X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fosd-ldiskfs%2Fosd_scrub.c;h=a9a191d409300170a66881a8a020ea3fcc86a373;hp=2f385d3d9a75d042fb1977efcd19628f61c8e7f7;hb=d11360f4cc5d38cd748a97ca05e10121353ae616;hpb=d01d4c697a3c4423587159d58da6e455a5a3551f diff --git a/lustre/osd-ldiskfs/osd_scrub.c b/lustre/osd-ldiskfs/osd_scrub.c index 2f385d3..a9a191d 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 Whamcloud, Inc. + * Copyright (c) 2012, 2014, Intel Corporation. */ /* * lustre/osd-ldiskfs/osd_scrub.c @@ -35,20 +35,34 @@ * 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 "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 + +#define SCRUB_NEXT_BREAK 1 /* exit current loop and process next group */ +#define SCRUB_NEXT_CONTINUE 2 /* skip current object and process next bit */ +#define SCRUB_NEXT_EXIT 3 /* exit all the loops */ +#define SCRUB_NEXT_WAIT 4 /* wait for free cache slot */ +#define SCRUB_NEXT_CRASH 5 /* simulate system crash during OI scrub */ +#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 */ static inline struct osd_device *osd_scrub2dev(struct osd_scrub *scrub) { @@ -60,6 +74,119 @@ static inline struct super_block *osd_scrub2sb(struct osd_scrub *scrub) return osd_sb(osd_scrub2dev(scrub)); } +static inline int osd_scrub_has_window(struct osd_scrub *scrub, + struct osd_otable_cache *ooc) +{ + 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 + * + * \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, bool force, + enum oi_check_flags flags) +{ + handle_t *th; + int rc; + ENTRY; + + if (dev->od_scrub.os_file.sf_param & SP_DRYRUN && !force) + RETURN(0); + + /* 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->oii_ino, id->oii_gen, rc); + RETURN(rc); + } + + 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 + * multiple objects to be mapped to the same + * IGIF formatted FID. Consider the following + * situations: + * + * 1) The MDT is upgrading from 1.8 device. + * The OI scrub generates IGIF FID1 for the + * OBJ1 and adds the OI mapping. + * + * 2) For some reason, the OI scrub does not + * process all the IGIF objects completely. + * + * 3) The MDT is backuped and restored against + * this device. + * + * 4) When the MDT mounts up, the OI scrub will + * try to rebuild the OI files. For some IGIF + * object, OBJ2, which was not processed by the + * OI scrub before the backup/restore, and the + * new generated IGIF formatted FID may be just + * the FID1, the same as OBJ1. + * + * Under such case, the OI scrub cannot know how + * to generate new FID for the OBJ2. + * + * Currently, we do nothing for that. One possible + * solution is to generate new normal FID for the + * conflict object. + * + * 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; + } + + 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->oii_ino, id->oii_gen, rc); + + RETURN(rc); +} + +/* OI_scrub file ops */ + static void osd_scrub_file_to_cpu(struct scrub_file *des, struct scrub_file *src) { @@ -91,6 +218,7 @@ 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); } @@ -125,6 +253,7 @@ 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); } @@ -142,11 +271,14 @@ 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; @@ -156,13 +288,16 @@ 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; } 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; @@ -172,16 +307,18 @@ 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 { @@ -204,138 +341,141 @@ 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; } osd_scrub_file_to_le(&scrub->os_file_disk, &scrub->os_file); rc = osd_ldiskfs_write_record(scrub->os_inode, &scrub->os_file_disk, - len, &pos, jh); + 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; } -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; - cfs_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, sf->sf_flags); + 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); + CDEBUG(D_LFSCK, "%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)) - scrub->os_full_speed = 1; - - scrub->os_in_prior = 0; - scrub->os_waiting = 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) 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) { - cfs_spin_lock(&scrub->os_lock); - thread_set_flags(thread, SVC_RUNNING); - cfs_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); } - cfs_up_write(&scrub->os_rwsem); - RETURN(rc); -} + /* 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; + } -static int -osd_scrub_error(struct osd_device *dev, struct osd_inode_id *lid, int rc) -{ - struct osd_scrub *scrub = &dev->od_scrub; - struct scrub_file *sf = &scrub->os_file; + GOTO(stop, rc); - cfs_down_write(&scrub->os_rwsem); - scrub->os_new_checked++; - sf->sf_items_failed++; - if (sf->sf_pos_first_inconsistent == 0 || - sf->sf_pos_first_inconsistent > lid->oii_ino) - sf->sf_pos_first_inconsistent = lid->oii_ino; - cfs_up_write(&scrub->os_rwsem); - return sf->sf_param & SP_FAILOUT ? rc : 0; +stop: + ldiskfs_journal_stop(jh); + if (rc < 0) + CDEBUG(D_LFSCK, "%s: fail to convert ff "DFID": rc = %d\n", + osd_name(dev), PFID(fid), rc); + return rc; } static int osd_scrub_check_update(struct osd_thread_info *info, struct osd_device *dev, - struct osd_idmap_cache *oic) + struct osd_idmap_cache *oic, int val) { struct osd_scrub *scrub = &dev->od_scrub; struct scrub_file *sf = &scrub->os_file; + struct lu_fid *fid = &oic->oic_fid; + struct osd_inode_id *lid = &oic->oic_lid; struct osd_inode_id *lid2 = &info->oti_id; - struct lu_fid *oi_fid = &info->oti_fid; - struct osd_inode_id *oi_id = &info->oti_id2; - handle_t *jh = NULL; struct osd_inconsistent_item *oii = NULL; struct inode *inode = NULL; - struct lu_fid *fid = &oic->oic_fid; - struct