X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fosd-ldiskfs%2Fosd_scrub.c;h=d830e226a3ef465e1e8dbaff211b1ab7a2980a45;hp=75b0b8a7fe360f21dfd0f416a2a16938c6f59e03;hb=76031fc50abd3a21395e6f4c09e4d9ab6c60354b;hpb=64af83190180a8f3e90aea270e2baaecb1ecdb93 diff --git a/lustre/osd-ldiskfs/osd_scrub.c b/lustre/osd-ldiskfs/osd_scrub.c index 75b0b8a..d830e22 100644 --- a/lustre/osd-ldiskfs/osd_scrub.c +++ b/lustre/osd-ldiskfs/osd_scrub.c @@ -20,7 +20,7 @@ * GPL HEADER END */ /* - * Copyright (c) 2012, Intel Corporation. + * Copyright (c) 2012, 2013, Intel Corporation. */ /* * lustre/osd-ldiskfs/osd_scrub.c @@ -35,20 +35,18 @@ * Author: Fan Yong */ -#ifndef EXPORT_SYMTAB -# define EXPORT_SYMTAB -#endif #define DEBUG_SUBSYSTEM S_MDS #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 @@ -60,6 +58,9 @@ #define SCRUB_NEXT_FATAL 6 /* simulate failure during OI scrub */ #define SCRUB_NEXT_NOSCRUB 7 /* new created object, no scrub on it */ #define SCRUB_NEXT_NOLMA 8 /* the inode has no FID-in-LMA */ +#define SCRUB_NEXT_OSTOBJ 9 /* for OST-object */ +#define SCRUB_NEXT_OSTOBJ_OLD 10 /* old OST-object, no LMA or no FID-on-OST + * flags in LMA */ /* misc functions */ @@ -79,46 +80,49 @@ static inline int osd_scrub_has_window(struct osd_scrub *scrub, return scrub->os_pos_current < ooc->ooc_pos_preload + SCRUB_WINDOW_SIZE; } +/** + * update/insert/delete the specified OI mapping (@fid @id) according to the ops + * + * \retval 1, changed nothing + * \retval 0, changed successfully + * \retval -ve, on error + */ 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) + const struct osd_inode_id *id, + int ops, bool force, + enum oi_check_flags flags) { - struct lu_fid *oi_fid = &info->oti_fid2; - struct osd_inode_id *oi_id = &info->oti_id2; - struct iam_container *bag; - struct iam_path_descr *ipd; - handle_t *jh; - int rc; + handle_t *th; + int rc; ENTRY; - fid_cpu_to_be(oi_fid, fid); - osd_id_pack(oi_id, id); - jh = ldiskfs_journal_start_sb(osd_sb(dev), - osd_dto_credits_noquota[ops]); - if (IS_ERR(jh)) { - rc = PTR_ERR(jh); - CERROR("%.16s: fail to start trans for scrub store: rc = %d\n", - LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, rc); - RETURN(rc); - } + if (dev->od_scrub.os_file.sf_param & SP_DRYRUN && !force) + RETURN(0); - bag = &osd_fid2oi(dev, fid)->oi_dir.od_container; - ipd = osd_idx_ipd_get(info->oti_env, bag); - if (unlikely(ipd == NULL)) { - ldiskfs_journal_stop(jh); - CERROR("%.16s: fail to get ipd for scrub store\n", - LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name); - RETURN(-ENOMEM); + /* 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); } - if (ops == DTO_INDEX_UPDATE) { - rc = iam_update(jh, bag, (const struct iam_key *)oi_fid, - (struct iam_rec *)oi_id, ipd); - } else { - rc = iam_insert(jh, bag, (const struct iam_key *)oi_fid, - (struct iam_rec *)oi_id, ipd); - if (rc == -EEXIST) { + switch (ops) { + case DTO_INDEX_UPDATE: + rc = osd_oi_update(info, dev, fid, id, th, flags); + if (unlikely(rc == -ENOENT)) { + /* Some unlink thread may removed the OI mapping. */ + rc = 1; + } + break; + case DTO_INDEX_INSERT: + rc = osd_oi_insert(info, dev, fid, id, th, flags); + if (unlikely(rc == -EEXIST)) { rc = 1; /* XXX: There are trouble things when adding OI * mapping for IGIF object, which may cause @@ -152,10 +156,22 @@ static int osd_scrub_refresh_mapping(struct osd_thread_info *info, * * Anyway, it is rare, only exists in theory. */ } + break; + case DTO_INDEX_DELETE: + rc = osd_oi_delete(info, dev, fid, th, flags); + if (rc == -ENOENT) { + /* It is normal that the unlink thread has removed the + * OI mapping already. */ + rc = 1; + } + break; + default: + LASSERTF(0, "Unexpected ops %d\n", ops); + break; } - osd_ipd_put(info->oti_env, bag, ipd); - ldiskfs_journal_stop(jh); - RETURN(rc); + + ldiskfs_journal_stop(th); + return rc; } /* OI_scrub file ops */ @@ -191,6 +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); } @@ -225,6 +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); } @@ -246,7 +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; @@ -306,7 +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", @@ -329,61 +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 (flags & SS_CLEAR_FAILOUT) - sf->sf_param &= ~SP_FAILOUT; - - if (flags & SS_RESET) - osd_scrub_file_reset(scrub, - LDISKFS_SB(osd_sb(dev))->s_es->s_uuid, 0); + if (dev->od_scrub.os_file.sf_param & SP_DRYRUN) + RETURN(0); - if (flags & SS_AUTO) { - scrub->os_full_speed = 1; - sf->sf_flags |= SF_AUTO; - } else { - scrub->os_full_speed = 0; + /* 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 (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE)) - scrub->os_full_speed = 1; + /* 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); - 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; + removed = true; + } else if (unlikely(rc == -ENODATA)) { + reset = false; + } else if (rc != sizeof(struct filter_fid)) { + GOTO(stop, rc = -EINVAL); + } - 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); + /* 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); + + /* 4) generate new XATTR_NAME_FID with the saved parent FID and add it*/ + if (size > 0) { + int rc1; + + 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 @@ -400,6 +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); @@ -417,68 +455,112 @@ osd_scrub_check_update(struct osd_thread_info *info, struct osd_device *dev, if (fid_is_igif(fid)) sf->sf_items_igif++; + if (val == SCRUB_NEXT_OSTOBJ_OLD) { + inode = osd_iget(info, dev, lid); + if (IS_ERR(inode)) { + rc = PTR_ERR(inode); + /* Someone removed the inode. */ + if (rc == -ENOENT || rc == -ESTALE) + rc = 0; + GOTO(out, rc); + } + + sf->sf_flags |= SF_UPGRADE; + 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)) goto iget; - /* XXX: Currently, no FID-in-LMA for OST object, so osd_oi_lookup() - * without checking FLD is enough. - * - * It should be updated if FID-in-LMA for OSD object introduced - * in the future. */ - rc = osd_oi_lookup(info, dev, fid, lid2, false); + rc = osd_oi_lookup(info, dev, fid, lid2, + (val == SCRUB_NEXT_OSTOBJ || + val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0); if (rc != 0) { - if (rc != -ENOENT) + if (rc != -ENOENT && rc != -ESTALE) GOTO(out, rc); iget: - inode = osd_iget(info, dev, lid); - if (IS_ERR(inode)) { - rc = PTR_ERR(inode); - /* Someone removed the inode. */ - if (rc == -ENOENT || rc == -ESTALE) - rc = 0; - GOTO(out, rc); - } - - /* Prevent the inode to be unlinked during OI scrub. */ - mutex_lock(&inode->i_mutex); - if (unlikely(inode->i_nlink == 0)) { - mutex_unlock(&inode->i_mutex); - iput(inode); - GOTO(out, rc = 0); + if (inode == NULL) { + inode = osd_iget(info, dev, lid); + if (IS_ERR(inode)) { + rc = PTR_ERR(inode); + /* Someone removed the inode. */ + if (rc == -ENOENT || rc == -ESTALE) + rc = 0; + GOTO(out, rc); + } } + scrub->os_full_speed = 1; ops = DTO_INDEX_INSERT; idx = osd_oi_fid2idx(dev, fid); - if (val == SCRUB_NEXT_NOLMA) { + switch (val) { + case SCRUB_NEXT_NOLMA: sf->sf_flags |= SF_UPGRADE; - scrub->os_full_speed = 1; - rc = osd_ea_fid_set(info, inode, fid, 0); - if (rc != 0) - GOTO(out, rc); - } else { - sf->sf_flags |= SF_RECREATED | SF_INCONSISTENT; - scrub->os_full_speed = 1; + 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; + break; + case SCRUB_NEXT_OSTOBJ: + sf->sf_flags |= SF_INCONSISTENT; + case SCRUB_NEXT_OSTOBJ_OLD: + break; + default: + sf->sf_flags |= SF_RECREATED; if (unlikely(!ldiskfs_test_bit(idx, sf->sf_oi_bitmap))) ldiskfs_set_bit(idx, sf->sf_oi_bitmap); + break; } } else if (osd_id_eq(lid, lid2)) { + if (converted) + sf->sf_items_updated++; + GOTO(out, rc = 0); } else { - sf->sf_flags |= SF_INCONSISTENT; scrub->os_full_speed = 1; + sf->sf_flags |= SF_INCONSISTENT; + + /* XXX: If the device is restored from file-level backup, then + * some IGIFs may have been already in OI files, and some + * may be not yet. Means upgrading from 1.8 may be partly + * processed, but some clients may hold some immobilized + * IGIFs, and use them to access related objects. Under + * such case, OSD does not know whether an given IGIF has + * been processed or to be processed, and it also cannot + * generate local ino#/gen# directly from the immobilized + * IGIF because of the backup/restore. Then force OSD to + * lookup the given IGIF in OI files, and if no entry, + * then