X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fosd-ldiskfs%2Fosd_scrub.c;h=257e18f600a3e00f3b89995a4126943f1f67c665;hb=842bda9c5b41eef9e43dc3e00f05767147611677;hp=a6248e42a49992df5b94e02a32e5281fce9ca022;hpb=c8fd9c3c38a66ab3dd479c48e0168577a4ff960e;p=fs%2Flustre-release.git diff --git a/lustre/osd-ldiskfs/osd_scrub.c b/lustre/osd-ldiskfs/osd_scrub.c index a6248e4..257e18f 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, 2013, Intel Corporation. + * Copyright (c) 2012, 2015, Intel Corporation. */ /* * lustre/osd-ldiskfs/osd_scrub.c @@ -35,20 +35,19 @@ * Author: Fan Yong */ -#ifndef EXPORT_SYMTAB -# define EXPORT_SYMTAB -#endif -#define DEBUG_SUBSYSTEM S_MDS +#define DEBUG_SUBSYSTEM S_LFSCK +#include #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 msecs_to_jiffies(MSEC_PER_SEC >> 1) #define OSD_OTABLE_MAX_HASH 0x00000000ffffffffULL @@ -60,6 +59,9 @@ #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 */ @@ -79,6 +81,11 @@ static inline int osd_scrub_has_window(struct osd_scrub *scrub, return scrub->os_pos_current < ooc->ooc_pos_preload + SCRUB_WINDOW_SIZE; } +static inline const char *osd_scrub2name(struct osd_scrub *scrub) +{ + return LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name; +} + /** * update/insert/delete the specified OI mapping (@fid @id) according to the ops * @@ -89,48 +96,40 @@ static inline int osd_scrub_has_window(struct osd_scrub *scrub, 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) -{ - 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; + const struct osd_inode_id *id, + int ops, bool force, + enum oi_check_flags flags, bool *exist) +{ + 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 = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, + osd_dto_credits_noquota[DTO_INDEX_INSERT]); + if (IS_ERR(th)) { + rc = PTR_ERR(th); + CDEBUG(D_LFSCK, "%s: fail to start trans for scrub op %d " + DFID" => %u/%u: rc = %d\n", osd_name(dev), ops, + PFID(fid), id ? id->oii_ino : -1, id ? id->oii_gen : -1, + rc); + RETURN(rc); } switch (ops) { case DTO_INDEX_UPDATE: - rc = iam_update(jh, bag, (const struct iam_key *)oi_fid, - (struct iam_rec *)oi_id, ipd); + 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 = iam_insert(jh, bag, (const struct iam_key *)oi_fid, - (struct iam_rec *)oi_id, ipd); + rc = osd_oi_insert(info, dev, fid, id, th, flags, exist); if (unlikely(rc == -EEXIST)) { rc = 1; /* XXX: There are trouble things when adding OI @@ -167,7 +166,7 @@ static int osd_scrub_refresh_mapping(struct osd_thread_info *info, } break; case DTO_INDEX_DELETE: - rc = iam_delete(jh, bag, (const struct iam_key *)oi_fid, ipd); + 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. */ @@ -178,8 +177,14 @@ static int osd_scrub_refresh_mapping(struct osd_thread_info *info, LASSERTF(0, "Unexpected ops %d\n", ops); break; } - osd_ipd_put(info->oti_env, bag, ipd); - ldiskfs_journal_stop(jh); + + ldiskfs_journal_stop(th); + if (rc < 0) + CDEBUG(D_LFSCK, "%s: fail to refresh OI map for scrub op %d " + DFID" => %u/%u: rc = %d\n", osd_name(dev), ops, + PFID(fid), id ? id->oii_ino : -1, id ? id->oii_gen : -1, + rc); + RETURN(rc); } @@ -216,6 +221,7 @@ 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); } @@ -250,6 +256,7 @@ 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); } @@ -267,11 +274,14 @@ void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags) { struct scrub_file *sf = &scrub->os_file; - CDEBUG(D_LFSCK, "Reset OI scrub file, flags = "LPX64"\n", flags); + CDEBUG(D_LFSCK, "%.16s: reset OI scrub file, old flags = " + LPX64", add flags = "LPX64"\n", + osd_scrub2name(scrub), sf->sf_flags, flags); + memcpy(sf->sf_uuid, uuid, 16); sf->sf_status = SS_INIT; sf->sf_flags |= flags; - sf->sf_param = 0; + sf->sf_flags &= ~SF_AUTO; sf->sf_run_time = 0; sf->sf_time_latest_start = 0; sf->sf_time_last_checkpoint = 0; @@ -281,7 +291,9 @@ void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags) sf->sf_items_checked = 0; sf->sf_items_updated = 0; sf->sf_items_failed = 0; - sf->sf_items_updated_prior = 0; + if (!scrub->os_in_join) + sf->sf_items_updated_prior = 0; + sf->sf_items_noscrub = 0; sf->sf_items_igif = 0; } @@ -289,7 +301,6 @@ void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags) static int osd_scrub_file_load(struct osd_scrub *scrub) { loff_t pos = 0; - char *name = LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name; int len = sizeof(scrub->os_file_disk); int rc; @@ -299,16 +310,18 @@ static int osd_scrub_file_load(struct osd_scrub *scrub) osd_scrub_file_to_cpu(sf, &scrub->os_file_disk); if (sf->sf_magic != SCRUB_MAGIC_V1) { - CWARN("%.16s: invalid scrub magic 0x%x != 0x%x\n,", - name, sf->sf_magic, SCRUB_MAGIC_V1); + CDEBUG(D_LFSCK, "%.16s: invalid scrub magic " + "0x%x != 0x%x\n", osd_scrub2name(scrub), + sf->sf_magic, SCRUB_MAGIC_V1); /* Process it as new scrub file. */ rc = -ENOENT; } else { rc = 0; } } else if (rc != 0) { - CERROR("%.16s: fail to load scrub file, expected = %d, " - "rc = %d\n", name, len, rc); + CDEBUG(D_LFSCK, "%.16s: fail to load scrub file, " + "expected = %d: rc = %d\n", + osd_scrub2name(scrub), len, rc); if (rc > 0) rc = -EFAULT; } else { @@ -331,11 +344,11 @@ 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", - LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name,rc); + CDEBUG(D_LFSCK, "%.16s: fail to start trans for scrub store: " + "rc = %d\n", osd_scrub2name(scrub), rc); return rc; } @@ -344,71 +357,103 @@ int osd_scrub_file_store(struct osd_scrub *scrub) len, 0, &pos, jh); ldiskfs_journal_stop(jh); if (rc != 0) - CERROR("%.16s: fail to store scrub file, expected = %d, " - "rc = %d\n", - LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name, - len, rc); + CDEBUG(D_LFSCK, "%.16s: fail to store scrub file, " + "expected = %d: rc = %d\n", + osd_scrub2name(scrub), len, rc); + scrub->os_time_last_checkpoint = cfs_time_current(); scrub->os_time_next_checkpoint = scrub->os_time_last_checkpoint + cfs_time_seconds(SCRUB_CHECKPOINT_INTERVAL); 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; + struct lu_fid *tfid = &info->oti_fid; + 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; + if (fid_is_idif(fid) && dev->od_index_in_idif == 0) { + struct ost_id *oi = &info->oti_ostid; - 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; + fid_to_ostid(fid, oi); + ostid_to_fid(tfid, oi, 0); } else { - scrub->os_full_speed = 0; + *tfid = *fid; } - if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE)) - scrub->os_full_speed = 1; + /* 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); + CDEBUG(D_LFSCK, "%s: fail to start trans for convert ff " + DFID": rc = %d\n", osd_name(dev), PFID(tfid), rc); + RETURN(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; + /* 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_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); + removed = true; + } else if (unlikely(rc == -ENODATA)) { + reset = false; + } else if (rc != sizeof(struct filter_fid)) { + GOTO(stop, rc = -EINVAL); } - up_write(&scrub->os_rwsem); - RETURN(rc); + /* 3) make new LMA and add it */ + rc = osd_ea_fid_set(info, inode, tfid, 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; + } + + GOTO(stop, rc); + +stop: + ldiskfs_journal_stop(jh); + if (rc < 0) + CDEBUG(D_LFSCK, "%s: fail to convert ff "DFID": rc = %d\n", + osd_name(dev), PFID(tfid), rc); + return rc; } static int @@ -423,8 +468,9 @@ osd_scrub_check_update(struct osd_thread_info *info, struct osd_device *dev, struct osd_inconsistent_item *oii = NULL; struct inode *inode = NULL; int ops = DTO_INDEX_UPDATE; - int idx; int rc; + bool converted = false; + bool exist = false; ENTRY; down_write(&scrub->os_rwsem); @@ -433,8 +479,8 @@ osd_scrub_check_update(struct osd_thread_info *info, struct osd_device *dev, GOTO(out, rc = val); if (scrub->os_in_prior) - oii = cfs_list_entry(oic, struct osd_inconsistent_item, - oii_cache); + oii = list_entry(oic, struct osd_inconsistent_item, + oii_cache); if (lid->oii_ino < sf->sf_pos_latest_start && oii == NULL) GOTO(out, rc = 0); @@ -442,24 +488,7 @@ 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_NOLMA) && - (!dev->od_handle_nolma || 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); - if (rc != 0) { - if (rc != -ENOENT) - GOTO(out, rc); - -iget: + if (val == SCRUB_NEXT_OSTOBJ_OLD) { inode = osd_iget(info, dev, lid); if (IS_ERR(inode)) { rc = PTR_ERR(inode); @@ -469,34 +498,87 @@ iget: GOTO(out, rc); } - /* Check whether the inode to be unlinked during OI scrub. */ - if (unlikely(inode->i_nlink == 0)) { - iput(inode); + /* The inode has been reused as EA inode, ignore it. */ + if (unlikely(osd_is_ea_inode(inode))) GOTO(out, rc = 0); - } + 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) && + (!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)) { ops = DTO_INDEX_INSERT; - idx = osd_oi_fid2idx(dev, fid); - if (val == 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 iget; + } + + 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) + ops = DTO_INDEX_INSERT; + else if (rc != -ESTALE) + GOTO(out, rc); + +iget: + 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); + } + + /* The inode has been reused as EA inode, ignore it. */ + if (unlikely(osd_is_ea_inode(inode))) + GOTO(out, rc = 0); + } + + if (!scrub->os_partial_scan) + scrub->os_full_speed = 1; + + switch (val) { + case SCRUB_NEXT_NOLMA: + sf->sf_flags |= SF_UPGRADE; + 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; - } else { - sf->sf_flags |= SF_RECREATED; - scrub->os_full_speed = 1; - if (unlikely(!ldiskfs_test_bit(idx, sf->sf_oi_bitmap))) - ldiskfs_set_bit(idx, sf->sf_oi_bitmap); + break; + case SCRUB_NEXT_OSTOBJ: + sf->sf_flags |= SF_INCONSISTENT; + case SCRUB_NEXT_OSTOBJ_OLD: + break; + default: + break; } } else if (osd_id_eq(lid, lid2)) { + if (converted) + sf->sf_items_updated++; + GOTO(out, rc = 0); } else { + if (!scrub->os_partial_scan) + scrub->os_full_speed = 1; + sf->sf_flags |= SF_INCONSISTENT; - scrub->os_full_speed = 1; /* XXX: If the device is restored from file-level backup, then * some IGIFs may have been already in OI files, and some @@ -513,12 +595,23 @@ 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, + &exist); if (rc == 0) { if (scrub->os_in_prior) sf->sf_items_updated_prior++; else sf->sf_items_updated++; + + if (ops == DTO_INDEX_INSERT && val == 0 && !exist) { + int idx = osd_oi_fid2idx(dev, fid); + + sf->sf_flags |= SF_RECREATED; + if (unlikely(!ldiskfs_test_bit(idx, sf->sf_oi_bitmap))) + ldiskfs_set_bit(idx, sf->sf_oi_bitmap); + } } GOTO(out, rc); @@ -533,27 +626,114 @@ out: rc = 0; } - if (ops == DTO_INDEX_INSERT) { - /* There may be conflict unlink during the OI scrub, - * if happend, then remove the new added OI mapping. */ - if (unlikely(inode->i_nlink == 0)) - osd_scrub_refresh_mapping(info, dev, fid, lid, - DTO_INDEX_DELETE); - 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(ldiskfs_test_inode_state(inode, + LDISKFS_STATE_LUSTRE_DESTROY))) + 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, NULL); up_write(&scrub->os_rwsem); - if (oii != NULL) { - LASSERT(!cfs_list_empty(&oii->oii_list)); + if (inode != NULL && !IS_ERR(inode)) + iput(inode); + if (oii != NULL) { spin_lock(&scrub->os_lock); - cfs_list_del_init(&oii->oii_list); + if (likely(!list_empty(&oii->oii_list))) + list_del(&oii->oii_list); spin_unlock(&scrub->os_lock); + OBD_FREE_PTR(oii); } + 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; + + CDEBUG(D_LFSCK, "%.16s: OI scrub prep, flags = 0x%x\n", + osd_scrub2name(scrub), flags); + + down_write(&scrub->os_rwsem); + if (flags & SS_SET_FAILOUT) + sf->sf_param |= SP_FAILOUT; + else if (flags & SS_CLEAR_FAILOUT) + sf->sf_param &= ~SP_FAILOUT; + + if (flags & SS_SET_DRYRUN) { + sf->sf_param |= SP_DRYRUN; + } else 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_FULL) { + scrub->os_full_speed = 1; + scrub->os_partial_scan = 0; + sf->sf_flags |= SF_AUTO; + } else if (flags & SS_AUTO_PARTIAL) { + scrub->os_full_speed = 0; + scrub->os_partial_scan = 1; + sf->sf_flags |= SF_AUTO; + } else if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | + SF_UPGRADE)) { + scrub->os_full_speed = 1; + scrub->os_partial_scan = 0; + } else { + scrub->os_full_speed = 0; + scrub->os_partial_scan = 0; + } + + spin_lock(&scrub->os_lock); + scrub->os_in_prior = 0; + scrub->os_waiting = 0; + scrub->os_paused = 0; + scrub->os_in_join = 0; + scrub->os_full_scrub = 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; + sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1; + 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; @@ -577,11 +757,15 @@ static int osd_scrub_checkpoint(struct osd_scrub *scrub) return rc; } -static void osd_scrub_post(struct osd_scrub *scrub, int result) +static int osd_scrub_post(struct osd_scrub *scrub, int result) { struct scrub_file *sf = &scrub->os_file; + int rc; ENTRY; + CDEBUG(D_LFSCK, "%.16s: OI scrub post, result = %d\n", + osd_scrub2name(scrub), result); + down_write(&scrub->os_rwsem); spin_lock(&scrub->os_lock); thread_set_flags(&scrub->os_thread, SVC_STOPPING); @@ -597,10 +781,13 @@ 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) { @@ -613,14 +800,10 @@ static void osd_scrub_post(struct osd_scrub *scrub, int result) } sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC - scrub->os_time_last_checkpoint); - result = osd_scrub_file_store(scrub); - if (result < 0) - CERROR("%.16s: fail to osd_scrub_post, rc = %d\n", - LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name, - result); + rc = osd_scrub_file_store(scrub); up_write(&scrub->os_rwsem); - EXIT; + RETURN(rc < 0 ? rc : result); } /* iteration engine */ @@ -637,13 +820,13 @@ typedef int (*osd_iit_next_policy)(struct osd_thread_info *info, struct osd_device *dev, struct osd_iit_param *param, struct osd_idmap_cache **oic, - int noslot); + const bool noslot); typedef int (*osd_iit_exec_policy)(struct osd_thread_info *info, struct osd_device *dev, struct osd_iit_param *param, struct osd_idmap_cache *oic, - int *noslot, int rc); + bool *noslot, int rc); static int osd_iit_next(struct osd_iit_param *param, __u32 *pos) { @@ -658,13 +841,124 @@ 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; + + /* Not handle the backend root object and agent parent object. + * They are neither visible to namespace nor have OI mappings. */ + if (unlikely(pos == osd_sb(dev)->s_root->d_inode->i_ino || + pos == osd_remote_parent_ino(dev))) + RETURN(SCRUB_NEXT_CONTINUE); osd_id_gen(lid, pos, OSD_OII_NOGEN); inode = osd_iget(info, dev, lid); @@ -673,50 +967,37 @@ 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; + CDEBUG(D_LFSCK, "%.