osd_inode_id *lid = &oic->oic_lid; - struct iam_container *bag; - struct iam_path_descr *ipd; int ops = DTO_INDEX_UPDATE; int idx; int rc; + bool converted = false; ENTRY; + down_write(&scrub->os_rwsem); + scrub->os_new_checked++; + if (val < 0) + 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); - cfs_down_write(&scrub->os_rwsem); - scrub->os_new_checked++; if (lid->oii_ino < sf->sf_pos_latest_start && oii == NULL) GOTO(out, rc = 0); - if (oii != NULL && oii->oii_insert) - goto iget; - - rc = osd_oi_lookup(info, dev, fid, lid2); - if (rc != 0) { - if (rc != -ENOENT) - GOTO(out, rc); + if (fid_is_igif(fid)) + sf->sf_items_igif++; -iget: + if (val == SCRUB_NEXT_OSTOBJ_OLD) { inode = osd_iget(info, dev, lid); if (IS_ERR(inode)) { rc = PTR_ERR(inode); @@ -345,52 +485,98 @@ iget: GOTO(out, rc); } - /* Prevent the inode to be unlinked during OI scrub. */ - cfs_mutex_lock(&inode->i_mutex); - if (unlikely(inode->i_nlink == 0)) { - cfs_mutex_unlock(&inode->i_mutex); - iput(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)) + 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 && 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); + } } + if (!scrub->os_partial_scan) + scrub->os_full_speed = 1; + ops = DTO_INDEX_INSERT; idx = osd_oi_fid2idx(dev, fid); - if (unlikely(!ldiskfs_test_bit(idx, sf->sf_oi_bitmap))) - ldiskfs_set_bit(idx, sf->sf_oi_bitmap); - sf->sf_flags |= SF_RECREATED; - } else if (osd_id_eq(lid, lid2)) { - GOTO(out, rc = 0); - } + 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); + } - sf->sf_flags |= SF_INCONSISTENT; - fid_cpu_to_be(oi_fid, fid); - osd_id_pack(oi_id, &oic->oic_lid); - 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); - 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++; - 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); - GOTO(out, rc = -ENOMEM); + GOTO(out, rc = 0); + } else { + if (!scrub->os_partial_scan) + 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; } - 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); - osd_ipd_put(info->oti_env, bag, ipd); - ldiskfs_journal_stop(jh); + 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++; @@ -401,37 +587,132 @@ iget: GOTO(out, rc); out: - if (rc != 0) { + if (rc < 0) { sf->sf_items_failed++; if (sf->sf_pos_first_inconsistent == 0 || sf->sf_pos_first_inconsistent > lid->oii_ino) sf->sf_pos_first_inconsistent = lid->oii_ino; + } else { + rc = 0; } - if (ops == DTO_INDEX_INSERT) { - cfs_mutex_unlock(&inode->i_mutex); + /* 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); - } - cfs_up_write(&scrub->os_rwsem); if (oii != NULL) { - LASSERT(!cfs_list_empty(&oii->oii_list)); + LASSERT(!list_empty(&oii->oii_list)); - cfs_spin_lock(&scrub->os_lock); - cfs_list_del_init(&oii->oii_list); - cfs_spin_unlock(&scrub->os_lock); + spin_lock(&scrub->os_lock); + list_del_init(&oii->oii_list); + spin_unlock(&scrub->os_lock); OBD_FREE_PTR(oii); } RETURN(sf->sf_param & SP_FAILOUT ? rc : 0); } -static int do_osd_scrub_checkpoint(struct osd_scrub *scrub) +/* 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; int rc; - ENTRY; - cfs_down_write(&scrub->os_rwsem); + if (likely(cfs_time_before(cfs_time_current(), + scrub->os_time_next_checkpoint) || + scrub->os_new_checked == 0)) + return 0; + + down_write(&scrub->os_rwsem); sf->sf_items_checked += scrub->os_new_checked; scrub->os_new_checked = 0; sf->sf_pos_last_checkpoint = scrub->os_pos_current; @@ -439,18 +720,9 @@ static int do_osd_scrub_checkpoint(struct osd_scrub *scrub) sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC - scrub->os_time_last_checkpoint); rc = osd_scrub_file_store(scrub); - cfs_up_write(&scrub->os_rwsem); - - RETURN(rc); -} + up_write(&scrub->os_rwsem); -static inline int osd_scrub_checkpoint(struct osd_scrub *scrub) -{ - if (unlikely(cfs_time_beforeq(scrub->os_time_next_checkpoint, - cfs_time_current()) && - scrub->os_new_checked > 0)) - return do_osd_scrub_checkpoint(scrub); - return 0; + return rc; } static void osd_scrub_post(struct osd_scrub *scrub, int result) @@ -458,10 +730,13 @@ static void osd_scrub_post(struct osd_scrub *scrub, int result) struct scrub_file *sf = &scrub->os_file; ENTRY; - cfs_down_write(&scrub->os_rwsem); - cfs_spin_lock(&scrub->os_lock); + 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); - cfs_spin_unlock(&scrub->os_lock); + spin_unlock(&scrub->os_lock); if (scrub->os_new_checked > 0) { sf->sf_items_checked += scrub->os_new_checked; scrub->os_new_checked = 0; @@ -469,34 +744,36 @@ 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_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) { - sf->sf_status = SS_PAUSED; + if (scrub->os_paused) + sf->sf_status = SS_PAUSED; + else + sf->sf_status = SS_STOPPED; } else { sf->sf_status = SS_FAILED; } 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); - cfs_up_write(&scrub->os_rwsem); + up_write(&scrub->os_rwsem); EXIT; } -#define SCRUB_NEXT_BREAK 1 /* exit current loop and process next group */ -#define SCRUB_NEXT_CONTINUE 2 /* skip current object and process next bit */ -#define SCRUB_NEXT_EXIT 3 /* exit all the loops */ -#define SCRUB_NEXT_WAIT 4 /* wait for free cache slot */ -#define SCRUB_NEXT_CRASH 5 /* simulate system crash during OI scrub */ -#define SCRUB_NEXT_FATAL 6 /* simulate failure during OI scrub */ +/* iteration engine */ struct osd_iit_param { struct super_block *sb; @@ -510,19 +787,13 @@ 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); - -static inline int osd_scrub_has_window(struct osd_scrub *scrub, - struct osd_otable_cache *ooc) -{ - return scrub->os_pos_current < ooc->ooc_pos_preload + SCRUB_WINDOW_SIZE; -} + bool *noslot, int rc); static int osd_iit_next(struct