X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fosd-ldiskfs%2Fosd_scrub.c;h=628771a0e8bfb714693977d62bbba9c9c376a889;hp=3cd4e8c7f820adafa16d14943145e61e358dca17;hb=1cab7fafd016621160f49030ea46f155fe34d70b;hpb=b384b414d88de3cfa4966e731b46e85f62c8e438 diff --git a/lustre/osd-ldiskfs/osd_scrub.c b/lustre/osd-ldiskfs/osd_scrub.c index 3cd4e8c..628771a 100644 --- a/lustre/osd-ldiskfs/osd_scrub.c +++ b/lustre/osd-ldiskfs/osd_scrub.c @@ -6,21 +6,21 @@ * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 only, * as published by the Free Software Foundation. - - * This program is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License version 2 for more details. A copy is - * included in the COPYING file that accompanied this code. - + * + * This program is distributed in the hope that it will be useful, but + * WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * General Public License version 2 for more details (a copy is included + * in the LICENSE file that accompanied this code). + * * You should have received a copy of the GNU General Public License - * along with this program; if not, write to the Free Software - * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA + * version 2 along with this program; If not, see + * http://www.gnu.org/licenses/gpl-2.0.html * * GPL HEADER END */ /* - * Copyright (c) 2012, 2015, Intel Corporation. + * Copyright (c) 2012, 2017, Intel Corporation. */ /* * lustre/osd-ldiskfs/osd_scrub.c @@ -38,54 +38,25 @@ #define DEBUG_SUBSYSTEM S_LFSCK #include -#include +#include #include #include #include +#include +#include #include "osd_internal.h" #include "osd_oi.h" #include "osd_scrub.h" -#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) -{ - return container_of0(scrub, struct osd_device, od_scrub); -} - -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, +static inline int osd_scrub_has_window(struct lustre_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 * @@ -104,7 +75,7 @@ static int osd_scrub_refresh_mapping(struct osd_thread_info *info, int rc; ENTRY; - if (dev->od_scrub.os_file.sf_param & SP_DRYRUN && !force) + if (dev->od_scrub.os_scrub.os_file.sf_param & SP_DRYRUN && !force) RETURN(0); /* DTO_INDEX_INSERT is enough for other two ops: @@ -115,7 +86,8 @@ static int osd_scrub_refresh_mapping(struct osd_thread_info *info, 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); + PFID(fid), id ? id->oii_ino : -1, id ? id->oii_gen : -1, + rc); RETURN(rc); } @@ -187,200 +159,20 @@ static int osd_scrub_refresh_mapping(struct osd_thread_info *info, RETURN(rc); } -/* OI_scrub file ops */ - -static void osd_scrub_file_to_cpu(struct scrub_file *des, - struct scrub_file *src) -{ - memcpy(des->sf_uuid, src->sf_uuid, 16); - des->sf_flags = le64_to_cpu(src->sf_flags); - des->sf_magic = le32_to_cpu(src->sf_magic); - des->sf_status = le16_to_cpu(src->sf_status); - des->sf_param = le16_to_cpu(src->sf_param); - des->sf_time_last_complete = - le64_to_cpu(src->sf_time_last_complete); - des->sf_time_latest_start = - le64_to_cpu(src->sf_time_latest_start); - des->sf_time_last_checkpoint = - le64_to_cpu(src->sf_time_last_checkpoint); - des->sf_pos_latest_start = - le64_to_cpu(src->sf_pos_latest_start); - des->sf_pos_last_checkpoint = - le64_to_cpu(src->sf_pos_last_checkpoint); - des->sf_pos_first_inconsistent = - le64_to_cpu(src->sf_pos_first_inconsistent); - des->sf_items_checked = - le64_to_cpu(src->sf_items_checked); - des->sf_items_updated = - le64_to_cpu(src->sf_items_updated); - des->sf_items_failed = - le64_to_cpu(src->sf_items_failed); - des->sf_items_updated_prior = - le64_to_cpu(src->sf_items_updated_prior); - 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); -} - -static void osd_scrub_file_to_le(struct scrub_file *des, - struct scrub_file *src) -{ - memcpy(des->sf_uuid, src->sf_uuid, 16); - des->sf_flags = cpu_to_le64(src->sf_flags); - des->sf_magic = cpu_to_le32(src->sf_magic); - des->sf_status = cpu_to_le16(src->sf_status); - des->sf_param = cpu_to_le16(src->sf_param); - des->sf_time_last_complete = - cpu_to_le64(src->sf_time_last_complete); - des->sf_time_latest_start = - cpu_to_le64(src->sf_time_latest_start); - des->sf_time_last_checkpoint = - cpu_to_le64(src->sf_time_last_checkpoint); - des->sf_pos_latest_start = - cpu_to_le64(src->sf_pos_latest_start); - des->sf_pos_last_checkpoint = - cpu_to_le64(src->sf_pos_last_checkpoint); - des->sf_pos_first_inconsistent = - cpu_to_le64(src->sf_pos_first_inconsistent); - des->sf_items_checked = - cpu_to_le64(src->sf_items_checked); - des->sf_items_updated = - cpu_to_le64(src->sf_items_updated); - des->sf_items_failed = - cpu_to_le64(src->sf_items_failed); - des->sf_items_updated_prior = - cpu_to_le64(src->sf_items_updated_prior); - 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); -} - -static void osd_scrub_file_init(struct osd_scrub *scrub, __u8 *uuid) -{ - struct scrub_file *sf = &scrub->os_file; - - memset(sf, 0, sizeof(*sf)); - memcpy(sf->sf_uuid, uuid, 16); - sf->sf_magic = SCRUB_MAGIC_V1; - sf->sf_status = SS_INIT; -} - -void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags) -{ - struct scrub_file *sf = &scrub->os_file; - - 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_flags &= ~SF_AUTO; - sf->sf_run_time = 0; - sf->sf_time_latest_start = 0; - sf->sf_time_last_checkpoint = 0; - sf->sf_pos_latest_start = 0; - sf->sf_pos_last_checkpoint = 0; - sf->sf_pos_first_inconsistent = 0; - sf->sf_items_checked = 0; - sf->sf_items_updated = 0; - sf->sf_items_failed = 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; - int len = sizeof(scrub->os_file_disk); - int rc; - - rc = osd_ldiskfs_read(scrub->os_inode, &scrub->os_file_disk, len, &pos); - if (rc == len) { - struct scrub_file *sf = &scrub->os_file; - - osd_scrub_file_to_cpu(sf, &scrub->os_file_disk); - if (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) { - 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 { - /* return -ENOENT for empty scrub file case. */ - rc = -ENOENT; - } - - return rc; -} - -int osd_scrub_file_store(struct osd_scrub *scrub) -{ - struct osd_device *dev; - handle_t *jh; - loff_t pos = 0; - int len = sizeof(scrub->os_file_disk); - int credits; - int rc; - - 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 = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, credits); - if (IS_ERR(jh)) { - rc = PTR_ERR(jh); - 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, 0, &pos, jh); - ldiskfs_journal_stop(jh); - if (rc != 0) - 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_convert_ff(struct osd_thread_info *info, struct osd_device *dev, struct inode *inode, const struct lu_fid *fid) { - struct filter_fid_old *ff = &info->oti_ff; - struct dentry *dentry = &info->oti_obj_dentry; - struct lu_fid *tfid = &info->oti_fid; - handle_t *jh; - int size = 0; - int rc; - bool removed = false; - bool reset = true; + struct filter_fid_18_23 *ff = &info->oti_ff_old; + struct dentry *dentry = &info->oti_obj_dentry; + struct lu_fid *tfid = &info->oti_fid; + bool fid_18_23 = false; + handle_t *jh; + int size = 0; + int rc; ENTRY; - if (dev->od_scrub.os_file.sf_param & SP_DRYRUN) + if (dev->od_scrub.os_scrub.os_file.sf_param & SP_DRYRUN) RETURN(0); if (fid_is_idif(fid) && dev->od_index_in_idif == 0) { @@ -415,25 +207,26 @@ osd_scrub_convert_ff(struct osd_thread_info *info, struct osd_device *dev, 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) + dquot_initialize(inode); + rc = osd_removexattr(dentry, inode, XATTR_NAME_FID); + if (rc) GOTO(stop, rc); - removed = true; - } else if (unlikely(rc == -ENODATA)) { - reset = false; - } else if (rc != sizeof(struct filter_fid)) { + fid_18_23 = true; + } else if (rc != -ENODATA && rc < (int)sizeof(struct filter_fid_24_29)) { GOTO(stop, rc = -EINVAL); } /* 3) make new LMA and add it */ rc = osd_ea_fid_set(info, inode, tfid, LMAC_FID_ON_OST, 0); - if (rc == 0 && reset) - size = sizeof(struct filter_fid); - else if (rc != 0 && removed) - /* If failed, we should try to add the old back. */ - size = sizeof(struct filter_fid_old); + if (fid_18_23) { + if (rc) + /* If failed, we should try to add the old back. */ + size = sizeof(*ff); + else + /* The new PFID EA will only contains ::ff_parent */ + size = sizeof(ff->ff_parent); + } /* 4) generate new XATTR_NAME_FID with the saved parent FID and add it*/ if (size > 0) { @@ -459,7 +252,7 @@ static int osd_scrub_check_update(struct osd_thread_info *info, struct osd_device *dev, struct osd_idmap_cache *oic, int val) { - struct osd_scrub *scrub = &dev->od_scrub; + struct lustre_scrub *scrub = &dev->od_scrub.os_scrub; struct scrub_file *sf = &scrub->os_file; struct lu_fid *fid = &oic->oic_fid; struct osd_inode_id *lid = &oic->oic_lid; @@ -641,7 +434,10 @@ out: iput(inode); if (oii != NULL) { - LASSERT(list_empty(&oii->oii_list)); + spin_lock(&scrub->os_lock); + if (likely(!list_empty(&oii->oii_list))) + list_del(&oii->oii_list); + spin_unlock(&scrub->os_lock); OBD_FREE_PTR(oii); } @@ -649,11 +445,9 @@ out: RETURN(sf->sf_param & SP_FAILOUT ? rc : 0); } -/* OI scrub APIs */ - -static int osd_scrub_prep(struct osd_device *dev) +static int osd_scrub_prep(const struct lu_env *env, struct osd_device *dev) { - struct osd_scrub *scrub = &dev->od_scrub; + struct lustre_scrub *scrub = &dev->od_scrub.os_scrub; struct ptlrpc_thread *thread = &scrub->os_thread; struct scrub_file *sf = &scrub->os_file; __u32 flags = scrub->os_start_flags; @@ -661,7 +455,7 @@ static int osd_scrub_prep(struct osd_device *dev) bool drop_dryrun = false; ENTRY; - CDEBUG(D_LFSCK, "%.16s: OI scrub prep, flags = 0x%x\n", + CDEBUG(D_LFSCK, "%s: OI scrub prep, flags = 0x%x\n", osd_scrub2name(scrub), flags); down_write(&scrub->os_rwsem); @@ -678,8 +472,7 @@ static int osd_scrub_prep(struct osd_device *dev) } if (flags & SS_RESET) - osd_scrub_file_reset(scrub, - LDISKFS_SB(osd_sb(dev))->s_es->s_uuid, 0); + scrub_file_reset(scrub, dev->od_uuid, 0); if (flags & SS_AUTO_FULL) { scrub->os_full_speed = 1; @@ -715,10 +508,10 @@ static int osd_scrub_prep(struct osd_device *dev) scrub->os_pos_current = sf->sf_pos_latest_start; sf->sf_status = SS_SCANNING; - sf->sf_time_latest_start = cfs_time_current_sec(); + sf->sf_time_latest_start = ktime_get_real_seconds(); sf->sf_time_last_checkpoint = sf->sf_time_latest_start; sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1; - rc = osd_scrub_file_store(scrub); + rc = scrub_file_store(env, scrub); if (rc == 0) { spin_lock(&scrub->os_lock); thread_set_flags(thread, SVC_RUNNING); @@ -730,36 +523,15 @@ static int osd_scrub_prep(struct osd_device *dev) RETURN(rc); } -static int osd_scrub_checkpoint(struct osd_scrub *scrub) -{ - struct scrub_file *sf = &scrub->os_file; - int rc; - - 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; - sf->sf_time_last_checkpoint = cfs_time_current_sec(); - sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC - - scrub->os_time_last_checkpoint); - rc = osd_scrub_file_store(scrub); - up_write(&scrub->os_rwsem); - - return rc; -} - -static int osd_scrub_post(struct osd_scrub *scrub, int result) +static int osd_scrub_post(const struct lu_env *env, struct osd_device *dev, + int result) { + struct lustre_scrub *scrub = &dev->od_scrub.os_scrub; struct scrub_file *sf = &scrub->os_file; int rc; ENTRY; - CDEBUG(D_LFSCK, "%.16s: OI scrub post, result = %d\n", + CDEBUG(D_LFSCK, "%s: OI scrub post with result = %d\n", osd_scrub2name(scrub), result); down_write(&scrub->os_rwsem); @@ -771,11 +543,8 @@ static int osd_scrub_post(struct osd_scrub *scrub, int result) scrub->os_new_checked = 0; sf->sf_pos_last_checkpoint = scrub->os_pos_current; } - sf->sf_time_last_checkpoint = cfs_time_current_sec(); + sf->sf_time_last_checkpoint = ktime_get_real_seconds(); if (result > 0) { - 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; @@ -794,9 +563,10 @@ static int osd_scrub_post(struct osd_scrub *scrub, int result) } else { sf->sf_status = SS_FAILED; } - sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC - - scrub->os_time_last_checkpoint); - rc = osd_scrub_file_store(scrub); + sf->sf_run_time += ktime_get_seconds() - + scrub->os_time_last_checkpoint; + + rc = scrub_file_store(env, scrub); up_write(&scrub->os_rwsem); RETURN(rc < 0 ? rc : result); @@ -804,14 +574,6 @@ static int osd_scrub_post(struct osd_scrub *scrub, int result) /* iteration engine */ -struct osd_iit_param { - struct super_block *sb; - struct buffer_head *bitmap; - ldiskfs_group_t bg; - __u32 gbase; - __u32 offset; -}; - typedef int (*osd_iit_next_policy)(struct osd_thread_info *info, struct osd_device *dev, struct osd_iit_param *param, @@ -824,17 +586,30 @@ typedef int (*osd_iit_exec_policy)(struct osd_thread_info *info, struct osd_idmap_cache *oic, bool *noslot, int rc); -static int osd_iit_next(struct osd_iit_param *param, __u32 *pos) +static int osd_iit_next(struct osd_iit_param *param, __u64 *pos) { + __u32 offset; + +again: param->offset = ldiskfs_find_next_bit(param->bitmap->b_data, LDISKFS_INODES_PER_GROUP(param->sb), param->offset); if (param->offset >= LDISKFS_INODES_PER_GROUP(param->sb)) { *pos = 1 + (param->bg+1) * LDISKFS_INODES_PER_GROUP(param->sb); return SCRUB_NEXT_BREAK; - } else { - *pos = param->gbase + param->offset; - return 0; } + + offset = param->offset++; + if (unlikely(*pos == param->gbase + offset && *pos != param->start)) { + /* We should NOT find the same object more than once. */ + CERROR("%s: scan the same object multiple times at the pos: " + "group = %u, base = %u, offset = %u, start = %u\n", + param->sb->s_id, (__u32)param->bg, param->gbase, + offset, param->start); + goto again; + } + + *pos = param->gbase + offset; + return 0; } /** @@ -851,19 +626,20 @@ static int osd_scrub_check_local_fldb(struct osd_thread_info *info, * quite possible for FID-on-MDT. */ if (dev->od_is_ost) return SCRUB_NEXT_OSTOBJ_OLD; - else - return 0; + + 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; + struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma; + bool has_lma = false; + int rc; - rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma); + rc = osd_get_lma(info, inode, &info->oti_obj_dentry, + &info->oti_ost_attrs); if (rc == 0) { has_lma = true; if (lma->lma_compat & LMAC_NOT_IN_OI || @@ -897,7 +673,7 @@ static int osd_scrub_get_fid(struct osd_thread_info *info, 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. */ + /* It is 2.3 or older OST-object. */ rc = SCRUB_NEXT_OSTOBJ_OLD; return rc; } @@ -908,7 +684,7 @@ static int osd_scrub_get_fid(struct osd_thread_info *info, * to generate its FID, ignore it directly. */ rc = SCRUB_NEXT_CONTINUE; else - /* It is 2.4 OST-object. */ + /* It is 2.4 or newer OST-object. */ rc = SCRUB_NEXT_OSTOBJ_OLD; return rc; } @@ -917,7 +693,7 @@ static int osd_scrub_get_fid(struct osd_thread_info *info, return rc; if (!has_lma) { - if (dev->od_scrub.os_convert_igif) { + if (dev->od_scrub.os_scrub.os_convert_igif) { lu_igif_build(fid, inode->i_ino, inode->i_generation); if (scrub) @@ -953,9 +729,16 @@ static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev, /* 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))) + is_remote_parent_ino(dev, pos))) RETURN(SCRUB_NEXT_CONTINUE); + /* Skip project quota inode since it is greater than s_first_ino. */ +#ifdef HAVE_PROJECT_QUOTA + if (ldiskfs_has_feature_project(sb) && + pos == le32_to_cpu(LDISKFS_SB(sb)->s_es->s_prj_quota_inum)) + RETURN(SCRUB_NEXT_CONTINUE); +#endif + osd_id_gen(lid, pos, OSD_OII_NOGEN); inode = osd_iget(info, dev, lid); if (IS_ERR(inode)) { @@ -965,9 +748,8 @@ static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev, if (rc == -ENOENT || rc == -ESTALE) RETURN(SCRUB_NEXT_CONTINUE); - CDEBUG(D_LFSCK, "%.16s: fail to read inode, ino# = %u: " - "rc = %d\n", LDISKFS_SB(sb)->s_es->s_volume_name, - pos, rc); + CDEBUG(D_LFSCK, "%s: fail to read inode, ino# = %u: " + "rc = %d\n", osd_dev2name(dev), pos, rc); RETURN(rc); } @@ -995,22 +777,18 @@ 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 lustre_scrub *scrub = &dev->od_scrub.os_scrub; struct ptlrpc_thread *thread = &scrub->os_thread; - struct lu_fid *fid; - struct osd_inode_id *lid; - int rc; - - if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && cfs_fail_val > 0) { - struct l_wait_info lwi; + struct lu_fid *fid; + struct osd_inode_id *lid; + int rc; - lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val), NULL, NULL); - 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_DELAY) && cfs_fail_val > 0) + wait_event_idle_timeout( + thread->t_ctl_waitq, + !list_empty(&scrub->os_inconsistent_items) || + !thread_is_running(thread), + cfs_time_seconds(cfs_fail_val)); if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) { spin_lock(&scrub->os_lock); @@ -1032,7 +810,6 @@ static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev, oii = list_entry(scrub->os_inconsistent_items.next, struct osd_inconsistent_item, oii_list); - list_del_init(&oii->oii_list); spin_unlock(&scrub->os_lock); *oic = &oii->oii_cache; @@ -1050,7 +827,7 @@ static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev, if (rc != 0) return rc; - *oic = &scrub->os_oic; + *oic = &dev->od_scrub.os_oic; fid = &(*oic)->oic_fid; lid = &(*oic)->oic_lid; rc = osd_iit_iget(info, dev, fid, lid, @@ -1062,33 +839,28 @@ static int osd_preload_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_otable_cache *ooc = &dev->od_otable_it->ooi_cache; - struct osd_scrub *scrub; - struct ptlrpc_thread *thread; - int rc; - - rc = osd_iit_next(param, &ooc->ooc_pos_preload); - if (rc != 0) - return rc; + struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache; + struct lustre_scrub *scrub = &dev->od_scrub.os_scrub; + struct ptlrpc_thread *thread = &scrub->os_thread; + int rc; - scrub = &dev->od_scrub; - thread = &scrub->os_thread; if (thread_is_running(thread) && ooc->ooc_pos_preload >= scrub->os_pos_current) return SCRUB_NEXT_EXIT; + rc = osd_iit_next(param, &ooc->ooc_pos_preload); + if (rc) + return rc; + 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, 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); return rc; } static inline int -osd_scrub_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it) +osd_scrub_wakeup(struct lustre_scrub *scrub, struct osd_otable_it *it) { spin_lock(&scrub->os_lock); if (osd_scrub_has_window(scrub, &it->ooi_cache) || @@ -1106,24 +878,21 @@ static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev, struct osd_iit_param *param, 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; + struct lustre_scrub *scrub = &dev->od_scrub.os_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; switch (rc) { - case SCRUB_NEXT_CONTINUE: - 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; + case SCRUB_NEXT_CONTINUE: + case SCRUB_NEXT_WAIT: + goto wait; } rc = osd_scrub_check_update(info, dev, oic, rc); @@ -1132,9 +901,9 @@ static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev, return rc; } - rc = osd_scrub_checkpoint(scrub); - if (rc != 0) { - CDEBUG(D_LFSCK, "%.16s: fail to checkpoint, pos = %u: " + rc = scrub_checkpoint(info->oti_env, scrub); + if (rc) { + CDEBUG(D_LFSCK, "%s: fail to checkpoint, pos = %llu: " "rc = %d\n", osd_scrub2name(scrub), scrub->os_pos_current, rc); /* Continue, as long as the scrub itself can go ahead. */ @@ -1145,9 +914,6 @@ static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev, return 0; } -next: - scrub->os_pos_current = param->gbase + ++(param->offset); - wait: if (it != NULL && it->ooi_waiting && ooc != NULL && ooc->ooc_pos_preload < scrub->os_pos_current) { @@ -1157,19 +923,19 @@ wait: spin_unlock(&scrub->os_lock); } - if (scrub->os_full_speed || rc == SCRUB_NEXT_CONTINUE) + if (rc == SCRUB_NEXT_CONTINUE) return 0; - if (ooc != NULL && osd_scrub_has_window(scrub, ooc)) { + if (scrub->os_full_speed || !ooc || osd_scrub_has_window(scrub, ooc)) { *noslot = false; return 0; } if (it != NULL) - l_wait_event(thread->t_ctl_waitq, osd_scrub_wakeup(scrub, it), - &lwi); + wait_event_idle(thread->t_ctl_waitq, + osd_scrub_wakeup(scrub, it)); - if (ooc != NULL && osd_scrub_has_window(scrub, ooc)) + if (!ooc || osd_scrub_has_window(scrub, ooc)) *noslot = false; else *noslot = true; @@ -1193,10 +959,10 @@ 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) +static void osd_scrub_join(const struct lu_env *env, struct osd_device *dev, + __u32 flags, bool inconsistent) { - struct osd_scrub *scrub = &dev->od_scrub; + struct lustre_scrub *scrub = &dev->od_scrub.os_scrub; struct ptlrpc_thread *thread = &scrub->os_thread; struct scrub_file *sf = &scrub->os_file; int rc; @@ -1217,22 +983,21 @@ static void osd_scrub_join(struct osd_device *dev, __u32 flags, 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); + scrub_file_reset(scrub, dev->od_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; + if (flags & SS_AUTO_FULL) { + sf->sf_flags |= SF_AUTO; + scrub->os_full_speed = 1; + } + scrub->os_new_checked = 0; if (sf->sf_pos_last_checkpoint != 0) sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1; @@ -1240,13 +1005,10 @@ static void osd_scrub_join(struct osd_device *dev, __u32 flags, 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_latest_start = ktime_get_real_seconds(); sf->sf_time_last_checkpoint = sf->sf_time_latest_start; sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1; - rc = 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); + rc = scrub_file_store(env, scrub); spin_lock(&scrub->os_lock); scrub->os_waiting = 0; @@ -1258,34 +1020,39 @@ static void osd_scrub_join(struct osd_device *dev, __u32 flags, wake_up_all(&thread->t_ctl_waitq); up_write(&scrub->os_rwsem); + CDEBUG(D_LFSCK, "%s: joined in the OI scrub with flag %u: rc = %d\n", + osd_scrub2name(scrub), flags, rc); + EXIT; } static int osd_inode_iteration(struct osd_thread_info *info, struct osd_device *dev, __u32 max, bool preload) { - struct osd_scrub *scrub = &dev->od_scrub; + struct lustre_scrub *scrub = &dev->od_scrub.os_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 = { NULL }; - struct l_wait_info lwi = { 0 }; - __u32 limit; - int rc; - bool noslot = true; + struct scrub_file *sf = &scrub->os_file; + osd_iit_next_policy next; + osd_iit_exec_policy exec; + __u64 *pos; + __u64 *count; + struct osd_iit_param *param; + __u32 limit; + int rc; + bool noslot = true; ENTRY; - param.sb = osd_sb(dev); if (preload) goto full; + param = &dev->od_scrub.os_iit_param; + memset(param, 0, sizeof(*param)); + param->sb = osd_sb(dev); + while (scrub->os_partial_scan && !scrub->os_in_join) { struct osd_idmap_cache *oic = NULL; - rc = osd_scrub_next(info, dev, ¶m, &oic, noslot); + rc = osd_scrub_next(info, dev, param, &oic, noslot); switch (rc) { case SCRUB_NEXT_EXIT: RETURN(0); @@ -1303,12 +1070,12 @@ static int osd_inode_iteration(struct osd_thread_info *info, if (dev->od_full_scrub_ratio == OFSR_DIRECTLY || scrub->os_full_scrub) { - osd_scrub_join(dev, SS_AUTO_FULL | SS_RESET, - true); + osd_scrub_join(info->oti_env, dev, + SS_AUTO_FULL | SS_RESET, true); goto full; } - rc = param.sb->s_op->statfs(param.sb->s_root, ksfs); + rc = param->sb->s_op->statfs(param->sb->s_root, ksfs); if (rc == 0) { __u64 used = ksfs->f_files - ksfs->f_ffree; @@ -1317,7 +1084,7 @@ static int osd_inode_iteration(struct osd_thread_info *info, * mappings during the partial scan, * then scan the device completely. */ if (used < dev->od_full_scrub_ratio) { - osd_scrub_join(dev, + osd_scrub_join(info->oti_env, dev, SS_AUTO_FULL | SS_RESET, true); goto full; } @@ -1332,12 +1099,12 @@ wait: 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); + wait_event_idle( + thread->t_ctl_waitq, + !thread_is_running(thread) || + !scrub->os_partial_scan || + scrub->os_in_join || + !list_empty(&scrub->os_inconsistent_items)); sf->sf_flags = saved_flags; sf->sf_status = SS_SCANNING; @@ -1352,16 +1119,16 @@ wait: default: LASSERTF(rc == 0, "rc = %d\n", rc); - osd_scrub_exec(info, dev, ¶m, oic, &noslot, rc); + osd_scrub_exec(info, dev, param, oic, &noslot, rc); break; } } full: if (!preload) { - l_wait_event(thread->t_ctl_waitq, - !thread_is_running(thread) || !scrub->os_in_join, - &lwi); + wait_event_idle(thread->t_ctl_waitq, + !thread_is_running(thread) || + !scrub->os_in_join); if (unlikely(!thread_is_running(thread))) RETURN(0); @@ -1373,6 +1140,12 @@ full: exec = osd_scrub_exec; pos = &scrub->os_pos_current; count = &scrub->os_new_checked; + param->start = *pos; + 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); } else { struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache; @@ -1380,80 +1153,92 @@ full: exec = osd_preload_exec; pos = &ooc->ooc_pos_preload; count = &ooc->ooc_cached_items; + param = &dev->od_otable_it->ooi_iit_param; } - limit = le32_to_cpu(LDISKFS_SB(param.sb)->s_es->s_inodes_count); + rc = 0; + limit = le32_to_cpu(LDISKFS_SB(osd_sb(dev))->s_es->s_inodes_count); while (*pos <= limit && *count < max) { - struct osd_idmap_cache *oic = NULL; struct ldiskfs_group_desc *desc; + bool next_group = false; - param.bg = (*pos - 1) / LDISKFS_INODES_PER_GROUP(param.sb); - desc = ldiskfs_get_group_desc(param.sb, param.bg, NULL); - if (desc == NULL) + desc = ldiskfs_get_group_desc(param->sb, param->bg, NULL); + if (!desc) 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; + next_group = true; + goto next_group; } - 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, " + param->bitmap = ldiskfs_read_inode_bitmap(param->sb, param->bg); + if (!param->bitmap) { + CERROR("%s: fail to read bitmap for %u, " "scrub will stop, urgent mode\n", - osd_scrub2name(scrub), (__u32)param.bg); + osd_scrub2name(scrub), (__u32)param->bg); RETURN(-EIO); } - while (param.offset < LDISKFS_INODES_PER_GROUP(param.sb) && - *count < max) { - if (param.offset + - ldiskfs_itable_unused_count(param.sb, desc) > - LDISKFS_INODES_PER_GROUP(param.sb)) + do { + struct osd_idmap_cache *oic = NULL; + + if (param->offset + + ldiskfs_itable_unused_count(param->sb, desc) >= + LDISKFS_INODES_PER_GROUP(param->sb)) { + next_group = true; goto next_group; + } - rc = next(info, dev, ¶m, &oic, noslot); + rc = next(info, dev, param, &oic, noslot); switch (rc) { case SCRUB_NEXT_BREAK: + next_group = true; goto next_group; case SCRUB_NEXT_EXIT: - brelse(param.bitmap); + brelse(param->bitmap); RETURN(0); case SCRUB_NEXT_CRASH: - brelse(param.bitmap); + brelse(param->bitmap); RETURN(SCRUB_IT_CRASH); case SCRUB_NEXT_FATAL: - brelse(param.bitmap); + brelse(param->bitmap); RETURN(-EINVAL); } - rc = exec(info, dev, ¶m, oic, &noslot, rc); - if (rc != 0) { - brelse(param.bitmap); - RETURN(rc); - } - } + rc = exec(info, dev, param, oic, &noslot, rc); + } while (!rc && *pos <= limit && *count < max); next_group: - brelse(param.bitmap); + if (param->bitmap) { + brelse(param->bitmap); + param->bitmap = NULL; + } + + if (rc < 0) + GOTO(out, rc); + + if (next_group) { + param->bg++; + param->offset = 0; + param->gbase = 1 + + param->bg * LDISKFS_INODES_PER_GROUP(param->sb); + *pos = param->gbase; + param->start = *pos; + } } if (*pos > limit) RETURN(SCRUB_IT_ALL); - RETURN(0); + +out: + RETURN(rc); } 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_device *dev = it->ooi_dev; + struct lustre_scrub *scrub = &dev->od_scrub.os_scrub; struct osd_otable_cache *ooc = &it->ooi_cache; int rc; ENTRY; @@ -1475,53 +1260,57 @@ static int osd_otable_it_preload(const struct lu_env *env, 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 lu_env env; + struct osd_device *dev = (struct osd_device *)args; + struct lustre_scrub *scrub = &dev->od_scrub.os_scrub; struct ptlrpc_thread *thread = &scrub->os_thread; - int rc; + int rc; ENTRY; - rc = lu_env_init(&env, LCT_LOCAL); + rc = lu_env_init(&env, LCT_LOCAL | LCT_DT_THREAD); if (rc != 0) { - CDEBUG(D_LFSCK, "%.16s: OI scrub fail to init env: rc = %d\n", + CDEBUG(D_LFSCK, "%s: OI scrub fail to init env: rc = %d\n", osd_scrub2name(scrub), rc); GOTO(noenv, rc); } - rc = osd_scrub_prep(dev); + rc = osd_scrub_prep(&env, dev); if (rc != 0) { - CDEBUG(D_LFSCK, "%.16s: OI scrub fail to scrub prep: rc = %d\n", + CDEBUG(D_LFSCK, "%s: 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); + wait_event_idle(thread->t_ctl_waitq, + it->ooi_user_ready || + !thread_is_running(thread)); 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", + CDEBUG(D_LFSCK, "%s: OI scrub start, flags = 0x%x, pos = %llu\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)) + if (unlikely(rc == SCRUB_IT_CRASH)) { + spin_lock(&scrub->os_lock); + thread_set_flags(&scrub->os_thread, SVC_STOPPING); + spin_unlock(&scrub->os_lock); GOTO(out, rc = -EINVAL); + } + GOTO(post, rc); post: - rc = osd_scrub_post(scrub, rc); - CDEBUG(D_LFSCK, "%.16s: OI scrub: stop, pos = %u: rc = %d\n", + rc = osd_scrub_post(&env, dev, rc); + CDEBUG(D_LFSCK, "%s: OI scrub: stop, pos = %llu: rc = %d\n", osd_scrub2name(scrub), scrub->os_pos_current, rc); out: @@ -1529,7 +1318,7 @@ out: struct osd_inconsistent_item *oii; oii = list_entry(scrub->os_inconsistent_items.next, - struct osd_inconsistent_item, oii_list); + struct osd_inconsistent_item, oii_list); list_del_init(&oii->oii_list); OBD_FREE_PTR(oii); } @@ -1583,14 +1372,6 @@ 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, - OLF_IDX_IN_FID = 0x0010, -}; - struct osd_lf_map { char *olm_name; struct lu_fid olm_fid; @@ -1603,145 +1384,336 @@ struct osd_lf_map { /* 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, - sizeof(CATLIST) - 1, NULL, NULL }, + { + .olm_name = CATLIST, + .olm_fid = { + .f_seq = FID_SEQ_LOCAL_FILE, + .f_oid = LLOG_CATALOGS_OID, + }, + .olm_flags = OLF_SHOW_NAME, + .olm_namelen = sizeof(CATLIST) - 1, + }, /* CONFIGS */ - { MOUNT_CONFIGS_DIR, { FID_SEQ_LOCAL_FILE, MGS_CONFIGS_OID, 0 }, - OLF_SCAN_SUBITEMS, sizeof(MOUNT_CONFIGS_DIR) - 1, - osd_ios_general_scan, osd_ios_varfid_fill }, + { + .olm_name = MOUNT_CONFIGS_DIR, + .olm_fid = { + .f_seq = FID_SEQ_LOCAL_FILE, + .f_oid = MGS_CONFIGS_OID, + }, + .olm_flags = OLF_SCAN_SUBITEMS, + .olm_namelen = sizeof(MOUNT_CONFIGS_DIR) - 1, + .olm_scandir = osd_ios_general_scan, + .olm_filldir = osd_ios_varfid_fill, + }, /* NIDTBL_VERSIONS */ - { MGS_NIDTBL_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, - sizeof(MGS_NIDTBL_DIR) - 1, osd_ios_general_scan, - osd_ios_varfid_fill }, + { + .olm_name = MGS_NIDTBL_DIR, + .olm_flags = OLF_SCAN_SUBITEMS, + .olm_namelen = sizeof(MGS_NIDTBL_DIR) - 1, + .olm_scandir = osd_ios_general_scan, + .olm_filldir = osd_ios_varfid_fill, + }, /* PENDING */ - { "PENDING", { 0, 0, 0 }, 0, sizeof("PENDING") - 1, NULL, NULL }, + { + .olm_name = MDT_ORPHAN_DIR, + .olm_namelen = sizeof(MDT_ORPHAN_DIR) - 1, + }, /* ROOT */ - { "ROOT", { FID_SEQ_ROOT, FID_OID_ROOT, 0 }, - OLF_SCAN_SUBITEMS | OLF_HIDE_FID, sizeof("ROOT") - 1, - osd_ios_ROOT_scan, NULL }, + { + .olm_name = "ROOT", + .olm_fid = { + .f_seq = FID_SEQ_ROOT, + .f_oid = FID_OID_ROOT, + }, + .olm_flags = OLF_SCAN_SUBITEMS | OLF_HIDE_FID, + .olm_namelen = sizeof("ROOT") - 1, + .olm_scandir = osd_ios_ROOT_scan, + }, /* changelog_catalog */ - { CHANGELOG_CATALOG, { 0, 0, 0 }, 0, sizeof(CHANGELOG_CATALOG) - 1, - NULL, NULL }, + { + .olm_name = CHANGELOG_CATALOG, + .olm_namelen = sizeof(CHANGELOG_CATALOG) - 1, + }, /* changelog_users */ - { CHANGELOG_USERS, { 0, 0, 0 }, 0, sizeof(CHANGELOG_USERS) - 1, - NULL, NULL }, + { + .olm_name = CHANGELOG_USERS, + .olm_namelen = sizeof(CHANGELOG_USERS) - 1, + }, /* fld */ - { "fld", { FID_SEQ_LOCAL_FILE, FLD_INDEX_OID, 0 }, OLF_SHOW_NAME, - sizeof("fld") - 1, NULL, NULL }, + { + .olm_name = "fld", + .olm_fid = { + .f_seq = FID_SEQ_LOCAL_FILE, + .f_oid = FLD_INDEX_OID, + }, + .olm_flags = OLF_SHOW_NAME, + .olm_namelen = sizeof("fld") - 1, + }, /* last_rcvd */ - { LAST_RCVD, { FID_SEQ_LOCAL_FILE, LAST_RECV_OID, 0 }, OLF_SHOW_NAME, - sizeof(LAST_RCVD) - 1, NULL, NULL }, + { + .olm_name = LAST_RCVD, + .olm_fid = { + .f_seq = FID_SEQ_LOCAL_FILE, + .f_oid = LAST_RECV_OID, + }, + .olm_flags = OLF_SHOW_NAME, + .olm_namelen = sizeof(LAST_RCVD) - 1, + }, /* reply_data */ - { REPLY_DATA, { FID_SEQ_LOCAL_FILE, REPLY_DATA_OID, 0 }, OLF_SHOW_NAME, - sizeof(REPLY_DATA) - 1, NULL, NULL }, + { + .olm_name = REPLY_DATA, + .olm_fid = { + .f_seq = FID_SEQ_LOCAL_FILE, + .f_oid = REPLY_DATA_OID, + }, + .olm_flags = OLF_SHOW_NAME, + .olm_namelen = sizeof(REPLY_DATA) - 1, + }, /* lov_objid */ - { LOV_OBJID, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OID, 0 }, OLF_SHOW_NAME, - sizeof(LOV_OBJID) - 1, NULL, NULL }, + { + .olm_name = LOV_OBJID, + .olm_fid = { + .f_seq = FID_SEQ_LOCAL_FILE, + .f_oid = MDD_LOV_OBJ_OID, + }, + .olm_flags = OLF_SHOW_NAME, + .olm_namelen = sizeof(LOV_OBJID) - 1, + }, /* lov_objseq */ - { LOV_OBJSEQ, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OSEQ, 0 }, - OLF_SHOW_NAME, sizeof(LOV_OBJSEQ) - 1, NULL, NULL }, + { + .olm_name = LOV_OBJSEQ, + .olm_fid = { + .f_seq = FID_SEQ_LOCAL_FILE, + .f_oid = MDD_LOV_OBJ_OSEQ, + }, + .olm_flags = OLF_SHOW_NAME, + .olm_namelen = sizeof(LOV_OBJSEQ) - 1, + }, /* quota_master */ - { QMT_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(QMT_DIR) - 1, - osd_ios_general_scan, osd_ios_varfid_fill }, + { + .olm_name = QMT_DIR, + .olm_flags = OLF_SCAN_SUBITEMS, + .olm_namelen = sizeof(QMT_DIR) - 1, + .olm_scandir = osd_ios_general_scan, + .olm_filldir = osd_ios_varfid_fill, + }, /* quota_slave */ - { QSD_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(QSD_DIR) - 1, - osd_ios_general_scan, osd_ios_varfid_fill }, + { + .olm_name = QSD_DIR, + .olm_flags = OLF_SCAN_SUBITEMS, + .olm_namelen = sizeof(QSD_DIR) - 1, + .olm_scandir = osd_ios_general_scan, + .olm_filldir = osd_ios_varfid_fill, + }, /* seq_ctl */ - { "seq_ctl", { FID_SEQ_LOCAL_FILE, FID_SEQ_CTL_OID, 0 }, - OLF_SHOW_NAME, sizeof("seq_ctl") - 1, NULL, NULL }, + { + .olm_name = "seq_ctl", + .olm_fid = { + .f_seq = FID_SEQ_LOCAL_FILE, + .f_oid = FID_SEQ_CTL_OID, + }, + .olm_flags = OLF_SHOW_NAME, + .olm_namelen = sizeof("seq_ctl") - 1, + }, /* seq_srv */ - { "seq_srv", { FID_SEQ_LOCAL_FILE, FID_SEQ_SRV_OID, 0 }, - OLF_SHOW_NAME, sizeof("seq_srv") - 