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; - } + /* It is an EA inode, no OI mapping for it, skip it. */ + if (osd_is_ea_inode(inode)) + 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); - 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)) || - (scrub && (lma->lma_incompat & LMAI_AGENT))) - 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; } static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev, struct osd_iit_param *param, - struct osd_idmap_cache **oic, int noslot) + struct osd_idmap_cache **oic, const bool noslot) { struct osd_scrub *scrub = &dev->od_scrub; struct ptlrpc_thread *thread = &scrub->os_thread; @@ -728,10 +1009,11 @@ static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev, struct l_wait_info lwi; lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val), NULL, NULL); - l_wait_event(thread->t_ctl_waitq, - !cfs_list_empty(&scrub->os_inconsistent_items) || - !thread_is_running(thread), - &lwi); + if (likely(lwi.lwi_timeout > 0)) + l_wait_event(thread->t_ctl_waitq, + !list_empty(&scrub->os_inconsistent_items) || + !thread_is_running(thread), + &lwi); } if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) { @@ -747,17 +1029,24 @@ static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev, if (unlikely(!thread_is_running(thread))) return SCRUB_NEXT_EXIT; - if (!cfs_list_empty(&scrub->os_inconsistent_items)) { - struct osd_inconsistent_item *oii; + if (!list_empty(&scrub->os_inconsistent_items)) { + spin_lock(&scrub->os_lock); + if (likely(!list_empty(&scrub->os_inconsistent_items))) { + struct osd_inconsistent_item *oii; - oii = cfs_list_entry(scrub->os_inconsistent_items.next, - struct osd_inconsistent_item, oii_list); - *oic = &oii->oii_cache; - scrub->os_in_prior = 1; - return 0; + oii = list_entry(scrub->os_inconsistent_items.next, + struct osd_inconsistent_item, oii_list); + spin_unlock(&scrub->os_lock); + + *oic = &oii->oii_cache; + scrub->os_in_prior = 1; + + return 0; + } + spin_unlock(&scrub->os_lock); } - if (noslot != 0) + if (noslot) return SCRUB_NEXT_WAIT; rc = osd_iit_next(param, &scrub->os_pos_current); @@ -774,7 +1063,7 @@ static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev, static int osd_preload_next(struct osd_thread_info *info, struct osd_device *dev, struct osd_iit_param *param, - struct osd_idmap_cache **oic, int noslot) + struct osd_idmap_cache **oic, const bool noslot) { struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache; struct osd_scrub *scrub; @@ -806,7 +1095,7 @@ osd_scrub_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it) { spin_lock(&scrub->os_lock); if (osd_scrub_has_window(scrub, &it->ooi_cache) || - !cfs_list_empty(&scrub->os_inconsistent_items) || + !list_empty(&scrub->os_inconsistent_items) || it->ooi_waiting || !thread_is_running(&scrub->os_thread)) scrub->os_waiting = 0; else @@ -818,7 +1107,7 @@ osd_scrub_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it) static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev, struct osd_iit_param *param, - struct osd_idmap_cache *oic, int *noslot, int rc) + struct osd_idmap_cache *oic, bool *noslot, int rc) { struct l_wait_info lwi = { 0 }; struct osd_scrub *scrub = &dev->od_scrub; @@ -841,13 +1130,15 @@ static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev, } rc = osd_scrub_check_update(info, dev, oic, rc); - if (rc != 0) + if (rc != 0) { + scrub->os_in_prior = 0; return rc; + } rc = osd_scrub_checkpoint(scrub); if (rc != 0) { - CERROR("%.16s: fail to checkpoint, pos = %u, rc = %d\n", - LDISKFS_SB(param->sb)->s_es->s_volume_name, + CDEBUG(D_LFSCK, "%.16s: fail to checkpoint, pos = %u: " + "rc = %d\n", osd_scrub2name(scrub), scrub->os_pos_current, rc); /* Continue, as long as the scrub itself can go ahead. */ } @@ -861,36 +1152,36 @@ 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); } if (scrub->os_full_speed || rc == SCRUB_NEXT_CONTINUE) return 0; - if (osd_scrub_has_window(scrub, ooc)) { - *noslot = 0; + if (ooc != NULL && osd_scrub_has_window(scrub, ooc)) { + *noslot = false; return 0; } - l_wait_event(thread->t_ctl_waitq, - osd_scrub_wakeup(scrub, it), - &lwi); + if (it != NULL) + l_wait_event(thread->t_ctl_waitq, osd_scrub_wakeup(scrub, it), + &lwi); - if (osd_scrub_has_window(scrub, ooc)) - *noslot = 0; + if (ooc != NULL && osd_scrub_has_window(scrub, ooc)) + *noslot = false; else - *noslot = 1; + *noslot = true; return 0; } static int osd_preload_exec(struct osd_thread_info *info, struct osd_device *dev, struct osd_iit_param *param, - struct osd_idmap_cache *oic, int *noslot, int rc) + struct osd_idmap_cache *oic, bool *noslot, int rc) { struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache; @@ -905,22 +1196,182 @@ static int osd_preload_exec(struct osd_thread_info *info, #define SCRUB_IT_ALL 1 #define SCRUB_IT_CRASH 2 +static void osd_scrub_join(struct osd_device *dev, __u32 flags, + bool inconsistent) +{ + struct osd_scrub *scrub = &dev->od_scrub; + struct ptlrpc_thread *thread = &scrub->os_thread; + struct scrub_file *sf = &scrub->os_file; + int rc; + ENTRY; + + LASSERT(!