock); CFS_INIT_LIST_HEAD(&scrub->os_inconsistent_items); - push_ctxt(&saved, ctxt, NULL); + push_ctxt(&saved, ctxt); filp = filp_open(osd_scrub_name, O_RDWR | O_CREAT, 0644); if (IS_ERR(filp)) { - pop_ctxt(&saved, ctxt, NULL); + pop_ctxt(&saved, ctxt); RETURN(PTR_ERR(filp)); } inode = filp->f_dentry->d_inode; - ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NO_OI); /* 'What the @fid is' is not imporatant, because the object * has no OI mapping, and only is visible inside the OSD.*/ lu_igif_build(fid, inode->i_ino, inode->i_generation); rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0); if (rc != 0) { filp_close(filp, 0); - pop_ctxt(&saved, ctxt, NULL); + pop_ctxt(&saved, ctxt); RETURN(rc); } scrub->os_inode = igrab(inode); filp_close(filp, 0); - pop_ctxt(&saved, ctxt, NULL); + pop_ctxt(&saved, ctxt); rc = osd_scrub_file_load(scrub); if (rc == -ENOENT) { osd_scrub_file_init(scrub, es->s_uuid); + /* If the "/O" dir does not exist when mount (indicated by + * osd_device::od_maybe_new), neither for the "/OI_scrub", + * then it is quite probably that the device is a new one, + * under such case, mark it as SIF_NO_HANDLE_OLD_FID. + * + * For the rare case that "/O" and "OI_scrub" both lost on + * an old device, it can be found and cleared later. + * + * For the system with "SIF_NO_HANDLE_OLD_FID", we do not + * need to check "filter_fid_old" and to convert it to + * "filter_fid" for each object, and all the IGIF should + * have their FID mapping in OI files already. */ + if (dev->od_maybe_new) + sf->sf_internal_flags = SIF_NO_HANDLE_OLD_FID; dirty = 1; } else if (rc != 0) { RETURN(rc); @@ -1919,12 +2151,17 @@ int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev) rc = osd_initial_OI_scrub(info, dev); if (rc == 0) { - if ((sf->sf_flags & SF_UPGRADE) && - !(sf->sf_flags & SF_INCONSISTENT)) - /* The 'od_igif_inoi' will be set after the - * upgrading completed, needs NOT remount. */ + if (sf->sf_flags & SF_UPGRADE || + !(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID || + sf->sf_success_count > 0)) { dev->od_igif_inoi = 0; - else + dev->od_check_ff = dev->od_is_ost; + } else { + dev->od_igif_inoi = 1; + dev->od_check_ff = 0; + } + + if (sf->sf_flags & SF_INCONSISTENT) /* The 'od_igif_inoi' will be set under the * following cases: * 1) new created system, or @@ -2011,6 +2248,13 @@ static struct dt_it *osd_otable_it_init(const struct lu_env *env, start |= SS_CLEAR_FAILOUT; } + if (valid & DOIV_DRYRUN) { + if (flags & DOIF_DRYRUN) + start |= SS_SET_DRYRUN; + else + start |= SS_CLEAR_DRYRUN; + } + rc = do_osd_scrub_start(dev, start); if (rc < 0 && rc != -EALREADY) { dev->od_otable_it = NULL; @@ -2101,7 +2345,7 @@ again: if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) { spin_lock(&scrub->os_lock); scrub->os_waiting = 0; - cfs_waitq_broadcast(&scrub->os_thread.t_ctl_waitq); + wake_up_all(&scrub->os_thread.t_ctl_waitq); spin_unlock(&scrub->os_lock); } @@ -2189,7 +2433,7 @@ static int osd_otable_it_load(const struct lu_env *env, it->ooi_user_ready = 1; if (!scrub->os_full_speed) - cfs_waitq_broadcast(&scrub->os_thread.t_ctl_waitq); + wake_up_all(&scrub->os_thread.t_ctl_waitq); /* Unplug OSD layer iteration by the first next() call. */ rc = osd_otable_it_next(env, (struct dt_it *)it); @@ -2251,7 +2495,7 @@ int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic, spin_unlock(&scrub->os_lock); if (wakeup != 0) - cfs_waitq_broadcast(&thread->t_ctl_waitq); + wake_up_all(&thread->t_ctl_waitq); RETURN(0); } @@ -2299,6 +2543,7 @@ static const char *scrub_flags_names[] = { static const char *scrub_param_names[] = { "failout", + "dryrun", NULL }; @@ -2448,7 +2693,7 @@ int osd_scrub_dump(struct osd_device *dev, char *buf, int len) if (thread_is_running(&scrub->os_thread)) { cfs_duration_t duration = cfs_time_current() - scrub->os_time_last_checkpoint; - __u64 new_checked = scrub->os_new_checked * CFS_HZ; + __u64 new_checked = scrub->os_new_checked * HZ; __u32 rtime = sf->sf_run_time + cfs_duration_sec(duration + HALF_SEC); @@ -2460,8 +2705,13 @@ int osd_scrub_dump(struct osd_device *dev, char *buf, int len) "run_time: %u seconds\n" "average_speed: "LPU64" objects/sec\n" "real-time_speed: "LPU64" objects/sec\n" - "current_position: %u\n", - rtime, speed, new_checked, scrub->os_pos_current); + "current_position: %u\n" + "lf_scanned: "LPU64"\n" + "lf_reparied: "LPU64"\n" + "lf_failed: "LPU64"\n", + rtime, speed, new_checked, scrub->os_pos_current, + scrub->os_lf_scanned, scrub->os_lf_repaired, + scrub->os_lf_failed); } else { if (sf->sf_run_time != 0) do_div(speed, sf->sf_run_time); @@ -2469,8 +2719,12 @@ int osd_scrub_dump(struct osd_device *dev, char *buf, int len) "run_time: %u seconds\n" "average_speed: "LPU64" objects/sec\n" "real-time_speed: N/A\n" - "current_position: N/A\n", - sf->sf_run_time, speed); + "current_position: N/A\n" + "lf_scanned: "LPU64"\n" + "lf_reparied: "LPU64"\n" + "lf_failed: "LPU64"\n", + sf->sf_run_time, speed, scrub->os_lf_scanned, + scrub->os_lf_repaired, scrub->os_lf_failed); } if (rc <= 0) goto out;