1, NULL, NULL }, + { + .olm_name = "seq_srv", + .olm_fid = { + .f_seq = FID_SEQ_LOCAL_FILE, + .f_oid = FID_SEQ_SRV_OID, + }, + .olm_flags = OLF_SHOW_NAME, + .olm_namelen = sizeof("seq_srv") - 1, + }, /* health_check */ - { HEALTH_CHECK, { FID_SEQ_LOCAL_FILE, OFD_HEALTH_CHECK_OID, 0 }, - OLF_SHOW_NAME, sizeof(HEALTH_CHECK) - 1, NULL, NULL }, + { + .olm_name = HEALTH_CHECK, + .olm_fid = { + .f_seq = FID_SEQ_LOCAL_FILE, + .f_oid = OFD_HEALTH_CHECK_OID, + }, + .olm_flags = OLF_SHOW_NAME, + .olm_namelen = sizeof(HEALTH_CHECK) - 1, + }, /* LFSCK */ - { LFSCK_DIR, { 0, 0, 0 }, 0, sizeof(LFSCK_DIR) - 1, - osd_ios_general_scan, osd_ios_varfid_fill }, + { + .olm_name = LFSCK_DIR, + .olm_flags = OLF_SCAN_SUBITEMS, + .olm_namelen = sizeof(LFSCK_DIR) - 1, + .olm_scandir = osd_ios_general_scan, + .olm_filldir = osd_ios_varfid_fill, + }, /* lfsck_bookmark */ - { LFSCK_BOOKMARK, { 0, 0, 0 }, 0, sizeof(LFSCK_BOOKMARK) - 1, - NULL, NULL }, + { + .olm_name = LFSCK_BOOKMARK, + .olm_namelen = sizeof(LFSCK_BOOKMARK) - 1, + }, /* lfsck_layout */ - { LFSCK_LAYOUT, { 0, 0, 0 }, 0, sizeof(LFSCK_LAYOUT) - 1, - NULL, NULL }, + { + .olm_name = LFSCK_LAYOUT, + .olm_namelen = sizeof(LFSCK_LAYOUT) - 1, + }, /* lfsck_namespace */ - { LFSCK_NAMESPACE, { 0, 0, 0 }, 0, sizeof(LFSCK_NAMESPACE) - 1, - NULL, NULL }, + { + .olm_name = LFSCK_NAMESPACE, + .olm_namelen = sizeof(LFSCK_NAMESPACE) - 1, + }, /* OBJECTS, upgrade from old device */ - { OBJECTS, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(OBJECTS) - 1, - osd_ios_OBJECTS_scan, NULL }, + { + .olm_name = OBJECTS, + .olm_flags = OLF_SCAN_SUBITEMS, + .olm_namelen = sizeof(OBJECTS) - 1, + .olm_scandir = osd_ios_OBJECTS_scan, + }, /* lquota_v2.user, upgrade from old device */ - { "lquota_v2.user", { 0, 0, 0 }, 0, sizeof("lquota_v2.user") - 1, - NULL, NULL }, + { + .olm_name = "lquota_v2.user", + .olm_namelen = sizeof("lquota_v2.user") - 1, + }, /* lquota_v2.group, upgrade from old device */ - { "lquota_v2.group", { 0, 0, 0 }, 0, sizeof("lquota_v2.group") - 1, - NULL, NULL }, + { + .olm_name = "lquota_v2.group", + .olm_namelen = sizeof("lquota_v2.group") - 1, + }, /* LAST_GROUP, upgrade from old device */ - { "LAST_GROUP", { FID_SEQ_LOCAL_FILE, OFD_LAST_GROUP_OID, 0 }, - OLF_SHOW_NAME, sizeof("LAST_GROUP") - 1, NULL, NULL }, + { + .olm_name = "LAST_GROUP", + .olm_fid = { + .f_seq = FID_SEQ_LOCAL_FILE, + .f_oid = OFD_LAST_GROUP_OID, + }, + .olm_flags = OLF_SHOW_NAME, + .olm_namelen = sizeof("LAST_GROUP") - 1, + }, /* committed batchid for cross-MDT operation */ - { "BATCHID", { FID_SEQ_LOCAL_FILE, BATCHID_COMMITTED_OID, 0 }, - OLF_SHOW_NAME, sizeof("BATCHID") - 1, NULL, NULL }, + { + .olm_name = "BATCHID", + .olm_fid = { + .f_seq = FID_SEQ_LOCAL_FILE, + .f_oid = BATCHID_COMMITTED_OID, + }, + .olm_flags = OLF_SHOW_NAME, + .olm_namelen = sizeof("BATCHID") - 1, + }, /* OSP update logs update_log{_dir} use f_seq = FID_SEQ_UPDATE_LOG{_DIR} * and f_oid = index for their log files. See lu_update_log{_dir}_fid() * for more details. */ /* update_log */ - { "update_log", { FID_SEQ_UPDATE_LOG, 0, 0 }, - OLF_SHOW_NAME | OLF_IDX_IN_FID, sizeof("update_log") - 1, - NULL, NULL }, + { + .olm_name = "update_log", + .olm_fid = { + .f_seq = FID_SEQ_UPDATE_LOG, + }, + .olm_flags = OLF_SHOW_NAME | OLF_IDX_IN_FID, + .olm_namelen = sizeof("update_log") - 1, + }, /* update_log_dir */ - { "update_log_dir", { FID_SEQ_UPDATE_LOG_DIR, 0, 0 }, - OLF_SHOW_NAME | OLF_SCAN_SUBITEMS | OLF_IDX_IN_FID, - sizeof("update_log_dir") - 1, - osd_ios_general_scan, osd_ios_uld_fill }, + { + .olm_name = "update_log_dir", + .olm_fid = { + .f_seq = FID_SEQ_UPDATE_LOG_DIR, + }, + .olm_flags = OLF_SHOW_NAME | OLF_SCAN_SUBITEMS | + OLF_IDX_IN_FID, + .olm_namelen = sizeof("update_log_dir") - 1, + .olm_scandir = osd_ios_general_scan, + .olm_filldir = osd_ios_uld_fill, + }, /* lost+found */ - { "lost+found", { FID_SEQ_LOCAL_FILE, OSD_LPF_OID, 0 }, - OLF_SCAN_SUBITEMS, sizeof("lost+found") - 1, - osd_ios_general_scan, osd_ios_lf_fill }, - - { NULL, { 0, 0, 0 }, 0, 0, NULL, NULL } + { + .olm_name = "lost+found", + .olm_fid = { + .f_seq = FID_SEQ_LOCAL_FILE, + .f_oid = OSD_LPF_OID, + }, + .olm_flags = OLF_SCAN_SUBITEMS, + .olm_namelen = sizeof("lost+found") - 1, + .olm_scandir = osd_ios_general_scan, + .olm_filldir = osd_ios_lf_fill, + }, + + /* hsm_actions */ + { + .olm_name = HSM_ACTIONS, + }, + + /* nodemap */ + { + .olm_name = LUSTRE_NODEMAP_NAME, + }, + + /* index_backup */ + { + .olm_name = INDEX_BACKUP_DIR, + .olm_fid = { + .f_seq = FID_SEQ_LOCAL_FILE, + .f_oid = INDEX_BACKUP_OID, + }, + .olm_flags = OLF_SCAN_SUBITEMS | OLF_NOT_BACKUP, + .olm_namelen = sizeof(INDEX_BACKUP_DIR) - 1, + .olm_scandir = osd_ios_general_scan, + .olm_filldir = osd_ios_varfid_fill, + }, + + { + .olm_name = NULL + } }; /* Add the new introduced files under .lustre/ in the list in the future. */ static const struct osd_lf_map osd_dl_maps[] = { /* .lustre/fid */ - { "fid", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_OBF, 0 }, 0, - sizeof("fid") - 1, NULL, NULL }, + { + .olm_name = "fid", + .olm_fid = { + .f_seq = FID_SEQ_DOT_LUSTRE, + .f_oid = FID_OID_DOT_LUSTRE_OBF, + }, + .olm_namelen = sizeof("fid") - 1, + }, /* .lustre/lost+found */ - { "lost+found", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_LPF, 0 }, 0, - sizeof("lost+found") - 1, NULL, NULL }, - - { NULL, { 0, 0, 0 }, 0, 0, NULL, NULL } + { + .olm_name = "lost+found", + .olm_fid = { + .f_seq = FID_SEQ_DOT_LUSTRE, + .f_oid = FID_OID_DOT_LUSTRE_LPF, + }, + .olm_namelen = sizeof("lost+found") - 1, + }, + + { + .olm_name = NULL + } }; struct osd_ios_item { @@ -1752,41 +1724,14 @@ struct osd_ios_item { }; struct osd_ios_filldir_buf { -#ifdef HAVE_DIR_CONTEXT /* please keep it as first member */ struct dir_context ctx; -#endif struct osd_thread_info *oifb_info; struct osd_device *oifb_dev; struct dentry *oifb_dentry; + int oifb_items; }; -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)) { - int rc = PTR_ERR(dentry); - - if (rc != -ENOENT) - CERROR("Fail to find %.