(flags & SS_AUTO_PARTIAL)); + + down_write(&scrub->os_rwsem); + scrub->os_in_join = 1; + if (flags & SS_SET_FAILOUT) + sf->sf_param |= SP_FAILOUT; + else if (flags & SS_CLEAR_FAILOUT) + sf->sf_param &= ~SP_FAILOUT; + + if (flags & SS_SET_DRYRUN) + sf->sf_param |= SP_DRYRUN; + else if (flags & SS_CLEAR_DRYRUN) + sf->sf_param &= ~SP_DRYRUN; + + if (flags & SS_RESET) { + osd_scrub_file_reset(scrub, + LDISKFS_SB(osd_sb(dev))->s_es->s_uuid, + inconsistent ? SF_INCONSISTENT : 0); + sf->sf_status = SS_SCANNING; + } + + if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE)) + scrub->os_full_speed = 1; + else + scrub->os_full_speed = 0; + + if (flags & SS_AUTO_FULL) { + sf->sf_flags |= SF_AUTO; + scrub->os_full_speed = 1; + } + + 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; + + scrub->os_pos_current = sf->sf_pos_latest_start; + sf->sf_time_latest_start = cfs_time_current_sec(); + sf->sf_time_last_checkpoint = sf->sf_time_latest_start; + sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1; + rc = osd_scrub_file_store(scrub); + if (rc != 0) + CDEBUG(D_LFSCK, "%.16s: fail to store scrub file when join " + "the OI scrub: rc = %d\n", osd_scrub2name(scrub), rc); + + spin_lock(&scrub->os_lock); + scrub->os_waiting = 0; + scrub->os_paused = 0; + scrub->os_partial_scan = 0; + scrub->os_in_join = 0; + scrub->os_full_scrub = 0; + spin_unlock(&scrub->os_lock); + wake_up_all(&thread->t_ctl_waitq); + up_write(&scrub->os_rwsem); + + EXIT; +} + static int osd_inode_iteration(struct osd_thread_info *info, struct osd_device *dev, __u32 max, bool preload) { + struct osd_scrub *scrub = &dev->od_scrub; + struct ptlrpc_thread *thread = &scrub->os_thread; + struct scrub_file *sf = &scrub->os_file; osd_iit_next_policy next; osd_iit_exec_policy exec; __u32 *pos; __u32 *count; - struct osd_iit_param param; + struct osd_iit_param param = { NULL }; + struct l_wait_info lwi = { 0 }; __u32 limit; - int noslot = 0; int rc; + bool noslot = true; ENTRY; + param.sb = osd_sb(dev); + if (preload) + goto full; + + while (scrub->os_partial_scan && !scrub->os_in_join) { + struct osd_idmap_cache *oic = NULL; + + rc = osd_scrub_next(info, dev, ¶m, &oic, noslot); + switch (rc) { + case SCRUB_NEXT_EXIT: + RETURN(0); + case SCRUB_NEXT_CRASH: + RETURN(SCRUB_IT_CRASH); + case SCRUB_NEXT_FATAL: + RETURN(-EINVAL); + case SCRUB_NEXT_WAIT: { + struct kstatfs *ksfs = &info->oti_ksfs; + __u64 saved_flags; + + if (dev->od_full_scrub_ratio == OFSR_NEVER || + unlikely(sf->sf_items_updated_prior == 0)) + goto wait; + + if (dev->od_full_scrub_ratio == OFSR_DIRECTLY || + scrub->os_full_scrub) { + osd_scrub_join(dev, SS_AUTO_FULL | SS_RESET, + true); + goto full; + } + + rc = param.sb->s_op->statfs(param.sb->s_root, ksfs); + if (rc == 0) { + __u64 used = ksfs->f_files - ksfs->f_ffree; + + do_div(used, sf->sf_items_updated_prior); + /* If we hit too much inconsistent OI + * mappings during the partial scan, + * then scan the device completely. */ + if (used < dev->od_full_scrub_ratio) { + osd_scrub_join(dev, + SS_AUTO_FULL | SS_RESET, true); + goto full; + } + } + +wait: + if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && + cfs_fail_val > 0) + continue; + + saved_flags = sf->sf_flags; + sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT | + SF_UPGRADE | SF_AUTO); + sf->sf_status = SS_COMPLETED; + l_wait_event(thread->t_ctl_waitq, + !thread_is_running(thread) || + !scrub->os_partial_scan || + scrub->os_in_join || + !list_empty(&scrub->os_inconsistent_items), + &lwi); + sf->sf_flags = saved_flags; + sf->sf_status = SS_SCANNING; + + if (unlikely(!thread_is_running(thread))) + RETURN(0); + + if (!scrub->os_partial_scan || scrub->os_in_join) + goto full; + + continue; + } + default: + LASSERTF(rc == 0, "rc = %d\n", rc); + + osd_scrub_exec(info, dev, ¶m, oic, &noslot, rc); + break; + } + } + +full: if (!preload) { - struct osd_scrub *scrub = &dev->od_scrub; + l_wait_event(thread->t_ctl_waitq, + !thread_is_running(thread) || !scrub->os_in_join, + &lwi); + + if (unlikely(!thread_is_running(thread))) + RETURN(0); + } + noslot = false; + if (!preload) { next = osd_scrub_next; exec = osd_scrub_exec; pos = &scrub->os_pos_current; @@ -933,26 +1384,46 @@ static int osd_inode_iteration(struct osd_thread_info *info, pos = &ooc->ooc_pos_preload; count = &ooc->ooc_cached_items; } - param.sb = osd_sb(dev); limit = le32_to_cpu(LDISKFS_SB(param.sb)->s_es->s_inodes_count); while (*pos <= limit && *count < max) { struct osd_idmap_cache *oic = NULL; + struct ldiskfs_group_desc *desc; param.bg = (*pos - 1) / LDISKFS_INODES_PER_GROUP(param.sb); + desc = ldiskfs_get_group_desc(param.sb, param.bg, NULL); + if (desc == NULL) + RETURN(-EIO); + + ldiskfs_lock_group(param.sb, param.bg); + if (desc->bg_flags & cpu_to_le16(LDISKFS_BG_INODE_UNINIT)) { + ldiskfs_unlock_group(param.sb, param.bg); + *pos = 1 + (param.bg + 1) * + LDISKFS_INODES_PER_GROUP(param.sb); + continue; + } + ldiskfs_unlock_group(param.sb, param.bg); + param.offset = (*pos - 1) % LDISKFS_INODES_PER_GROUP(param.sb); param.gbase = 1 + param.bg * LDISKFS_INODES_PER_GROUP(param.sb); param.bitmap = ldiskfs_read_inode_bitmap(param.sb, param.bg); if (param.bitmap == NULL) { - CERROR("%.16s: fail to read bitmap for %u, " + CDEBUG(D_LFSCK, "%.16s: fail to read bitmap for %u, " "scrub will stop, urgent mode\n", - LDISKFS_SB(param.sb)->s_es->s_volume_name, - (__u32)param.bg); + osd_scrub2name(scrub), (__u32)param.bg); RETURN(-EIO); } while (param.offset < LDISKFS_INODES_PER_GROUP(param.sb) && *count < max) { + if (param.offset + + ldiskfs_itable_unused_count(param.sb, desc) > + LDISKFS_INODES_PER_GROUP(param.sb)) { + *pos = 1 + (param.bg + 1) * + LDISKFS_INODES_PER_GROUP(param.sb); + goto next_group; + } + rc = next(info, dev, ¶m, &oic, noslot); switch (rc) { case SCRUB_NEXT_BREAK: @@ -998,13 +1469,11 @@ static int osd_otable_it_preload(const struct lu_env *env, if (rc == SCRUB_IT_ALL) it->ooi_all_cached = 1; - CDEBUG(D_LFSCK, "OSD pre-loaded: max = %u, preload = %u, rc = %d\n", - le32_to_cpu(LDISKFS_SB(osd_sb(dev))->s_es->s_inodes_count), - 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); @@ -1016,25 +1485,24 @@ static int osd_scrub_main(void *args) struct osd_device *dev = (struct osd_device *)args; struct osd_scrub *scrub = &dev->od_scrub; struct ptlrpc_thread *thread = &scrub->os_thread; - struct super_block *sb = osd_sb(dev); int rc; ENTRY; - 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); + CDEBUG(D_LFSCK, "%.16s: OI scrub fail to init env: rc = %d\n", + osd_scrub2name(scrub), rc); GOTO(noenv, rc); } rc = osd_scrub_prep(dev); if (rc != 0) { - CERROR("%.16s: OI scrub, fail to scrub prep, rc = %d\n", - LDISKFS_SB(sb)->s_es->s_volume_name, rc); + CDEBUG(D_LFSCK, "%.16s: OI scrub fail to scrub prep: rc = %d\n", + osd_scrub2name(scrub), rc); GOTO(out, rc); } - if (!scrub->os_full_speed) { + if (!scrub->os_full_speed && !scrub->os_partial_scan) { struct l_wait_info lwi = { 0 }; struct osd_otable_it *it = dev->od_otable_it; struct osd_otable_cache *ooc = &it->ooi_cache; @@ -1048,8 +1516,9 @@ static int osd_scrub_main(void *args) scrub->os_pos_current = ooc->ooc_pos_preload; } - CDEBUG(D_LFSCK, "OI scrub: flags = 0x%x, pos = %u\n", - scrub->os_start_flags, scrub->os_pos_current); + CDEBUG(D_LFSCK, "%.16s: OI scrub start, flags = 0x%x, pos = %u\n", + osd_scrub2name(scrub), scrub->os_start_flags, + scrub->os_pos_current); rc = osd_inode_iteration(osd_oti_get(&env), dev, ~0U, false); if (unlikely(rc == SCRUB_IT_CRASH)) @@ -1057,17 +1526,17 @@ static int osd_scrub_main(void *args) GOTO(post, rc); post: - osd_scrub_post(scrub, rc); - CDEBUG(D_LFSCK, "OI scrub: stop, rc = %d, pos = %u\n", - rc, scrub->os_pos_current); + rc = osd_scrub_post(scrub, rc); + CDEBUG(D_LFSCK, "%.16s: OI scrub: stop, pos = %u: rc = %d\n", + osd_scrub2name(scrub), scrub->os_pos_current, rc); out: - while (!cfs_list_empty(&scrub->os_inconsistent_items)) { + while (!list_empty(&scrub->os_inconsistent_items)) { struct osd_inconsistent_item *oii; - oii = cfs_list_entry(scrub->os_inconsistent_items.next, + oii = list_entry(scrub->os_inconsistent_items.next, struct osd_inconsistent_item, oii_list); - cfs_list_del_init(&oii->oii_list); + list_del_init(&oii->oii_list); OBD_FREE_PTR(oii); } lu_env_fini(&env); @@ -1075,7 +1544,7 @@ 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; } @@ -1085,8 +1554,29 @@ noenv: typedef int (*scandir_t)(struct osd_thread_info *, struct osd_device *, struct dentry *, filldir_t filldir); +#ifdef HAVE_FILLDIR_USE_CTX +static int osd_ios_varfid_fill(struct dir_context *buf, const char *name, + int namelen, loff_t offset, __u64 ino, + unsigned d_type); +static int osd_ios_lf_fill(struct dir_context *buf, const char *name, + int namelen, loff_t offset, __u64 ino, + unsigned d_type); +static int osd_ios_dl_fill(struct dir_context *buf, const char *name, + int namelen, loff_t offset, __u64 ino, + unsigned d_type); +static int osd_ios_uld_fill(struct dir_context *buf, const char *name, + int namelen, loff_t offset, __u64 ino, + unsigned d_type); +#else 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_uld_fill(void *buf, const char *name, int namelen, + loff_t offset, __u64 ino, unsigned d_type); +#endif static int osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev, @@ -1103,12 +1593,15 @@ enum osd_lf_flags { OLF_SCAN_SUBITEMS = 0x0001, OLF_HIDE_FID = 0x0002, OLF_SHOW_NAME = 0x0004, + OLF_NO_OI = 0x0008, + OLF_IDX_IN_FID = 0x0010, }; struct osd_lf_map { char *olm_name; struct lu_fid olm_fid; __u16 olm_flags; + __u16 olm_namelen; scandir_t olm_scandir; filldir_t olm_filldir; }; @@ -1117,96 +1610,158 @@ struct osd_lf_map { static const struct osd_lf_map osd_lf_maps[] = { /* CATALOGS */ { CATLIST, { FID_SEQ_LOCAL_FILE, LLOG_CATALOGS_OID, 0 }, OLF_SHOW_NAME, - NULL, NULL }, + sizeof(CATLIST) - 1, NULL, NULL }, /* CONFIGS */ { MOUNT_CONFIGS_DIR, { FID_SEQ_LOCAL_FILE, MGS_CONFIGS_OID, 0 }, - OLF_SCAN_SUBITEMS, osd_ios_general_scan, - osd_ios_varfid_fill }, + OLF_SCAN_SUBITEMS, sizeof(MOUNT_CONFIGS_DIR) - 1, + osd_ios_general_scan, osd_ios_varfid_fill }, /* NIDTBL_VERSIONS */ { MGS_NIDTBL_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, - osd_ios_general_scan, osd_ios_varfid_fill }, + sizeof(MGS_NIDTBL_DIR) - 1, osd_ios_general_scan, + osd_ios_varfid_fill }, /* PENDING */ - { "PENDING", { 0, 0, 0 }, 0, NULL, NULL }, + { "PENDING", { 0, 0, 0 }, 0, sizeof("PENDING") - 1, NULL, NULL }, /* ROOT */ - { "ROOT", { FID_SEQ_ROOT, 1, 0 }, - OLF_SCAN_SUBITEMS | OLF_HIDE_FID, osd_ios_ROOT_scan, NULL }, + { "ROOT", { FID_SEQ_ROOT, FID_OID_ROOT, 0 }, + OLF_SCAN_SUBITEMS | OLF_HIDE_FID, sizeof("ROOT") - 1, + osd_ios_ROOT_scan, NULL }, /* changelog_catalog */ - { CHANGELOG_CATALOG, { 0, 0, 0 }, 0, NULL, NULL }, + { CHANGELOG_CATALOG, { 0, 0, 0 }, 0, sizeof(CHANGELOG_CATALOG) - 1, + NULL, NULL }, /* changelog_users */ - { CHANGELOG_USERS, { 0, 0, 0 }, 0, NULL, NULL }, + { CHANGELOG_USERS, { 0, 0, 0 }, 0, sizeof(CHANGELOG_USERS) - 1, + NULL, NULL }, /* fld */ { "fld", { FID_SEQ_LOCAL_FILE, FLD_INDEX_OID, 0 }, OLF_SHOW_NAME, - NULL, NULL }, + sizeof("fld") - 1, NULL, NULL }, /* last_rcvd */ { LAST_RCVD, { FID_SEQ_LOCAL_FILE, LAST_RECV_OID, 0 }, OLF_SHOW_NAME, - NULL, NULL }, + sizeof(LAST_RCVD) - 1, NULL, NULL }, - /* lfsck_bookmark */ - { "lfsck_bookmark", { 0, 0, 0 }, 0, NULL, NULL }, + /* reply_data */ + { REPLY_DATA, { FID_SEQ_LOCAL_FILE, REPLY_DATA_OID, 0 }, OLF_SHOW_NAME, + sizeof(REPLY_DATA) - 1, NULL, NULL }, /* lov_objid */ { LOV_OBJID, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OID, 0 }, OLF_SHOW_NAME, - NULL, NULL }, + sizeof(LOV_OBJID) - 1, NULL, NULL }, /* lov_objseq */ { LOV_OBJSEQ, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OSEQ, 0 }, - OLF_SHOW_NAME, NULL, NULL }, + OLF_SHOW_NAME, sizeof(LOV_OBJSEQ) - 1, NULL, NULL }, /* quota_master */ - { QMT_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, + { QMT_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(QMT_DIR) - 1, osd_ios_general_scan, osd_ios_varfid_fill }, /* quota_slave */ - { QSD_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, + { QSD_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(QSD_DIR) - 1, osd_ios_general_scan, osd_ios_varfid_fill }, /* seq_ctl */ { "seq_ctl", { FID_SEQ_LOCAL_FILE, FID_SEQ_CTL_OID, 0 }, - OLF_SHOW_NAME, NULL, NULL }, + OLF_SHOW_NAME, sizeof("seq_ctl") - 1, NULL, NULL }, /* seq_srv */ { "seq_srv", { FID_SEQ_LOCAL_FILE, FID_SEQ_SRV_OID, 0 }, - OLF_SHOW_NAME, NULL, NULL }, + OLF_SHOW_NAME, sizeof("seq_srv") - 1, NULL, NULL }, /* health_check */ { HEALTH_CHECK, { FID_SEQ_LOCAL_FILE, OFD_HEALTH_CHECK_OID, 0 }, - OLF_SHOW_NAME, NULL, NULL }, + OLF_SHOW_NAME, sizeof(HEALTH_CHECK) - 1, NULL, NULL }, + + /* LFSCK */ + { LFSCK_DIR, { 0, 0, 0 }, 0, sizeof(LFSCK_DIR) - 1, + osd_ios_general_scan, osd_ios_varfid_fill }, + + /* lfsck_bookmark */ + { LFSCK_BOOKMARK, { 0, 0, 0 }, 0, sizeof(LFSCK_BOOKMARK) - 1, + NULL, NULL }, + + /* lfsck_layout */ + { LFSCK_LAYOUT, { 0, 0, 0 }, 0, sizeof(LFSCK_LAYOUT) - 1, + NULL, NULL }, /* lfsck_namespace */ - { "lfsck_namespace", { 0, 0, 0 }, 0, NULL, NULL }, + { LFSCK_NAMESPACE, { 0, 0, 0 }, 0, sizeof(LFSCK_NAMESPACE) - 1, + NULL, NULL }, /* OBJECTS, upgrade from old device */ - { OBJECTS, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, osd_ios_OBJECTS_scan, NULL }, + { OBJECTS, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(OBJECTS) - 1, + osd_ios_OBJECTS_scan, NULL }, /* lquota_v2.user, upgrade from old device */ - { "lquota_v2.user", { 0, 0, 0 }, 0, NULL, NULL }, + { "lquota_v2.user", { 0, 0, 0 }, 0, sizeof("lquota_v2.user") - 1, + NULL, NULL }, /* lquota_v2.group, upgrade from old device */ - { "lquota_v2.group", { 0, 0, 0 }, 0, NULL, NULL }, + { "lquota_v2.group", { 0, 0, 0 }, 0, sizeof("lquota_v2.group") - 1, + NULL, NULL }, /* LAST_GROUP, upgrade from old device */ { "LAST_GROUP", { FID_SEQ_LOCAL_FILE, OFD_LAST_GROUP_OID, 0 }, - OLF_SHOW_NAME, NULL, NULL }, + OLF_SHOW_NAME, sizeof("LAST_GROUP") - 1, NULL, NULL }, + + /* committed batchid for cross-MDT operation */ + { "BATCHID", { FID_SEQ_LOCAL_FILE, BATCHID_COMMITTED_OID, 0 }, + OLF_SHOW_NAME, sizeof("BATCHID") - 1, NULL, NULL }, + + /* OSP update logs update_log{_dir} use f_seq = FID_SEQ_UPDATE_LOG{_DIR} + * and f_oid = index for their log files. See lu_update_log{_dir}_fid() + * for more details. */ + + /* update_log */ + { "update_log", { FID_SEQ_UPDATE_LOG, 0, 0 }, + OLF_SHOW_NAME | OLF_IDX_IN_FID, sizeof("update_log") - 1, + NULL, NULL }, + + /* update_log_dir */ + { "update_log_dir", { FID_SEQ_UPDATE_LOG_DIR, 0, 0 }, + OLF_SHOW_NAME | OLF_SCAN_SUBITEMS | OLF_IDX_IN_FID, + sizeof("update_log_dir") - 1, + osd_ios_general_scan, osd_ios_uld_fill }, + + /* lost+found */ + { "lost+found", { FID_SEQ_LOCAL_FILE, OSD_LPF_OID, 0 }, + OLF_SCAN_SUBITEMS, sizeof("lost+found") - 1, + osd_ios_general_scan, osd_ios_lf_fill }, - { NULL, { 0, 0, 0 }, 0, NULL, NULL } + { NULL, { 0, 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, + sizeof("fid") - 1, NULL, NULL }, + + /* .lustre/lost+found */ + { "lost+found", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_LPF, 0 }, 0, + sizeof("lost+found") - 1, NULL, NULL }, + + { NULL, { 0, 0, 0 }, 0, 0, NULL, NULL } }; struct osd_ios_item { - cfs_list_t oii_list; + struct list_head oii_list; struct dentry *oii_dentry; scandir_t oii_scandir; filldir_t oii_filldir; }; 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; @@ -1218,7 +1773,19 @@ osd_ios_lookup_one_len(const char *name, struct dentry *parent, int namelen) struct dentry *dentry; dentry = ll_lookup_one_len(name, parent, namelen); - if (!IS_ERR(dentry) && dentry->d_inode == NULL) { + if (IS_ERR(dentry)) { + int rc = PTR_ERR(dentry); + + if (rc != -ENOENT) + CERROR("Fail to find %.