osd_iit_param *param, __u32 *pos) { @@ -537,56 +808,181 @@ 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, struct inode **pinode) + struct super_block *sb, bool scrub) { 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_fid(info, dev, lid, fid); + inode = osd_iget(info, dev, lid); if (IS_ERR(inode)) { rc = PTR_ERR(inode); /* The inode may be removed after bitmap searching, or the * file is new created without inode initialized yet. */ if (rc == -ENOENT || rc == -ESTALE) - return SCRUB_NEXT_CONTINUE; + 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); } - *pinode = inode; - return 0; -} + if (scrub && + ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB)) { + /* Only skip it for the first OI scrub accessing. */ + ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB); + GOTO(put, rc = SCRUB_NEXT_NOSCRUB); + } -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) + 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, const bool noslot) { struct osd_scrub *scrub = &dev->od_scrub; struct ptlrpc_thread *thread = &scrub->os_thread; struct lu_fid *fid; struct osd_inode_id *lid; - struct inode *inode; int rc; if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && cfs_fail_val > 0) { 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)) { - cfs_spin_lock(&scrub->os_lock); + spin_lock(&scrub->os_lock); thread_set_flags(thread, SVC_STOPPING); - cfs_spin_unlock(&scrub->os_lock); + spin_unlock(&scrub->os_lock); return SCRUB_NEXT_CRASH; } @@ -596,17 +992,17 @@ 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)) { + if (!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); + oii = list_entry(scrub->os_inconsistent_items.next, + struct osd_inconsistent_item, oii_list); *oic = &oii->oii_cache; scrub->os_in_prior = 1; return 0; } - if (noslot != 0) + if (noslot) return SCRUB_NEXT_WAIT; rc = osd_iit_next(param, &scrub->os_pos_current); @@ -617,24 +1013,17 @@ static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev, fid = &(*oic)->oic_fid; lid = &(*oic)->oic_lid; rc = osd_iit_iget(info, dev, fid, lid, - scrub->os_pos_current, param->sb, &inode); - if (rc != 0) - return rc; - - if (!fid_is_norm(fid) || inode->i_state & I_LUSTRE_NOSCRUB) - rc = SCRUB_NEXT_CONTINUE; - iput(inode); + scrub->os_pos_current, param->sb, true); return rc; } 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; struct ptlrpc_thread *thread; - struct inode *inode; int rc; rc = osd_iit_next(param, &ooc->ooc_pos_preload); @@ -650,21 +1039,35 @@ static int osd_preload_next(struct osd_thread_info *info, rc = osd_iit_iget(info, dev, &ooc->ooc_cache[ooc->ooc_producer_idx].oic_fid, &ooc->ooc_cache[ooc->ooc_producer_idx].oic_lid, - ooc->ooc_pos_preload, param->sb, &inode); + ooc->ooc_pos_preload, param->sb, false); /* If succeed, it needs to move forward; otherwise up layer LFSCK may * ignore the failure, so it still need to skip the inode next time. */ ooc->ooc_pos_preload = param->gbase + ++(param->offset); - if (rc == 0) - iput(inode); return rc; } +static inline int +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) || + !list_empty(&scrub->os_inconsistent_items) || + it->ooi_waiting || !thread_is_running(&scrub->os_thread)) + scrub->os_waiting = 0; + else + scrub->os_waiting = 1; + spin_unlock(&scrub->os_lock); + + return !scrub->os_waiting; +} + 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; + struct scrub_file *sf = &scrub->os_file; struct ptlrpc_thread *thread = &scrub->os_thread; struct osd_otable_it *it = dev->od_otable_it; struct osd_otable_cache *ooc = it ? &it->ooi_cache : NULL; @@ -674,21 +1077,22 @@ static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev, goto next; case SCRUB_NEXT_WAIT: goto wait; + case SCRUB_NEXT_NOSCRUB: + down_write(&scrub->os_rwsem); + scrub->os_new_checked++; + sf->sf_items_noscrub++; + up_write(&scrub->os_rwsem); + goto next; } - LASSERTF(rc <= 0, "unexpected rc = %d\n", rc); - - if (rc != 0) - rc = osd_scrub_error(dev, &oic->oic_lid, rc); - else - rc = osd_scrub_check_update(info, dev, oic); + rc = osd_scrub_check_update(info, dev, oic, rc); if (rc != 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. */ } @@ -700,39 +1104,38 @@ static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev, next: scrub->os_pos_current = param->gbase + ++(param->offset); - if (it != NULL && it->ooi_waiting && + +wait: + 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; -wait: - if (osd_scrub_has_window(scrub, ooc)) { - *noslot = 0; + if (ooc != NULL && osd_scrub_has_window(scrub, ooc)) { + *noslot = false; return 0; } - scrub->os_waiting = 1; - l_wait_event(thread->t_ctl_waitq, - osd_scrub_has_window(scrub, ooc) || - !cfs_list_empty(&scrub->os_inconsistent_items) || - !thread_is_running(thread), - &lwi); - scrub->os_waiting = 0; + 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; @@ -747,186 +1150,1114 @@ 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 (flags & SS_AUTO_FULL) { + sf->sf_flags |= SF_AUTO; + scrub->os_full_speed = 1; + } + + if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE)) + scrub->os_full_speed = 1; + else + scrub->os_full_speed = 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; + + 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, int preload) + 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) { + 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; + count = &scrub->os_new_checked; + } else { + struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache; + + next = osd_preload_next; + exec = osd_preload_exec; + pos = &ooc->ooc_pos_preload; + count = &ooc->ooc_cached_items; + } + 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) { + CDEBUG(D_LFSCK, "%.16s: fail to read bitmap for %u, " + "scrub will stop, urgent mode\n", + 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)) + goto next_group; + + rc = next(info, dev, ¶m, &oic, noslot); + switch (rc) { + case SCRUB_NEXT_BREAK: + goto next_group; + case SCRUB_NEXT_EXIT: + brelse(param.bitmap); + RETURN(0); + case SCRUB_NEXT_CRASH: + brelse(param.bitmap); + RETURN(SCRUB_IT_CRASH); + case SCRUB_NEXT_FATAL: + brelse(param.bitmap); + RETURN(-EINVAL); + } + + rc = exec(info, dev, ¶m, oic, &noslot, rc); + if (rc != 0) { + brelse(param.bitmap); + RETURN(rc); + } + } + +next_group: + brelse(param.bitmap); + } + + if (*pos > limit) + RETURN(SCRUB_IT_ALL); + RETURN(0); +} + +static int osd_otable_it_preload(const struct lu_env *env, + struct osd_otable_it *it) +{ + struct osd_device *dev = it->ooi_dev; + struct osd_scrub *scrub = &dev->od_scrub; + struct osd_otable_cache *ooc = &it->ooi_cache; + int rc; + ENTRY; + + rc = osd_inode_iteration(osd_oti_get(env), dev, + OSD_OTABLE_IT_CACHE_SIZE, true); + if (rc == SCRUB_IT_ALL) + it->ooi_all_cached = 1; + + if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) { + spin_lock(&scrub->os_lock); + scrub->os_waiting = 0; + wake_up_all(&scrub->os_thread.t_ctl_waitq); + spin_unlock(&scrub->os_lock); + } + + RETURN(rc < 0 ? rc : ooc->ooc_cached_items); +} + +static int osd_scrub_main(void *args) +{ + struct lu_env env; + struct osd_device *dev = (struct osd_device *)args; + struct osd_scrub *scrub = &dev->od_scrub; + struct ptlrpc_thread *thread = &scrub->os_thread; + int rc; + ENTRY; + + rc = lu_env_init(&env, LCT_LOCAL); + if (rc != 0) { + 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) { + 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 && !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; + + l_wait_event(thread->t_ctl_waitq, + it->ooi_user_ready || !thread_is_running(thread), + &lwi); + if (unlikely(!thread_is_running(thread))) + GOTO(post, rc = 0); + + scrub->os_pos_current = ooc->ooc_pos_preload; + } + + 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)) + GOTO(out, rc = -EINVAL); + GOTO(post, rc); + +post: + 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 (!list_empty(&scrub->os_inconsistent_items)) { + struct osd_inconsistent_item *oii; + + oii = list_entry(scrub->os_inconsistent_items.next, + struct osd_inconsistent_item, oii_list); + list_del_init(&oii->oii_list); + OBD_FREE_PTR(oii); + } + lu_env_fini(&env); + +noenv: + spin_lock(&scrub->os_lock); + thread_set_flags(thread, SVC_STOPPED); + wake_up_all(&thread->t_ctl_waitq); + spin_unlock(&scrub->os_lock); + return rc; +} + +/* initial OI scrub */ + +typedef int (*scandir_t)(struct osd_thread_info *, struct osd_device *, + struct dentry *, filldir_t filldir); + +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, + struct dentry *dentry, filldir_t filldir); +static int +osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev, + struct dentry *dentry, filldir_t filldir); + +static int +osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev, + struct dentry *dentry, filldir_t filldir); + +enum osd_lf_flags { + OLF_SCAN_SUBITEMS = 0x0001, + OLF_HIDE_FID = 0x0002, + OLF_SHOW_NAME = 0x0004, + OLF_NO_OI = 0x0008, +}; + +struct osd_lf_map { + char *olm_name; + struct lu_fid olm_fid; + __u16 olm_flags; + scandir_t olm_scandir; + filldir_t olm_filldir; +}; + +/* Add the new introduced local files in the list in the future. */ +static const struct osd_lf_map osd_lf_maps[] = { + /* CATALOGS */ + { CATLIST, { FID_SEQ_LOCAL_FILE, LLOG_CATALOGS_OID, 0 }, OLF_SHOW_NAME, + 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 }, + + /* NIDTBL_VERSIONS */ + { MGS_NIDTBL_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, + osd_ios_general_scan, osd_ios_varfid_fill }, + + /* PENDING */ + { "PENDING", { 0, 0, 0 }, 0, NULL, NULL }, + + /* ROOT */ + { "ROOT", { FID_SEQ_ROOT, FID_OID_ROOT, 0 }, + OLF_SCAN_SUBITEMS | OLF_HIDE_FID, osd_ios_ROOT_scan, NULL }, + + /* changelog_catalog */ + { CHANGELOG_CATALOG, { 0, 0, 0 }, 0, NULL, NULL }, + + /* changelog_users */ + { CHANGELOG_USERS, { 0, 0, 0 }, 0, NULL, NULL }, + + /* fld */ + { "fld", { FID_SEQ_LOCAL_FILE, FLD_INDEX_OID, 0 }, OLF_SHOW_NAME, + NULL, NULL }, + + /* last_rcvd */ + { LAST_RCVD, { FID_SEQ_LOCAL_FILE, LAST_RECV_OID, 0 }, OLF_SHOW_NAME, + NULL, NULL }, + + /* lfsck_bookmark */ + { "lfsck_bookmark", { 0, 0, 0 }, 0, NULL, NULL }, + + /* lov_objid */ + { LOV_OBJID, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OID, 0 }, OLF_SHOW_NAME, + NULL, NULL }, + + /* lov_objseq */ + { LOV_OBJSEQ, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OSEQ, 0 }, + OLF_SHOW_NAME, NULL, NULL }, + + /* quota_master */ + { QMT_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, + osd_ios_general_scan, osd_ios_varfid_fill }, + + /* quota_slave */ + { QSD_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, + 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 }, + + /* seq_srv */ + { "seq_srv", { FID_SEQ_LOCAL_FILE, FID_SEQ_SRV_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", { 0, 0, 0 }, 0, NULL, NULL }, + + /* OBJECTS, upgrade from old device */ + { OBJECTS, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, osd_ios_OBJECTS_scan, NULL }, + + /* lquota_v2.user, upgrade from old device */ + { "lquota_v2.user", { 0, 0, 0 }, 0, NULL, NULL }, + + /* 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", { FID_SEQ_LOCAL_FILE, OSD_LPF_OID, 0 }, + OLF_SCAN_SUBITEMS, 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 } +}; + +struct osd_ios_item { + 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; +}; + +static inline struct dentry * +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) { + dput(dentry); + return ERR_PTR(-ENOENT); + } + + return dentry; +} + +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); + + INIT_LIST_HEAD(&item->oii_list); + item->oii_dentry = dget(dentry); + item->oii_scandir = scandir; + item->oii_filldir = filldir; + list_add_tail(&item->oii_list, &dev->od_ios_list); + + RETURN(0); +} + +/** + * osd_ios_scan_one() - check/fix LMA FID and OI entry for one inode + * + * The passed \a inode's \a fid is verified against the LMA FID. If the \a fid + * is NULL or is empty the IGIF FID is used. The FID is verified in the OI to + * reference the inode, or fixed if it is missing or references another inode. + */ +static int +osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev, + struct inode *inode, const struct lu_fid *fid, int flags) +{ + struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs; + struct osd_inode_id *id = &info->oti_id; + struct osd_inode_id *id2 = &info->oti_id2; + struct osd_scrub *scrub = &dev->od_scrub; + struct scrub_file *sf = &scrub->os_file; + struct lu_fid tfid; + int rc; + ENTRY; + + rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma); + 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) + lu_igif_build(&tfid, inode->i_ino, inode->i_generation); + else + tfid = *fid; + 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, 0); + if (rc != 0) { + if (rc != -ENOENT) + RETURN(rc); + + rc = osd_scrub_refresh_mapping(info, dev, &tfid, id, + DTO_INDEX_INSERT, true, 0); + if (rc > 0) + rc = 0; + + RETURN(rc); + } + + if (osd_id_eq_strict(id, id2)) + RETURN(0); + + if (!(sf->sf_flags & SF_INCONSISTENT)) { + osd_scrub_file_reset(scrub, + LDISKFS_SB(osd_sb(dev))->s_es->s_uuid, + SF_INCONSISTENT); + rc = osd_scrub_file_store(scrub); + if (rc != 0) + RETURN(rc); + } + + 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)) { + 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; +} + +static int osd_ios_varfid_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; + struct dentry *child; + int rc; + ENTRY; + + /* skip any '.' started names */ + 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)); + + rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode, + NULL, 0); + if (rc == 0 && S_ISDIR(child->d_inode->i_mode)) + rc = osd_ios_new_item(dev, child, osd_ios_general_scan, + osd_ios_varfid_fill); + dput(child); + + 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) +{ + 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_lf_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)); + + 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); + dput(child); + + RETURN(rc); +} + +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 = { +#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; + ENTRY; + + LASSERT(filldir != NULL); + + filp->f_pos = 0; + filp->f_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); +} + +static int +osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev, + struct dentry *dentry, filldir_t filldir) +{ + struct osd_scrub *scrub = &dev->od_scrub; + struct scrub_file *sf = &scrub->os_file; + struct dentry *child; + int rc; + ENTRY; + + /* 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)) { + rc = PTR_ERR(child); + if (rc == -ENOENT) { + /* It is 1.8 MDT device. */ + if (!(sf->sf_flags & SF_UPGRADE)) { + 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 { + /* 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); + } + + RETURN(rc); +} + +static int +osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev, + struct dentry *dentry, filldir_t filldir) +{ + struct osd_scrub *scrub = &dev->od_scrub; + struct scrub_file *sf = &scrub->os_file; + struct dentry *child; + int rc; ENTRY; - if (preload == 0) { - struct osd_scrub *scrub = &dev->od_scrub; + 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); + } - next = osd_scrub_next; - exec = osd_scrub_exec; - pos = &scrub->os_pos_current; - count = &scrub->os_new_checked; + 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); + dput(child); } else { - struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache; + rc = PTR_ERR(child); + } - next = osd_preload_next; - exec = osd_preload_exec; - pos = &ooc->ooc_pos_preload; - count = &ooc->ooc_cached_items; + if (rc != 0 && rc != -ENOENT) + RETURN(rc); + + child = osd_ios_lookup_one_len(ADMIN_GRP, dentry, strlen(ADMIN_GRP)); + if (!IS_ERR(child)) { + rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0); + dput(child); + } else { + rc = PTR_ERR(child); } - 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; + if (rc == -ENOENT) + rc = 0; - param.bg = (*pos - 1) / LDISKFS_INODES_PER_GROUP(param.sb); - 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, " - "scrub will stop, urgent mode\n", - LDISKFS_SB(param.sb)->s_es->s_volume_name, - (__u32)param.bg); - RETURN(-EIO); - } + RETURN(rc); +} - while (param.offset < LDISKFS_INODES_PER_GROUP(param.sb) && - *count < max) { - rc = next(info, dev, ¶m, &oic, noslot); - switch (rc) { - case SCRUB_NEXT_BREAK: - goto next_group; - case SCRUB_NEXT_EXIT: - brelse(param.bitmap); - RETURN(0); - case SCRUB_NEXT_CRASH: - brelse(param.bitmap); - RETURN(SCRUB_IT_CRASH); - case SCRUB_NEXT_FATAL: - brelse(param.bitmap); - RETURN(-EINVAL); - } +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; + const struct osd_lf_map *map = osd_lf_maps; + int rc; + ENTRY; - rc = exec(info, dev, ¶m, oic, &noslot, rc); - if (rc != 0) { - brelse(param.bitmap); - RETURN(rc); - } + /* 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) { + dput(item->oii_dentry); + OBD_FREE_PTR(item); } -next_group: - brelse(param.bitmap); - } + if (rc != 0) + break; - if (*pos > limit) - RETURN(SCRUB_IT_ALL); - RETURN(0); -} + if (list_empty(&dev->od_ios_list)) + break; -static int osd_scrub_main(void *args) -{ - struct lu_env env; - 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; + item = list_entry(dev->od_ios_list.next, + struct osd_ios_item, oii_list); + list_del_init(&item->oii_list); - cfs_daemonize("OI_scrub"); - rc = lu_env_init(&env, LCT_DT_THREAD); - if (rc != 0) { - CERROR("%.16s: OI scrub, fail to init env, rc = %d\n", - LDISKFS_SB(sb)->s_es->s_volume_name, rc); - GOTO(noenv, rc); + LASSERT(item->oii_scandir != NULL); + scandir = item->oii_scandir; + filldir = item->oii_filldir; + dentry = item->oii_dentry; } - 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); - GOTO(out, rc); + 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); } - if (!scrub->os_full_speed) { - struct l_wait_info lwi = { 0 }; - struct osd_otable_it *it = dev->od_otable_it; - struct osd_otable_cache *ooc = &it->ooi_cache; + if (rc != 0) + RETURN(rc); - l_wait_event(thread->t_ctl_waitq, - it->ooi_user_ready || !thread_is_running(thread), - &lwi); - if (unlikely(!thread_is_running(thread))) - GOTO(post, rc = 0); + /* 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; - LASSERT(scrub->os_pos_current >= ooc->ooc_pos_preload); - scrub->os_pos_current = ooc->ooc_pos_preload; - } + if (fid_is_zero(&map->olm_fid)) { + map++; + continue; + } - CDEBUG(D_LFSCK, "OI scrub: flags = 0x%x, pos = %u\n", - scrub->os_start_flags, scrub->os_pos_current); + 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++; + } - rc = osd_inode_iteration(osd_oti_get(&env), dev, ~0U, 0); - if (unlikely(rc == SCRUB_IT_CRASH)) - GOTO(out, rc = -EINVAL); - GOTO(post, rc); + RETURN(0); +} -post: - osd_scrub_post(scrub, rc); - CDEBUG(D_LFSCK, "OI scrub: stop, rc = %d, pos = %u\n", - rc, scrub->os_pos_current); +char *osd_lf_fid2name(const struct lu_fid *fid) +{ + const struct osd_lf_map *map = osd_lf_maps; -out: - while (!cfs_list_empty(&scrub->os_inconsistent_items)) { - struct osd_inconsistent_item *oii; + while (map->olm_name != NULL) { + if (!lu_fid_eq(fid, &map->olm_fid)) { + map++; + continue; + } - oii = cfs_list_entry(scrub->os_inconsistent_items.next, - struct osd_inconsistent_item, oii_list); - cfs_list_del_init(&oii->oii_list); - OBD_FREE_PTR(oii); + if (map->olm_flags & OLF_SHOW_NAME) + return map->olm_name; + else + return ""; } - lu_env_fini(&env); -noenv: - cfs_spin_lock(&scrub->os_lock); - thread_set_flags(thread, SVC_STOPPED); - cfs_waitq_broadcast(&thread->t_ctl_waitq); - cfs_spin_unlock(&scrub->os_lock); - return rc; + return NULL; } +/* OI scrub start/stop */ + 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 */ - cfs_spin_lock(&scrub->os_lock); + spin_lock(&scrub->os_lock); + +again: if (thread_is_running(thread)) { - cfs_spin_unlock(&scrub->os_lock); - RETURN(-EALREADY); - } else if (unlikely(thread_is_stopping(thread))) { - cfs_spin_unlock(&scrub->os_lock); + spin_unlock(&scrub->os_lock); + if (!(scrub->os_file.sf_flags & SF_AUTO) || + (flags & (SS_AUTO_FULL | 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; } - cfs_spin_unlock(&scrub->os_lock); + spin_unlock(&scrub->os_lock); + + 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 = cfs_create_thread(osd_scrub_main, dev, 0); - if (rc < 0) { - 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); } @@ -937,19 +2268,15 @@ again: RETURN(0); } -int osd_scrub_start(struct osd_device *dev) +int osd_scrub_start(struct osd_device *dev, __u32 flags) { - __u32 flags = SS_AUTO; - int rc; + int rc; ENTRY; - if (dev->od_scrub.os_file.sf_status == SS_COMPLETED) - flags |= SS_RESET; - /* od_otable_mutex: prevent curcurrent start/stop */ - cfs_mutex_lock(&dev->od_otable_mutex); + mutex_lock(&dev->od_otable_mutex); rc = do_osd_scrub_start(dev, flags); - cfs_mutex_unlock(&dev->od_otable_mutex); + mutex_unlock(&dev->od_otable_mutex); RETURN(rc == -EALREADY ? 0 : rc); } @@ -960,29 +2287,32 @@ static void do_osd_scrub_stop(struct osd_scrub *scrub) struct l_wait_info lwi = { 0 }; /* os_lock: sync status between stop and scrub thread */ - cfs_spin_lock(&scrub->os_lock); + spin_lock(&scrub->os_lock); if (!thread_is_init(thread) && !thread_is_stopped(thread)) { thread_set_flags(thread, SVC_STOPPING); - cfs_spin_unlock(&scrub->os_lock); - cfs_waitq_broadcast(&thread->t_ctl_waitq); + spin_unlock(&scrub->os_lock); + wake_up_all(&thread->t_ctl_waitq); l_wait_event(thread->t_ctl_waitq, thread_is_stopped(thread), &lwi); /* Do not skip the last lock/unlock, which can guarantee that * the caller cannot return until the OI scrub thread exit. */ - cfs_spin_lock(&scrub->os_lock); + spin_lock(&scrub->os_lock); } - cfs_spin_unlock(&scrub->os_lock); + spin_unlock(&scrub->os_lock); } static void osd_scrub_stop(struct osd_device *dev) { /* od_otable_mutex: prevent curcurrent start/stop */ - cfs_mutex_lock(&dev->od_otable_mutex); + mutex_lock(&dev->od_otable_mutex); + dev->od_scrub.os_paused = 1; do_osd_scrub_stop(&dev->od_scrub); - cfs_mutex_unlock(&dev->od_otable_mutex); + mutex_unlock(&dev->od_otable_mutex); } +/* OI scrub setup/cleanup */ + static const char osd_scrub_name[] = "OI_scrub"; int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev) @@ -991,45 +2321,69 @@ int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev) struct osd_scrub *scrub = &dev->od_scrub; struct lvfs_run_ctxt *ctxt = &scrub->os_ctxt; struct scrub_file *sf = &scrub->os_file; - struct osd_inode_id *id = &scrub->os_oic.oic_lid; struct super_block *sb = osd_sb(dev); struct ldiskfs_super_block *es = LDISKFS_SB(sb)->s_es; - struct inode *inode; struct lvfs_run_ctxt saved; struct file *filp; + struct inode *inode; + struct lu_fid *fid = &info->oti_fid; int dirty = 0; - int init = 0; int rc = 0; ENTRY; + memset(scrub, 0, sizeof(*scrub)); OBD_SET_CTXT_MAGIC(ctxt); ctxt->pwdmnt = dev->od_mnt; ctxt->pwd = dev->od_mnt->mnt_root; ctxt->fs = get_ds(); - cfs_waitq_init(&scrub->os_thread.t_ctl_waitq); - cfs_init_rwsem(&scrub->os_rwsem); - cfs_spin_lock_init(&scrub->os_lock); - CFS_INIT_LIST_HEAD(&scrub->os_inconsistent_items); - if (get_mount_flags(dev->od_mount->lmi_sb) & LMD_FLG_NOSCRUB) - scrub->os_no_scrub = 1; + init_waitqueue_head(&scrub->os_thread.t_ctl_waitq); + init_rwsem(&scrub->os_rwsem); + spin_lock_init(&scrub->os_lock); + 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); - filp_close(filp, 0); - pop_ctxt(&saved, ctxt, NULL); + 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, NULL); + pop_ctxt(&saved, ctxt); + RETURN(rc); + } + + 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); + /* 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; - init = 1; } else if (rc != 0) { - RETURN(rc); + GOTO(cleanup_inode, rc); } else { if (memcmp(sf->sf_uuid, es->s_uuid, 16) != 0) { osd_scrub_file_reset(scrub, es->s_uuid,SF_INCONSISTENT); @@ -1043,47 +2397,72 @@ int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev) if (sf->sf_pos_last_checkpoint != 0) scrub->os_pos_current = sf->sf_pos_last_checkpoint + 1; else - scrub->os_pos_current = LDISKFS_FIRST_INO(sb); + scrub->os_pos_current = LDISKFS_FIRST_INO(sb) + 1; if (dirty != 0) { rc = osd_scrub_file_store(scrub); if (rc != 0) - RETURN(rc); + GOTO(cleanup_inode, rc); } /* Initialize OI files. */ rc = osd_oi_init(info, dev); if (rc < 0) - RETURN(rc); + GOTO(cleanup_inode, rc); - if (init != 0) { - rc = __osd_oi_lookup(info, dev, &LU_DOT_LUSTRE_FID, id); - if (rc == 0) { - inode = osd_iget(info, dev, id); - if (IS_ERR(inode)) { - rc = PTR_ERR(inode); - /* It is restored from old 2.x backup. */ - if (rc == -ENOENT || rc == -ESTALE) { - osd_scrub_file_reset(scrub, es->s_uuid, - SF_INCONSISTENT); - rc = osd_scrub_file_store(scrub); - } - } else { - iput(inode); - } - } else if (rc == -ENOENT) { - rc = 0; - } + rc = osd_initial_OI_scrub(info, dev); + if (rc != 0) + GOTO(cleanup_oi, rc); + + 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 (rc == 0 && !scrub->os_no_scrub && - ((sf->sf_status == SS_CRASHED && - sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_AUTO)) || + 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)))) - rc = osd_scrub_start(dev); + sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | + SF_UPGRADE)))) + rc = osd_scrub_start(dev, SS_AUTO_FULL); - RETURN(rc); + 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) @@ -1101,6 +2480,8 @@ void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev) osd_oi_fini(osd_oti_get(env), 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) @@ -1115,7 +2496,7 @@ static struct dt_it *osd_otable_it_init(const struct lu_env *env, ENTRY; /* od_otable_mutex: prevent curcurrent init/fini */ - cfs_mutex_lock(&dev->od_otable_mutex); + mutex_lock(&dev->od_otable_mutex); if (dev->od_otable_it != NULL) GOTO(out, it = ERR_PTR(-EALREADY)); @@ -1139,21 +2520,26 @@ 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 (rc == -EALREADY) { - it->ooi_cache.ooc_pos_preload = scrub->os_pos_current - 1; - } else if (rc < 0) { + 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); - GOTO(out, it = ERR_PTR(-EALREADY)); - } else { - it->ooi_cache.ooc_pos_preload = scrub->os_pos_current; + GOTO(out, it = ERR_PTR(rc)); } + it->ooi_cache.ooc_pos_preload = scrub->os_pos_current; + GOTO(out, it); out: - cfs_mutex_unlock(&dev->od_otable_mutex); + mutex_unlock(&dev->od_otable_mutex); return (struct dt_it *)it; } @@ -1163,72 +2549,38 @@ static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di) struct osd_device *dev = it->ooi_dev; /* od_otable_mutex: prevent curcurrent init/fini */ - cfs_mutex_lock(&dev->od_otable_mutex); + mutex_lock(&dev->od_otable_mutex); do_osd_scrub_stop(&dev->od_scrub); LASSERT(dev->od_otable_it == it); dev->od_otable_it = NULL; - cfs_mutex_unlock(&dev->od_otable_mutex); + mutex_unlock(&dev->od_otable_mutex); OBD_FREE_PTR(it); } -/** - * Set the OSD layer iteration start position as the specified key. - * - * The LFSCK out of OSD layer does not know the detail of the key, so if there - * are several keys, they cannot be compared out of OSD, so call "::get()" for - * each key, and OSD will select the smallest one by itself. - */ static int osd_otable_it_get(const struct lu_env *env, struct dt_it *di, const struct dt_key *key) { - struct osd_otable_it *it = (struct osd_otable_it *)di; - struct osd_otable_cache *ooc = &it->ooi_cache; - const char *str = (const char *)key; - __u32 ino; - ENTRY; - - /* Forbid to set iteration position after iteration started. */ - if (it->ooi_user_ready) - RETURN(-EPERM); - - if (str[0] == '\0') - RETURN(-EINVAL); - - if (sscanf(str, "%u", &ino) <= 0) - RETURN(-EINVAL); - - /* Skip the one that has been processed last time. */ - if (ooc->ooc_pos_preload > ++ino) - ooc->ooc_pos_preload = ino; - - RETURN(0); + return 0; } -static int osd_otable_it_preload(const struct lu_env *env, - struct osd_otable_it *it) +static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di) { - struct osd_device *dev = it->ooi_dev; - struct osd_scrub *scrub = &dev->od_scrub; - struct osd_otable_cache *ooc = &it->ooi_cache; - int rc; - ENTRY; - - rc = osd_inode_iteration(osd_oti_get(env), dev, - OSD_OTABLE_IT_CACHE_SIZE, 1); - 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)) { - scrub->os_waiting = 0; - cfs_waitq_broadcast(&scrub->os_thread.t_ctl_waitq); - } +static inline int +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 || + !thread_is_running(&scrub->os_thread)) + it->ooi_waiting = 0; + else + it->ooi_waiting = 1; + spin_unlock(&scrub->os_lock); - RETURN(rc < 0 ? rc : ooc->ooc_cached_items); + return !it->ooi_waiting; } static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di) @@ -1262,12 +2614,17 @@ again: RETURN(1); } - it->ooi_waiting = 1; - l_wait_event(thread->t_ctl_waitq, - ooc->ooc_pos_preload < scrub->os_pos_current || - !thread_is_running(thread), - &lwi); - it->ooi_waiting = 0; + if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) { + spin_lock(&scrub->os_lock); + scrub->os_waiting = 0; + wake_up_all(&scrub->os_thread.t_ctl_waitq); + spin_unlock(&scrub->os_lock); + } + + if (it->ooi_cache.ooc_pos_preload >= scrub->os_pos_current) + l_wait_event(thread->t_ctl_waitq, + osd_otable_it_wakeup(scrub, it), + &lwi); if (!thread_is_running(thread) && !it->ooi_used_outside) RETURN(1); @@ -1282,18 +2639,13 @@ again: static struct dt_key *osd_otable_it_key(const struct lu_env *env, const struct dt_it *di) { - struct osd_otable_it *it = (struct osd_otable_it *)di; - struct osd_otable_cache *ooc = &it->ooi_cache; - - sprintf(it->ooi_key, "%u", - ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino); - return (struct dt_key *)it->ooi_key; + return NULL; } static int osd_otable_it_key_size(const struct lu_env *env, const struct dt_it *di) { - return sizeof(((struct osd_otable_it *)di)->ooi_key); + return sizeof(__u64); } static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di, @@ -1303,9 +2655,33 @@ static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di, struct osd_otable_cache *ooc = &it->ooi_cache; *(struct lu_fid *)rec = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_fid; + + /* Filter out Invald FID already. */ + LASSERTF(fid_is_sane((struct lu_fid *)rec), + "Invalid FID "DFID", p_idx = %d, c_idx = %d\n", + PFID((struct lu_fid *)rec), + ooc->ooc_producer_idx, ooc->ooc_consumer_idx); + return 0; } +static __u64 osd_otable_it_store(const struct lu_env *env, + const struct dt_it *di) +{ + struct osd_otable_it *it = (struct osd_otable_it *)di; + struct osd_otable_cache *ooc = &it->ooi_cache; + __u64 hash; + + if (it->ooi_user_ready && ooc->ooc_consumer_idx != -1) + hash = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino; + else + hash = ooc->ooc_pos_preload; + return hash; +} + +/** + * Set the OSD layer iteration start position as the specified hash. + */ static int osd_otable_it_load(const struct lu_env *env, const struct dt_it *di, __u64 hash) { @@ -1313,18 +2689,34 @@ static int osd_otable_it_load(const struct lu_env *env, struct osd_device *dev = it->ooi_dev; struct osd_otable_cache *ooc = &it->ooi_cache; struct osd_scrub *scrub = &dev->od_scrub; + int rc; + ENTRY; + /* Forbid to set iteration position after iteration started. */ if (it->ooi_user_ready) - return 0; + RETURN(-EPERM); + + if (hash > OSD_OTABLE_MAX_HASH) + hash = OSD_OTABLE_MAX_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; - if (ooc->ooc_pos_preload < LDISKFS_FIRST_INO(osd_sb(dev))) - ooc->ooc_pos_preload = LDISKFS_FIRST_INO(osd_sb(dev)); 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. */ - return osd_otable_it_next(env, (struct dt_it *)it); + rc = osd_otable_it_next(env, (struct dt_it *)it); + + RETURN(rc); +} + +static int osd_otable_it_key_rec(const struct lu_env *env, + const struct dt_it *di, void *key_rec) +{ + return 0; } const struct dt_index_operations osd_otable_ops = { @@ -1332,14 +2724,21 @@ const struct dt_index_operations osd_otable_ops = { .init = osd_otable_it_init, .fini = osd_otable_it_fini, .get = osd_otable_it_get, + .put = osd_otable_it_put, .next = osd_otable_it_next, - .key = osd_otable_it_key, + .key = osd_otable_it_key, .key_size = osd_otable_it_key_size, .rec = osd_otable_it_rec, + .store = osd_otable_it_store, .load = osd_otable_it_load, + .key_rec = osd_otable_it_key_rec, } }; +/* 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) { @@ -1353,24 +2752,44 @@ 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; - cfs_spin_lock(&scrub->os_lock); + 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))) { - cfs_spin_unlock(&scrub->os_lock); + spin_unlock(&scrub->os_lock); OBD_FREE_PTR(oii); 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); - cfs_spin_unlock(&scrub->os_lock); + 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); } @@ -1382,24 +2801,27 @@ int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid, struct osd_inconsistent_item *oii; ENTRY; - cfs_spin_lock(&scrub->os_lock); - cfs_list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) { + spin_lock(&scrub->os_lock); + 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; - cfs_spin_unlock(&scrub->os_lock); + spin_unlock(&scrub->os_lock); RETURN(0); } } - cfs_spin_unlock(&scrub->os_lock); + spin_unlock(&scrub->os_lock); RETURN(-ENOENT); } +/* OI scrub dump */ + static const char *scrub_status_names[] = { "init", "scanning", "completed", "failed", + "stopped", "paused", "crashed", NULL @@ -1409,158 +2831,140 @@ static const char *scrub_flags_names[] = { "recreated", "inconsistent", "auto", + "upgrade", NULL }; 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; - cfs_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) + down_read(&scrub->os_rwsem); + 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" + "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_success_count); - if (rc <= 0) + 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; 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); @@ -1568,30 +2972,31 @@ 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: - cfs_up_read(&scrub->os_rwsem); - return ret; + up_read(&scrub->os_rwsem); + return (rc < 0 ? -ENOSPC : 0); }