ask the client to retry after upgrading completed. + * No better choice. */ + 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) { if (scrub->os_in_prior) sf->sf_items_updated_prior++; else sf->sf_items_updated++; + + /* The target has been changed, need to be re-loaded. */ + lu_object_purge(info->oti_env, osd2lu_dev(dev), fid); } GOTO(out, rc); @@ -493,12 +575,20 @@ out: rc = 0; } - if (ops == DTO_INDEX_INSERT) { - mutex_unlock(&inode->i_mutex); - iput(inode); - } + /* There may be conflict unlink during the OI scrub, + * if happend, then remove the new added OI mapping. */ + 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, false, + (val == SCRUB_NEXT_OSTOBJ || + val == SCRUB_NEXT_OSTOBJ_OLD) ? + OI_KNOWN_ON_OST : 0); up_write(&scrub->os_rwsem); + if (inode != NULL && !IS_ERR(inode)) + iput(inode); + if (oii != NULL) { LASSERT(!cfs_list_empty(&oii->oii_list)); @@ -510,6 +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; @@ -549,10 +711,17 @@ static void osd_scrub_post(struct osd_scrub *scrub, int result) } sf->sf_time_last_checkpoint = cfs_time_current_sec(); if (result > 0) { + struct osd_device *dev = + 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) { @@ -610,13 +779,118 @@ static int osd_iit_next(struct osd_iit_param *param, __u32 *pos) } } +/** + * \retval SCRUB_NEXT_OSTOBJ_OLD: FID-on-OST + * \retval 0: FID-on-MDT + */ +static int osd_scrub_check_local_fldb(struct osd_thread_info *info, + struct osd_device *dev, + struct lu_fid *fid) +{ + /* XXX: The initial OI scrub will scan the top level /O to generate + * 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. */ + if (dev->od_is_ost) + return SCRUB_NEXT_OSTOBJ_OLD; + else + return 0; +} + +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; + 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 || + lma->lma_incompat & LMAI_AGENT) + return SCRUB_NEXT_CONTINUE; + + *fid = lma->lma_self_fid; + if (!scrub) + return 0; + + if (lma->lma_compat & LMAC_FID_ON_OST) + return SCRUB_NEXT_OSTOBJ; + + if (fid_is_idif(fid)) + return SCRUB_NEXT_OSTOBJ_OLD; + + /* For local object. */ + if (fid_is_internal(fid)) + return 0; + + /* 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) { + rc = osd_get_idif(info, inode, &info->oti_obj_dentry, fid); + if (rc == 0) { + if (scrub) + /* It is old 2.x (x <= 3) or 1.8 OST-object. */ + rc = SCRUB_NEXT_OSTOBJ_OLD; + return rc; + } + + if (rc > 0) { + if (!has_lma) + /* It is FID-on-OST, but we do not know how + * to generate its FID, ignore it directly. */ + rc = SCRUB_NEXT_CONTINUE; + else + /* It is 2.4 OST-object. */ + rc = SCRUB_NEXT_OSTOBJ_OLD; + return rc; + } + + if (rc != -ENODATA) + return rc; + + if (!has_lma) { + if (dev->od_scrub.os_convert_igif) { + lu_igif_build(fid, inode->i_ino, + inode->i_generation); + if (scrub) + rc = SCRUB_NEXT_NOLMA; + else + rc = 0; + } else { + /* It may be FID-on-OST, or may be FID for + * non-MDT0, anyway, we do not know how to + * generate its FID, ignore it directly. */ + rc = SCRUB_NEXT_CONTINUE; + } + return rc; + } + + /* For OI scrub case only: the object has LMA but has no ff + * (or ff crashed). It may be MDT-object, may be OST-object + * with crashed ff. The last check is local FLDB. */ + 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 lustre_mdt_attrs *lma = &info->oti_mdt_attrs; - struct inode *inode; - int rc; + struct inode *inode; + int rc; + ENTRY; osd_id_gen(lid, pos, OSD_OII_NOGEN); inode = osd_iget(info, dev, lid); @@ -625,42 +899,25 @@ static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev, /* 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; + 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)) { - iput(inode); - return SCRUB_NEXT_CONTINUE; + 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); - iput(inode); - return SCRUB_NEXT_NOSCRUB; + GOTO(put, rc = SCRUB_NEXT_NOSCRUB); } - rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma); - if (rc == 0) { - if (fid_is_llog(&lma->lma_self_fid) || - (!scrub && fid_is_internal(&lma->lma_self_fid))) - rc = SCRUB_NEXT_CONTINUE; - else - *fid = lma->lma_self_fid; - } else if (rc == -ENODATA) { - lu_igif_build(fid, inode->i_ino, inode->i_generation); - if (scrub) - rc = SCRUB_NEXT_NOLMA; - else - rc = 0; - } + rc = osd_scrub_get_fid(info, dev, inode, fid, scrub); + + GOTO(put, rc); + +put: iput(inode); return rc; } @@ -812,11 +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); } @@ -924,18 +1181,10 @@ static int osd_inode_iteration(struct osd_thread_info *info, brelse(param.bitmap); RETURN(rc); } - - if (preload && dev->od_otable_it->ooi_stopping) { - brelse(param.bitmap); - RETURN(0); - } } next_group: brelse(param.bitmap); - - if (preload && dev->od_otable_it->ooi_stopping) - RETURN(0); } if (*pos > limit) @@ -962,8 +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); @@ -979,8 +1230,7 @@ static int osd_scrub_main(void *args) int rc; ENTRY; - cfs_daemonize("OI_scrub"); - rc = lu_env_init(&env, LCT_DT_THREAD); + rc = lu_env_init(&env, LCT_LOCAL); if (rc != 0) { CERROR("%.16s: OI scrub, fail to init env, rc = %d\n", LDISKFS_SB(sb)->s_es->s_volume_name, rc); @@ -1005,7 +1255,6 @@ static int osd_scrub_main(void *args) if (unlikely(!thread_is_running(thread))) GOTO(post, rc = 0); - LASSERT(scrub->os_pos_current >= ooc->ooc_pos_preload); scrub->os_pos_current = ooc->ooc_pos_preload; } @@ -1036,7 +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; } @@ -1048,6 +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, @@ -1064,6 +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 { @@ -1090,16 +1344,12 @@ static const struct osd_lf_map osd_lf_maps[] = { osd_ios_general_scan, osd_ios_varfid_fill }, /* PENDING */ - { "PENDING", { FID_SEQ_LOCAL_FILE, MDD_ORPHAN_OID, 0 }, 0, NULL, NULL }, + { "PENDING", { 0, 0, 0 }, 0, NULL, NULL }, /* ROOT */ - { "ROOT", { FID_SEQ_LOCAL_FILE, MDD_ROOT_INDEX_OID, 0 }, + { "ROOT", { FID_SEQ_ROOT, 1, 0 }, OLF_SCAN_SUBITEMS | OLF_HIDE_FID, osd_ios_ROOT_scan, NULL }, - /* capa_keys */ - { CAPA_KEYS, { FID_SEQ_LOCAL_FILE, MDD_CAPA_KEYS_OID, 0 }, 0, - NULL, NULL }, - /* changelog_catalog */ { CHANGELOG_CATALOG, { 0, 0, 0 }, 0, NULL, NULL }, @@ -1115,8 +1365,7 @@ static const struct osd_lf_map osd_lf_maps[] = { NULL, NULL }, /* lfsck_bookmark */ - { "lfsck_bookmark", { FID_SEQ_LOCAL_FILE, LFSCK_BOOKMARK_OID, 0 }, 0, - NULL, NULL }, + { "lfsck_bookmark", { 0, 0, 0 }, 0, NULL, NULL }, /* lov_objid */ { LOV_OBJID, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OID, 0 }, OLF_SHOW_NAME, @@ -1134,10 +1383,6 @@ static const struct osd_lf_map osd_lf_maps[] = { { QSD_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, osd_ios_general_scan, osd_ios_varfid_fill }, - /* seq-200000003-lastid */ - { "seq-200000003-lastid", { FID_SEQ_LOCAL_NAME, 1, 0 }, 0, - NULL, NULL }, - /* seq_ctl */ { "seq_ctl", { FID_SEQ_LOCAL_FILE, FID_SEQ_CTL_OID, 0 }, OLF_SHOW_NAME, NULL, NULL }, @@ -1146,17 +1391,12 @@ static const struct osd_lf_map osd_lf_maps[] = { { "seq_srv", { FID_SEQ_LOCAL_FILE, FID_SEQ_SRV_OID, 0 }, OLF_SHOW_NAME, NULL, NULL }, - /* LAST_GROUP */ - { "LAST_GROUP", { FID_SEQ_LOCAL_FILE, OFD_LAST_GROUP_OID, 0 }, - OLF_SHOW_NAME, NULL, NULL }, - /* health_check */ { HEALTH_CHECK, { FID_SEQ_LOCAL_FILE, OFD_HEALTH_CHECK_OID, 0 }, OLF_SHOW_NAME, NULL, NULL }, /* lfsck_namespace */ - { "lfsck_namespace", { FID_SEQ_LOCAL_FILE, LFSCK_NAMESPACE_OID, 0 }, 0, - NULL, NULL }, + { "lfsck_namespace", { 0, 0, 0 }, 0, NULL, NULL }, /* OBJECTS, upgrade from old device */ { OBJECTS, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, osd_ios_OBJECTS_scan, NULL }, @@ -1167,6 +1407,31 @@ static const struct osd_lf_map osd_lf_maps[] = { /* lquota_v2.group, upgrade from old device */ { "lquota_v2.group", { 0, 0, 0 }, 0, NULL, NULL }, + /* LAST_GROUP, upgrade from old device */ + { "LAST_GROUP", { FID_SEQ_LOCAL_FILE, OFD_LAST_GROUP_OID, 0 }, + OLF_SHOW_NAME, NULL, NULL }, + + /* SLAVE_LOG, llog for destroy slave stripes of striped dir */ + { "SLAVE_LOG", { FID_SEQ_LOCAL_FILE, SLAVE_LLOG_CATALOGS_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 }, + + /* .lustre/lost+found */ + { "lost+found", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_LPF, 0 }, 0, + NULL, NULL }, + { NULL, { 0, 0, 0 }, 0, NULL, NULL } }; @@ -1178,6 +1443,10 @@ struct osd_ios_item { }; struct osd_ios_filldir_buf { +#ifdef HAVE_DIR_CONTEXT + /* please keep it