*s in %.*s (%lu/%u): rc = %d\n", + namelen, name, parent->d_name.len, + parent->d_name.name, parent->d_inode->i_ino, + parent->d_inode->i_generation, rc); + + return dentry; + } + + if (dentry->d_inode == NULL) { dput(dentry); return ERR_PTR(-ENOENT); } @@ -1226,48 +1793,24 @@ 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); + 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; + list_add_tail(&item->oii_list, &dev->od_ios_list); + + RETURN(0); } /** @@ -1291,29 +1834,49 @@ osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev, ENTRY; rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma); - if (rc != 0 && rc != -ENODATA) + if (rc != 0 && rc != -ENODATA) { + CDEBUG(D_LFSCK, "%s: fail to get lma for init OI scrub: " + "rc = %d\n", osd_name(dev), rc); + RETURN(rc); + } osd_id_gen(id, inode->i_ino, inode->i_generation); if (rc == -ENODATA) { - if (fid == NULL || fid_is_zero(fid) || flags & OLF_HIDE_FID) + if (fid == NULL || fid_is_zero(fid) || flags & OLF_HIDE_FID) { lu_igif_build(&tfid, inode->i_ino, inode->i_generation); - else + } else { tfid = *fid; - rc = osd_ea_fid_set(info, inode, &tfid, 0); - if (rc != 0) + if (flags & OLF_IDX_IN_FID) { + LASSERT(dev->od_index >= 0); + + tfid.f_oid = dev->od_index; + } + } + rc = osd_ea_fid_set(info, inode, &tfid, 0, 0); + if (rc != 0) { + CDEBUG(D_LFSCK, "%s: fail to set LMA for init OI " + "scrub: rc = %d\n", osd_name(dev), rc); + 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, NULL); + if (rc > 0) + rc = 0; + RETURN(rc); } @@ -1329,15 +1892,102 @@ 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, NULL); + if (rc > 0) + rc = 0; RETURN(rc); } -static int osd_ios_varfid_fill(void *buf, const char *name, int namelen, +/** + * 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. + */ +#ifdef HAVE_FILLDIR_USE_CTX +static int osd_ios_lf_fill(struct dir_context *buf, +#else +static int osd_ios_lf_fill(void *buf, +#endif + const char *name, int namelen, + loff_t offset, __u64 ino, unsigned d_type) +{ + struct osd_ios_filldir_buf *fill_buf = + (struct osd_ios_filldir_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)) { + CDEBUG(D_LFSCK, "%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) + CDEBUG(D_LFSCK, "%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 { + CDEBUG(D_LFSCK, "%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; +} + +#ifdef HAVE_FILLDIR_USE_CTX +static int osd_ios_varfid_fill(struct dir_context *buf, +#else +static int osd_ios_varfid_fill(void *buf, +#endif + const char *name, int namelen, loff_t offset, __u64 ino, unsigned d_type) { - struct osd_ios_filldir_buf *fill_buf = buf; + struct osd_ios_filldir_buf *fill_buf = + (struct osd_ios_filldir_buf *)buf; struct osd_device *dev = fill_buf->oifb_dev; struct dentry *child; int rc; @@ -1361,10 +2011,16 @@ static int osd_ios_varfid_fill(void *buf, const char *name, int namelen, RETURN(rc); } -static int osd_ios_root_fill(void *buf, const char *name, int namelen, - loff_t offset, __u64 ino, unsigned d_type) +#ifdef HAVE_FILLDIR_USE_CTX +static int osd_ios_dl_fill(struct dir_context *buf, +#else +static int osd_ios_dl_fill(void *buf, +#endif + const char *name, int namelen, + loff_t offset, __u64 ino, unsigned d_type) { - struct osd_ios_filldir_buf *fill_buf = buf; + struct osd_ios_filldir_buf *fill_buf = + (struct osd_ios_filldir_buf *)buf; struct osd_device *dev = fill_buf->oifb_dev; const struct osd_lf_map *map; struct dentry *child; @@ -1375,8 +2031,8 @@ static int osd_ios_root_fill(void *buf, const char *name, int namelen, if (name[0] == '.') RETURN(0); - for (map = osd_lf_maps; map->olm_name != NULL; map++) { - if (strlen(map->olm_name) != namelen) + for (map = osd_dl_maps; map->olm_name != NULL; map++) { + if (map->olm_namelen != namelen) continue; if (strncmp(map->olm_name, name, namelen) == 0) @@ -1392,6 +2048,84 @@ static int osd_ios_root_fill(void *buf, const char *name, int namelen, rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode, &map->olm_fid, map->olm_flags); + dput(child); + + RETURN(rc); +} + +#ifdef HAVE_FILLDIR_USE_CTX +static int osd_ios_uld_fill(struct dir_context *buf, +#else +static int osd_ios_uld_fill(void *buf, +#endif + const char *name, int namelen, + loff_t offset, __u64 ino, unsigned d_type) +{ + struct osd_ios_filldir_buf *fill_buf = + (struct osd_ios_filldir_buf *)buf; + struct dentry *child; + struct lu_fid tfid; + int rc = 0; + ENTRY; + + /* skip any non-DFID format name */ + if (name[0] != '[') + RETURN(0); + + child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen); + if (IS_ERR(child)) + RETURN(PTR_ERR(child)); + + /* skip the start '[' */ + sscanf(&name[1], SFID, RFID(&tfid)); + if (fid_is_sane(&tfid)) + rc = osd_ios_scan_one(fill_buf->oifb_info, fill_buf->oifb_dev, + child->d_inode, &tfid, 0); + else + rc = -EIO; + dput(child); + + RETURN(rc); +} + +#ifdef HAVE_FILLDIR_USE_CTX +static int osd_ios_root_fill(struct dir_context *buf, +#else +static int osd_ios_root_fill(void *buf, +#endif + const char *name, int namelen, + loff_t offset, __u64 ino, unsigned d_type) +{ + struct osd_ios_filldir_buf *fill_buf = + (struct osd_ios_filldir_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_lf_maps; map->olm_name != NULL; map++) { + if (map->olm_namelen != 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)); + + 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); @@ -1404,8 +2138,14 @@ 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 file *filp = &info->oti_it_ea.oie_file; + 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_file; struct inode *inode = dentry->d_inode; const struct file_operations *fops = inode->i_fop; int rc; @@ -1414,13 +2154,20 @@ osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev, LASSERT(filldir != NULL); filp->f_pos = 0; - filp->f_dentry = dentry; + filp->f_path.dentry = dentry; filp->f_mode = FMODE_64BITHASH; 