as first member */ + struct dir_context ctx; +#endif struct osd_thread_info *oifb_info; struct osd_device *oifb_dev; struct dentry *oifb_dentry; @@ -1188,6 +1457,9 @@ 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); @@ -1197,48 +1469,24 @@ 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); } /** @@ -1261,6 +1509,8 @@ 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); @@ -1271,20 +1521,26 @@ osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev, lu_igif_build(&tfid, inode->i_ino, inode->i_generation); else tfid = *fid; - rc = osd_ea_fid_set(info, inode, &tfid, 0); + rc = osd_ea_fid_set(info, inode, &tfid, 0, 0); if (rc != 0) RETURN(rc); } else { + if (lma->lma_compat & LMAC_NOT_IN_OI) + RETURN(0); + tfid = lma->lma_self_fid; } - rc = __osd_oi_lookup(info, dev, &tfid, id2); + rc = osd_oi_lookup(info, dev, &tfid, id2, 0); if (rc != 0) { if (rc != -ENOENT) RETURN(rc); rc = osd_scrub_refresh_mapping(info, dev, &tfid, id, - DTO_INDEX_INSERT); + DTO_INDEX_INSERT, true, 0); + if (rc > 0) + rc = 0; + RETURN(rc); } @@ -1300,11 +1556,85 @@ 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_UPDATE); + rc = osd_scrub_refresh_mapping(info, dev, &tfid, id, + 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) { @@ -1332,6 +1662,42 @@ 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) { @@ -1361,8 +1727,9 @@ 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); @@ -1375,7 +1742,13 @@ static int osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev, struct dentry *dentry, filldir_t filldir) { - struct osd_ios_filldir_buf buf = { info, dev, dentry }; + struct osd_ios_filldir_buf buf = { +#ifdef HAVE_DIR_CONTEXT + .ctx.actor = filldir, +#endif + .oifb_info = info, + .oifb_dev = dev, + .oifb_dentry = dentry }; struct file *filp = &info->oti_it_ea.oie_file; struct inode *inode = dentry->d_inode; const struct file_operations *fops = inode->i_fop; @@ -1390,8 +1763,15 @@ 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); +#ifdef HAVE_DIR_CONTEXT + buf.ctx.pos = filp->f_pos; + rc = fops->iterate(filp, &buf.ctx); + filp->f_pos = buf.ctx.pos; +#else rc = fops->readdir(filp, &buf, filldir); +#endif fops->release(inode, filp); RETURN(rc); @@ -1407,8 +1787,20 @@ osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev, int rc; ENTRY; - /* It is existing MDT device. */ - dev->od_handle_nolma = 1; + /* It is existing MDT0 device. We only allow the case of object without + * LMA to happen on the MDT0, which is usually for old 1.8 MDT. Then we + * can generate IGIF mode FID for the object and related OI mapping. If + * it is on other MDTs, then becuase file-level backup/restore, related + * OI mapping may be invalid already, we do not know which is the right + * FID for the object. We only allow IGIF objects to reside on the MDT0. + * + * XXX: For the case of object on non-MDT0 device with neither LMA nor + * "fid" xattr, then something crashed. We cannot re-generate the + * FID directly, instead, the OI scrub will scan the OI structure + * 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. */ + scrub->os_convert_igif = 1; child = osd_ios_lookup_one_len(dot_lustre_name, dentry, strlen(dot_lustre_name)); if (IS_ERR(child)) { @@ -1419,13 +1811,41 @@ 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; } } } else { - rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0); + /* For lustre-2.x (x <= 3), the ".lustre" has NO FID-in-LMA, + * so the client will get IGIF for the ".lustre" object when + * the MDT restart. + * + * From the OI scrub view, when the MDT upgrade to Lustre-2.4, + * it does not know whether there are some old clients cached + * the ".lustre" IGIF during the upgrading. Two choices: + * + * 1) Generate IGIF-in-LMA and IGIF-in-OI for the ".lustre". + * It will allow the old connected clients to access the + * ".lustre" with cached IGIF. But it will cause others + * on the MDT failed to check "fid_is_dot_lustre()". + * + * 2) Use fixed FID {FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE, 0} + * for ".lustre" in spite of whether there are some clients + * cached the ".lustre" IGIF or not. It enables the check + * "fid_is_dot_lustre()" on the MDT, although