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); @@ -1436,8 +2183,20 @@ 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)) { @@ -1448,6 +2207,7 @@ 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; @@ -1479,6 +2239,9 @@ 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); } @@ -1489,10 +2252,19 @@ 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); @@ -1521,13 +2293,17 @@ 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) { @@ -1538,12 +2314,12 @@ static int osd_initial_OI_scrub(struct osd_thread_info *info, if (rc != 0) break; - if (cfs_list_empty(&dev->od_ios_list)) + if (list_empty(&dev->od_ios_list)) break; - item = cfs_list_entry(dev->od_ios_list.next, - struct osd_ios_item, oii_list); - cfs_list_del_init(&item->oii_list); + item = list_entry(dev->od_ios_list.next, + struct osd_ios_item, oii_list); + list_del_init(&item->oii_list); LASSERT(item->oii_scandir != NULL); scandir = item->oii_scandir; @@ -1551,15 +2327,41 @@ static int osd_initial_OI_scrub(struct osd_thread_info *info, dentry = item->oii_dentry; } - while (!cfs_list_empty(&dev->od_ios_list)) { - item = cfs_list_entry(dev->od_ios_list.next, - struct osd_ios_item, oii_list); - cfs_list_del_init(&item->oii_list); + while (!list_empty(&dev->od_ios_list)) { + item = list_entry(dev->od_ios_list.next, + struct osd_ios_item, oii_list); + list_del_init(&item->oii_list); dput(item->oii_dentry); 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, + map->olm_namelen); + 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, NULL); + map++; + } + + RETURN(0); } char *osd_lf_fid2name(const struct lu_fid *fid) @@ -1588,33 +2390,57 @@ 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; -again: /* os_lock: sync status between stop and scrub thread */ spin_lock(&scrub->os_lock); + +again: if (thread_is_running(thread)) { spin_unlock(&scrub->os_lock); - RETURN(-EALREADY); - } else if (unlikely(thread_is_stopping(thread))) { + if (!(scrub->os_file.sf_flags & SF_AUTO || + scrub->os_partial_scan) || + (flags & SS_AUTO_PARTIAL)) + RETURN(-EALREADY); + + osd_scrub_join(dev, flags, false); + spin_lock(&scrub->os_lock); + if (!thread_is_running(thread)) + goto again; + + spin_unlock(&scrub->os_lock); + RETURN(0); + } + + if (unlikely(thread_is_stopping(thread))) { spin_unlock(&scrub->os_lock); l_wait_event(thread->t_ctl_waitq, thread_is_stopped(thread), &lwi); + spin_lock(&scrub->os_lock); goto again; } spin_unlock(&scrub->os_lock); - if (scrub->os_file.sf_status == SS_COMPLETED) + if (scrub->os_file.sf_status == SS_COMPLETED) { + if (!(flags & SS_SET_FAILOUT)) + flags |= SS_CLEAR_FAILOUT; + + if (!(flags & SS_SET_DRYRUN)) + flags |= SS_CLEAR_DRYRUN; + flags |= SS_RESET; + } scrub->os_start_flags = flags; thread_set_flags(thread, 0); - rc = PTR_ERR(kthread_run(osd_scrub_main, dev, "OI_scrub")); - if (IS_ERR_VALUE(rc)) { - CERROR("%.16s: cannot start iteration thread, rc = %d\n", - LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, rc); + 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", + osd_scrub2name(scrub), rc); RETURN(rc); } @@ -1625,14 +2451,14 @@ again: RETURN(0); } -int osd_scrub_start(struct osd_device *dev) +int osd_scrub_start(struct osd_device *dev, __u32 flags) { int rc; ENTRY; /* od_otable_mutex: prevent curcurrent start/stop */ mutex_lock(&dev->od_otable_mutex); - rc = do_osd_scrub_start(dev, SS_AUTO); + rc = do_osd_scrub_start(dev, flags); mutex_unlock(&dev->od_otable_mutex); RETURN(rc == -EALREADY ? 0 : rc); @@ -1648,7 +2474,7 @@ 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); @@ -1682,7 +2508,10 @@ 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; - int dirty = 0; + struct inode *inode; + struct lu_fid *fid = &info->oti_fid; + bool dirty = false; + bool restored = false; int rc = 0; ENTRY; @@ -1692,35 +2521,82 @@ 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); + 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)); + } + + inode = file_inode(filp); + /* '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, NULL); + pop_ctxt(&saved, ctxt); + RETURN(rc); + } - scrub->os_inode = igrab(filp->f_dentry->d_inode); - filp_close(filp, 0); - pop_ctxt(&saved, ctxt, NULL); - ldiskfs_set_inode_state(scrub->os_inode, - LDISKFS_STATE_LUSTRE_NO_OI); + scrub->os_inode = igrab(inode); + filp_close(filp, NULL); + pop_ctxt(&saved, ctxt); rc = osd_scrub_file_load(scrub); if (rc == -ENOENT) { osd_scrub_file_init(scrub, es->s_uuid); - dirty = 1; + /* 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 = true; } else if (rc != 0) { - RETURN(rc); + GOTO(cleanup_inode, rc); } else { if (memcmp(sf->sf_uuid, es->s_uuid, 16) != 0) { + struct obd_uuid *old_uuid; + struct obd_uuid *new_uuid; + + OBD_ALLOC_PTR(old_uuid); + OBD_ALLOC_PTR(new_uuid); + if (old_uuid == NULL || new_uuid == NULL) { + CERROR("%.16s: UUID has been changed, but" + "failed to allocate RAM for report\n", + LDISKFS_SB(sb)->s_es->s_volume_name); + } else { + class_uuid_unparse(sf->sf_uuid, old_uuid); + class_uuid_unparse(es->s_uuid, new_uuid); + CERROR("%.16s: UUID has been changed from " + "%s to %s\n", + LDISKFS_SB(sb)->s_es->s_volume_name, + old_uuid->uuid, new_uuid->uuid); + } osd_scrub_file_reset(scrub, es->s_uuid,SF_INCONSISTENT); - dirty = 1; + dirty = true; + restored = true; + if (old_uuid != NULL) + OBD_FREE_PTR(old_uuid); + if (new_uuid != NULL) + OBD_FREE_PTR(new_uuid); } else if (sf->sf_status == SS_SCANNING) { sf->sf_status = SS_CRASHED; - dirty = 1; + dirty = true; } } @@ -1729,47 +2605,70 @@ int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev) else scrub->os_pos_current = LDISKFS_FIRST_INO(sb) + 1; - if (dirty != 0) { + if (dirty) { rc = osd_scrub_file_store(scrub); if (rc != 0) - RETURN(rc); + GOTO(cleanup_inode, rc); } /* Initialize OI files. */ - rc = osd_oi_init(info, dev); + rc = osd_oi_init(info, dev, restored); if (rc < 0) - RETURN(rc); + GOTO(cleanup_inode, rc); 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. */ - dev->od_igif_inoi = 0; - else - /* 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 (rc != 0) + GOTO(cleanup_oi, rc); - 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); + 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; } - RETURN(rc); + 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, SS_AUTO_FULL); + + if (rc != 0) + GOTO(cleanup_oi, rc); + + /* 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(0); +cleanup_oi: + osd_oi_fini(info, dev); +cleanup_inode: + iput(scrub->os_inode); + scrub->os_inode = NULL; + + return rc; } void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev) @@ -1790,8 +2689,7 @@ void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev) /* object table based iteration APIs */ static struct dt_it *osd_otable_it_init(const struct lu_env *env, - struct dt_object *dt, __u32 attr, - struct lustre_capa *capa) + struct dt_object *dt, __u32 