it will cause + * that the old connected clients cannot access the ".lustre" + * with the cached IGIF. + * + * Usually, it is rare case for the old connected clients + * to access the ".lustre" with cached IGIF. So we prefer + * 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); } @@ -1436,10 +1856,19 @@ 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); @@ -1468,13 +1897,17 @@ osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev, static int osd_initial_OI_scrub(struct osd_thread_info *info, struct osd_device *dev) { - struct osd_ios_item *item = NULL; - scandir_t scandir = osd_ios_general_scan; - filldir_t filldir = osd_ios_root_fill; - struct dentry *dentry = osd_sb(dev)->s_root; - int rc; + struct osd_ios_item *item = NULL; + scandir_t scandir = osd_ios_general_scan; + filldir_t filldir = osd_ios_root_fill; + struct dentry *dentry = osd_sb(dev)->s_root; + const struct osd_lf_map *map = osd_lf_maps; + 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) { @@ -1506,7 +1939,33 @@ static int osd_initial_OI_scrub(struct osd_thread_info *info, OBD_FREE_PTR(item); } - RETURN(rc); + if (rc != 0) + RETURN(rc); + + /* There maybe the case that the object has been removed, but its OI + * mapping is still in the OI file, such as the "CATALOGS" after MDT + * file-level backup/restore. So here cleanup the stale OI mappings. */ + while (map->olm_name != NULL) { + struct dentry *child; + + if (fid_is_zero(&map->olm_fid)) { + map++; + continue; + } + + child = osd_ios_lookup_one_len(map->olm_name, + osd_sb(dev)->s_root, + strlen(map->olm_name)); + if (!IS_ERR(child)) + dput(child); + else if (PTR_ERR(child) == -ENOENT) + osd_scrub_refresh_mapping(info, dev, &map->olm_fid, + NULL, DTO_INDEX_DELETE, + true, 0); + map++; + } + + RETURN(0); } char *osd_lf_fid2name(const struct lu_fid *fid) @@ -1535,6 +1994,7 @@ static int do_osd_scrub_start(struct osd_device *dev, __u32 flags) struct osd_scrub *scrub = &dev->od_scrub; struct ptlrpc_thread *thread = &scrub->os_thread; struct l_wait_info lwi = { 0 }; + struct task_struct *task; int rc; ENTRY; @@ -1558,9 +2018,10 @@ again: scrub->os_start_flags = flags; thread_set_flags(thread, 0); - rc = cfs_create_thread(osd_scrub_main, dev, 0); - if (rc < 0) { - CERROR("%.16s: cannot start iteration thread, rc = %d\n", + task = kthread_run(osd_scrub_main, dev, "OI_scrub"); + if (IS_ERR(task)) { + rc = PTR_ERR(task); + CERROR("%.16s: cannot start iteration thread: rc = %d\n", LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, rc); RETURN(rc); } @@ -1595,7 +2056,7 @@ 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); @@ -1629,6 +2090,8 @@ int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev) struct ldiskfs_super_block *es = LDISKFS_SB(sb)->s_es; struct lvfs_run_ctxt saved; struct file *filp; + struct inode *inode; + struct lu_fid *fid = &info->oti_fid; int dirty = 0; int rc = 0; ENTRY; @@ -1639,25 +2102,50 @@ 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)) + if (IS_ERR(filp)) { + pop_ctxt(&saved, ctxt); RETURN(PTR_ERR(filp)); + } - scrub->os_inode = igrab(filp->f_dentry->d_inode); + inode = filp->f_dentry->d_inode; + /* '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); + RETURN(rc); + } + + scrub->os_inode = igrab(inode); filp_close(filp, 0); - pop_ctxt(&saved, ctxt, NULL); - ldiskfs_set_inode_state(scrub->os_inode, - LDISKFS_STATE_LUSTRE_NO_OI); + 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); @@ -1688,14 +2176,45 @@ int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev) RETURN(rc); rc = osd_initial_OI_scrub(info, dev); - if (rc == 0 && !dev->od_noscrub && - ((sf->sf_status == SS_PAUSED) || - (sf->sf_status == SS_CRASHED && - sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE | - SF_AUTO)) || - (sf->sf_status == SS_INIT && - sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE)))) - rc = osd_scrub_start(dev); + if (rc == 0) { + 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; + 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 + * 2) restored from file-level backup, or + * 3) the upgrading completed. + * + * The 'od_igif_inoi' may be cleared by OI scrub + * later if found that the system is upgrading. */ + dev->od_igif_inoi = 1; + + if (!dev->od_noscrub && + ((sf->sf_status == SS_PAUSED) || + (sf->sf_status == SS_CRASHED && + sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | + SF_UPGRADE | SF_AUTO)) || + (sf->sf_status == SS_INIT && + sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | + SF_UPGRADE)))) + rc = osd_scrub_start(dev); + } + + /* it is possible that dcache entries may keep objects after they are + * deleted by OSD. While it looks safe this can cause object data to + * stay until umount causing failures in tests calculating free space, + * e.g. replay-ost-single. Since those dcache entries are not used + * anymore let's just free them after use here */ + shrink_dcache_sb(sb); RETURN(rc); } @@ -1741,7 +2260,6 @@ static struct dt_it *osd_otable_it_init(const struct lu_env *env, dev->od_otable_it = it; it->ooi_dev = dev; - it->ooi_pid = cfs_curproc_pid(); it->ooi_cache.ooc_consumer_idx = -1; if (flags & DOIF_OUTUSED) it->ooi_used_outside = 1; @@ -1756,6 +2274,13 @@ 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; @@ -1793,30 +2318,8 @@ static int osd_otable_it_get(const struct lu_env *env, return 0; } -/** - * It is hack here: - * - * Sometimes the otable-based iteration driver (LFSCK) may be blocked in OSD - * layer when someone wants to stop/pause the iteration. Under such case, we - * need some mechanism to notify the event and wakeup the blocker. - */ static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di) { - struct osd_otable_it *it = (struct osd_otable_it *)di; - struct osd_device *dev = it->ooi_dev; - - /* od_otable_mutex: prevent curcurrent init/fini */ - mutex_lock(&dev->od_otable_mutex); - if (it->ooi_pid == cfs_curproc_pid()) { - dev->od_scrub.os_paused = 1; - } else { - struct ptlrpc_thread *thread = &dev->od_scrub.os_thread; - - it->ooi_stopping = 1; - if (it->ooi_waiting) - cfs_waitq_broadcast(&thread->t_ctl_waitq); - } - mutex_unlock(&dev->od_otable_mutex); } static inline int @@ -1824,7 +2327,7 @@ osd_otable_it_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it) { spin_lock(&scrub->os_lock); if (it->ooi_cache.ooc_pos_preload < scrub->os_pos_current || - scrub->os_waiting || it->ooi_stopping || + scrub->os_waiting || !thread_is_running(&scrub->os_thread)) it->ooi_waiting = 0; else @@ -1868,7 +2371,7 @@ 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); } @@ -1880,9 +2383,6 @@ again: if (!thread_is_running(thread) && !it->ooi_used_outside) RETURN(1); - if (it->ooi_stopping) - RETURN(0); - rc = osd_otable_it_preload(env, it); if (rc >= 0) goto again; @@ -1926,12 +2426,10 @@ static __u64 osd_otable_it_store(const struct lu_env *env, struct osd_otable_cache *ooc = &it->ooi_cache; __u64 hash; - if (it->ooi_user_ready) - hash = ooc->ooc_pos_preload; - else if (likely(ooc->ooc_consumer_idx != -1)) + if (it->ooi_user_ready && ooc->ooc_consumer_idx != -1) hash = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino; else - hash = 0; + hash = ooc->ooc_pos_preload; return hash; } @@ -1955,16 +2453,13 @@ static int osd_otable_it_load(const struct lu_env *env, if (hash > OSD_OTABLE_MAX_HASH) hash = OSD_OTABLE_MAX_HASH; - /* Skip the one that has been processed last time. */ - if (ooc->ooc_pos_preload > hash) - ooc->ooc_pos_preload = hash; - + ooc->ooc_pos_preload = hash; if (ooc->ooc_pos_preload <= LDISKFS_FIRST_INO(osd_sb(dev))) ooc->ooc_pos_preload = LDISKFS_FIRST_INO(osd_sb(dev)) + 1; 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); @@ -2026,7 +2521,7 @@ 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); } @@ -2074,137 +2569,116 @@ static const char *scrub_flags_names[] = { static const char *scrub_param_names[] = { "failout", + "dryrun", NULL }; -static int scrub_bits_dump(char **buf, int *len, int bits, const char *names[], +static int scrub_bits_dump(struct seq_file *m, int bits, const char *names[], const char *prefix) { - int save = *len; int flag; int rc; int i; - rc = snprintf(*buf, *len, "%s:%c", prefix, bits != 0 ? ' ' : '\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; - rc = snprintf(*buf, *len, "%s%c", names[i], - bits != 0 ? ',' : '\n'); - if (rc <= 0) - return -ENOSPC; - - *buf += rc; - *len -= rc; + rc = seq_printf(m, "%s%c", names[i], + bits != 0 ? ',' : '\n'); + if (rc < 0) + return rc; } } - return save - *len; + return 0; } -static int scrub_time_dump(char **buf, int *len, __u64 time, const char *prefix) +static int scrub_time_dump(struct seq_file *m, __u64 time, const char *prefix) { int rc; if (time != 0) - rc = snprintf(*buf, *len, "%s: "LPU64" seconds\n", prefix, + rc = seq_printf(m, "%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; - - *buf += rc; - *len -= rc; + rc = seq_printf(m, "%s: N/A\n", prefix); return rc; } -static int scrub_pos_dump(char **buf, int *len, __u64 pos, const char *prefix) +static int scrub_pos_dump(struct seq_file *m, __u64 pos, const char *prefix) { int rc; if (pos != 0) - rc = snprintf(*buf, *len, "%s: "LPU64"\n", prefix, pos); + rc = seq_printf(m, "%s: "LPU64"\n", prefix, pos); else - rc = snprintf(*buf, *len, "%s: N/A\n", prefix); - if (rc <= 0) - return -ENOSPC; - - *buf += rc; - *len -= rc; + rc = seq_printf(m, "%s: N/A\n", prefix); return rc; } -int osd_scrub_dump(struct osd_device *dev, char *buf, int len) +int osd_scrub_dump(struct seq_file *m, struct osd_device *dev) { struct osd_scrub *scrub = &dev->od_scrub; struct scrub_file *sf = &scrub->os_file; __u64 checked; __u64 speed; - int save = len; - int ret = -ENOSPC; int rc; down_read(&scrub->os_rwsem); - rc = snprintf(buf, len, - "name: OI_scrub\n" - "magic: 0x%x\n" - "oi_files: %d\n" - "status: %s\n", - sf->sf_magic, (int)sf->sf_oi_count, - scrub_status_names[sf->sf_status]); - if (rc <= 0) + rc = seq_printf(m, "name: OI_scrub\n" + "magic: 0x%x\n" + "oi_files: %d\n" + "status: %s\n", + sf->sf_magic, (int)sf->sf_oi_count, + scrub_status_names[sf->sf_status]); + if (rc < 0) goto out; - buf += rc; - len -= rc; - rc = scrub_bits_dump(&buf, &len, sf->sf_flags, scrub_flags_names, + rc = scrub_bits_dump(m, sf->sf_flags, scrub_flags_names, "flags"); if (rc < 0) goto out; - rc = scrub_bits_dump(&buf, &len, sf->sf_param, scrub_param_names, + rc = scrub_bits_dump(m, sf->sf_param, scrub_param_names, "param"); if (rc < 0) goto out; - rc = scrub_time_dump(&buf, &len, sf->sf_time_last_complete, + rc = scrub_time_dump(m, sf->sf_time_last_complete, "time_since_last_completed"); if (rc < 0) goto out; - rc = scrub_time_dump(&buf, &len, sf->sf_time_latest_start, + rc = scrub_time_dump(m, sf->sf_time_latest_start, "time_since_latest_start"); if (rc < 0) goto out; - rc = scrub_time_dump(&buf, &len, sf->sf_time_last_checkpoint, + rc = scrub_time_dump(m, sf->sf_time_last_checkpoint, "time_since_last_checkpoint"); if (rc < 0) goto out; - rc = scrub_pos_dump(&buf, &len, sf->sf_pos_latest_start, + rc = scrub_pos_dump(m, sf->sf_pos_latest_start, "latest_start_position"); if (rc < 0) goto out; - rc = scrub_pos_dump(&buf, &len, sf->sf_pos_last_checkpoint, + rc = scrub_pos_dump(m, sf->sf_pos_last_checkpoint, "last_checkpoint_position"); if (rc < 0) goto out; - rc = scrub_pos_dump(&buf, &len, sf->sf_pos_first_inconsistent, + rc = scrub_pos_dump(m, sf->sf_pos_first_inconsistent, "first_failure_position"); if (rc < 0) goto out; checked = sf->sf_items_checked + scrub->os_new_checked; - rc = snprintf(buf, len, - "checked: "LPU64"\n" + rc = seq_printf(m, "checked: "LPU64"\n" "updated: "LPU64"\n" "failed: "LPU64"\n" "prior_updated: "LPU64"\n" @@ -2214,16 +2688,14 @@ int osd_scrub_dump(struct osd_device *dev, char *buf, int len) checked, sf->sf_items_updated, sf->sf_items_failed, sf->sf_items_updated_prior, sf->sf_items_noscrub, sf->sf_items_igif, sf->sf_success_count); - if (rc <= 0) + if (rc < 0) goto out; - buf += rc; - len -= rc; speed = checked; 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); @@ -2231,30 +2703,31 @@ int osd_scrub_dump(struct osd_device *dev, char *buf, int len) do_div(new_checked, duration); if (rtime != 0) do_div(speed, rtime); - rc = snprintf(buf, len, - "run_time: %u seconds\n" + rc = seq_printf(m, "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); - rc = snprintf(buf, len, - "run_time: %u seconds\n" + rc = seq_printf(m, "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; - - buf += rc; - len -= rc; - ret = save - len; out: up_read(&scrub->os_rwsem); - return ret; + return (rc < 0 ? -ENOSPC : 0); }