attr) { enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT; enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK; @@ -1827,7 +2725,14 @@ static struct dt_it *osd_otable_it_init(const struct lu_env *env, start |= SS_CLEAR_FAILOUT; } - rc = do_osd_scrub_start(dev, start); + if (valid & DOIV_DRYRUN) { + if (flags & DOIF_DRYRUN) + start |= SS_SET_DRYRUN; + else + start |= SS_CLEAR_DRYRUN; + } + + rc = do_osd_scrub_start(dev, start & ~SS_AUTO_PARTIAL); if (rc < 0 && rc != -EALREADY) { dev->od_otable_it = NULL; OBD_FREE_PTR(it); @@ -1917,7 +2822,7 @@ 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); } @@ -1996,6 +2901,8 @@ static int osd_otable_it_load(const struct lu_env *env, if (it->ooi_user_ready) RETURN(-EPERM); + LASSERT(!scrub->os_partial_scan); + if (hash > OSD_OTABLE_MAX_HASH) hash = OSD_OTABLE_MAX_HASH; @@ -2005,7 +2912,7 @@ 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); @@ -2037,6 +2944,8 @@ const struct dt_index_operations osd_otable_ops = { /* high priority inconsistent items list APIs */ +#define SCRUB_BAD_OIMAP_DECAY_INTERVAL 60 + int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic, int insert) { @@ -2050,10 +2959,30 @@ int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic, if (unlikely(oii == NULL)) RETURN(-ENOMEM); - CFS_INIT_LIST_HEAD(&oii->oii_list); + INIT_LIST_HEAD(&oii->oii_list); oii->oii_cache = *oic; oii->oii_insert = insert; + if (scrub->os_partial_scan) { + __u64 now = cfs_time_current_sec(); + + /* If there haven't been errors in a long time, + * decay old count until either the errors are + * gone or we reach the current interval. */ + while (unlikely(scrub->os_bad_oimap_count > 0 && + scrub->os_bad_oimap_time + + SCRUB_BAD_OIMAP_DECAY_INTERVAL < now)) { + scrub->os_bad_oimap_count >>= 1; + scrub->os_bad_oimap_time += + SCRUB_BAD_OIMAP_DECAY_INTERVAL; + } + + scrub->os_bad_oimap_time = now; + if (++scrub->os_bad_oimap_count > + dev->od_full_scrub_threshold_rate) + scrub->os_full_scrub = 1; + } + spin_lock(&scrub->os_lock); if (unlikely(!thread_is_running(thread))) { spin_unlock(&scrub->os_lock); @@ -2061,13 +2990,13 @@ int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic, RETURN(-EAGAIN); } - if (cfs_list_empty(&scrub->os_inconsistent_items)) + if (list_empty(&scrub->os_inconsistent_items)) wakeup = 1; - cfs_list_add_tail(&oii->oii_list, &scrub->os_inconsistent_items); + list_add_tail(&oii->oii_list, &scrub->os_inconsistent_items); spin_unlock(&scrub->os_lock); if (wakeup != 0) - cfs_waitq_broadcast(&thread->t_ctl_waitq); + wake_up_all(&thread->t_ctl_waitq); RETURN(0); } @@ -2080,7 +3009,7 @@ int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid, ENTRY; spin_lock(&scrub->os_lock); - cfs_list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) { + list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) { if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) { *id = oii->oii_cache.oic_lid; spin_unlock(&scrub->os_lock); @@ -2115,156 +3044,99 @@ 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[], - const char *prefix) +static void 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; + seq_printf(m, "%s:%c", prefix, bits != 0 ? ' ' : '\n'); - *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; + seq_printf(m, "%s%c", names[i], + bits != 0 ? ',' : '\n'); } } - return save - *len; } -static int scrub_time_dump(char **buf, int *len, __u64 time, const char *prefix) +static void 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, - cfs_time_current_sec() - time); + 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; - return rc; + seq_printf(m, "%s: N/A\n", prefix); } -static int scrub_pos_dump(char **buf, int *len, __u64 pos, const char *prefix) +static void 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); + 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; - return rc; + seq_printf(m, "%s: N/A\n", prefix); } -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) - goto out; - - buf += rc; - len -= rc; - rc = scrub_bits_dump(&buf, &len, sf->sf_flags, scrub_flags_names, - "flags"); - if (rc < 0) - goto out; + 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]); - rc = scrub_bits_dump(&buf, &len, sf->sf_param, scrub_param_names, - "param"); - if (rc < 0) - goto out; + scrub_bits_dump(m, sf->sf_flags, scrub_flags_names, "flags"); - rc = scrub_time_dump(&buf, &len, sf->sf_time_last_complete, - "time_since_last_completed"); - if (rc < 0) - goto out; + scrub_bits_dump(m, sf->sf_param, scrub_param_names, "param"); - rc = scrub_time_dump(&buf, &len, sf->sf_time_latest_start, - "time_since_latest_start"); - if (rc < 0) - goto out; + scrub_time_dump(m, sf->sf_time_last_complete, + "time_since_last_completed"); - rc = scrub_time_dump(&buf, &len, sf->sf_time_last_checkpoint, - "time_since_last_checkpoint"); - if (rc < 0) - goto out; + scrub_time_dump(m, sf->sf_time_latest_start, + "time_since_latest_start"); - rc = scrub_pos_dump(&buf, &len, sf->sf_pos_latest_start, - "latest_start_position"); - if (rc < 0) - goto out; + scrub_time_dump(m, sf->sf_time_last_checkpoint, + "time_since_last_checkpoint"); - rc = scrub_pos_dump(&buf, &len, sf->sf_pos_last_checkpoint, - "last_checkpoint_position"); - if (rc < 0) - goto out; + scrub_pos_dump(m, sf->sf_pos_latest_start, + "latest_start_position"); - rc = scrub_pos_dump(&buf, &len, sf->sf_pos_first_inconsistent, - "first_failure_position"); - if (rc < 0) - goto out; + scrub_pos_dump(m, sf->sf_pos_last_checkpoint, + "last_checkpoint_position"); + + scrub_pos_dump(m, sf->sf_pos_first_inconsistent, + "first_failure_position"); checked = sf->sf_items_checked + scrub->os_new_checked; - rc = snprintf(buf, len, - "checked: "LPU64"\n" - "updated: "LPU64"\n" - "failed: "LPU64"\n" - "prior_updated: "LPU64"\n" - "noscrub: "LPU64"\n" - "igif: "LPU64"\n" - "success_count: %u\n", - 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) - goto out; - - buf += rc; - len -= rc; + seq_printf(m, "checked: "LPU64"\n" + "updated: "LPU64"\n" + "failed: "LPU64"\n" + "prior_updated: "LPU64"\n" + "noscrub: "LPU64"\n" + "igif: "LPU64"\n" + "success_count: %u\n", + 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); + 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 = msecs_to_jiffies(scrub->os_new_checked * + MSEC_PER_SEC); __u32 rtime = sf->sf_run_time + cfs_duration_sec(duration + HALF_SEC); @@ -2272,30 +3144,30 @@ 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" - "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); + 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" + "lf_scanned: "LPU64"\n" + "lf_repaired: "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" - "average_speed: "LPU64" objects/sec\n" - "real-time_speed: N/A\n" - "current_position: N/A\n", - sf->sf_run_time, speed); + 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" + "lf_scanned: "LPU64"\n" + "lf_repaired: "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 0; }