*s in %.*s (%lu/%u): rc = %d\n", - namelen, name, parent->d_name.len, - parent->d_name.name, parent->d_inode->i_ino, - parent->d_inode->i_generation, rc); - - return dentry; - } - - if (dentry->d_inode == NULL) { - dput(dentry); - return ERR_PTR(-ENOENT); - } - - return dentry; -} - static int osd_ios_new_item(struct osd_device *dev, struct dentry *dentry, scandir_t scandir, filldir_t filldir) @@ -1807,6 +1752,119 @@ osd_ios_new_item(struct osd_device *dev, struct dentry *dentry, RETURN(0); } +static bool osd_index_need_recreate(const struct lu_env *env, + struct osd_device *dev, struct inode *inode) +{ + struct osd_directory *iam = &osd_oti_get(env)->oti_iam; + struct iam_container *bag = &iam->od_container; + int rc; + ENTRY; + + rc = iam_container_init(bag, &iam->od_descr, inode); + if (rc) + RETURN(true); + + rc = iam_container_setup(bag); + iam_container_fini(bag); + if (rc) + RETURN(true); + + RETURN(false); +} + +static void osd_ios_index_register(const struct lu_env *env, + struct osd_device *osd, + const struct lu_fid *fid, + struct inode *inode) +{ + struct osd_directory *iam = &osd_oti_get(env)->oti_iam; + struct iam_container *bag = &iam->od_container; + struct super_block *sb = osd_sb(osd); + struct iam_descr *descr; + __u32 keysize = 0; + __u32 recsize = 0; + int rc; + ENTRY; + + /* Index must be a regular file. */ + if (!S_ISREG(inode->i_mode)) + RETURN_EXIT; + + /* Index's size must be block aligned. */ + if (inode->i_size < sb->s_blocksize || + (inode->i_size & (sb->s_blocksize - 1)) != 0) + RETURN_EXIT; + + iam_container_init(bag, &iam->od_descr, inode); + rc = iam_container_setup(bag); + if (rc) + GOTO(fini, rc = 1); + + descr = bag->ic_descr; + /* May be regular file with IAM_LFIX_ROOT_MAGIC matched + * coincidentally, or corrupted index object, skip it. */ + if (descr->id_ptr_size != 4) + GOTO(fini, rc = 1); + + keysize = descr->id_key_size; + recsize = descr->id_rec_size; + rc = osd_index_register(osd, fid, keysize, recsize); + + GOTO(fini, rc); + +fini: + iam_container_fini(bag); + if (!rc) + CDEBUG(D_LFSCK, "%s: index object "DFID" (%u/%u) registered\n", + osd_name(osd), PFID(fid), keysize, recsize); +} + +static void osd_index_restore(const struct lu_env *env, struct osd_device *dev, + struct lustre_index_restore_unit *liru, + void *buf, int bufsize) +{ + struct osd_thread_info *info = osd_oti_get(env); + struct osd_inode_id *id = &info->oti_id; + struct lu_fid *tgt_fid = &liru->liru_cfid; + struct inode *bak_inode = NULL; + struct ldiskfs_dir_entry_2 *de = NULL; + struct buffer_head *bh = NULL; + struct dentry *dentry; + char *name = buf; + struct lu_fid bak_fid; + int rc; + ENTRY; + + lustre_fid2lbx(name, tgt_fid, bufsize); + dentry = osd_child_dentry_by_inode(env, dev->od_index_backup_inode, + name, strlen(name)); + bh = osd_ldiskfs_find_entry(dev->od_index_backup_inode, + &dentry->d_name, &de, NULL, NULL); + if (IS_ERR(bh)) + GOTO(log, rc = PTR_ERR(bh)); + + osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN); + brelse(bh); + bak_inode = osd_iget_fid(info, dev, id, &bak_fid); + if (IS_ERR(bak_inode)) + GOTO(log, rc = PTR_ERR(bak_inode)); + + iput(bak_inode); + /* The OI mapping for index may be invalid, since it will be + * re-created, not update the OI mapping, just cache it in RAM. */ + osd_id_gen(id, liru->liru_clid, OSD_OII_NOGEN); + osd_add_oi_cache(info, dev, id, tgt_fid); + rc = lustre_index_restore(env, &dev->od_dt_dev, &liru->liru_pfid, + tgt_fid, &bak_fid, liru->liru_name, + &dev->od_index_backup_list, &dev->od_lock, + buf, bufsize); + GOTO(log, rc); + +log: + CDEBUG(D_WARNING, "%s: restore index '%s' with "DFID": rc = %d\n", + osd_name(dev), liru->liru_name, PFID(tgt_fid), rc); +} + /** * osd_ios_scan_one() - check/fix LMA FID and OI entry for one inode * @@ -1816,18 +1874,27 @@ osd_ios_new_item(struct osd_device *dev, struct dentry *dentry, */ 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 inode *parent, struct inode *inode, + const struct lu_fid *fid, const char *name, + int namelen, int flags) { - struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs; + struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma; struct osd_inode_id *id = &info->oti_id; struct osd_inode_id *id2 = &info->oti_id2; - struct osd_scrub *scrub = &dev->od_scrub; + struct lustre_scrub *scrub = &dev->od_scrub.os_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 (!inode) { + CDEBUG(D_INODE, "%s: child '%.*s' lacks inode: rc = -2\n", + osd_name(dev), namelen, name); + RETURN(-ENOENT); + } + + rc = osd_get_lma(info, inode, &info->oti_obj_dentry, + &info->oti_ost_attrs); if (rc != 0 && rc != -ENODATA) { CDEBUG(D_LFSCK, "%s: fail to get lma for init OI scrub: " "rc = %d\n", osd_name(dev), rc); @@ -1859,9 +1926,32 @@ osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev, RETURN(0); tfid = lma->lma_self_fid; + if (lma->lma_compat & LMAC_IDX_BACKUP && + osd_index_need_recreate(info->oti_env, dev, inode)) { + struct lu_fid *pfid = &info->oti_fid3; + + if (parent == osd_sb(dev)->s_root->d_inode) { + lu_local_obj_fid(pfid, OSD_FS_ROOT_OID); + } else { + rc = osd_scrub_get_fid(info, dev, parent, pfid, + false); + if (rc) + RETURN(rc); + } + + rc = lustre_liru_new(&dev->od_index_restore_list, pfid, + &tfid, inode->i_ino, name, namelen); + + RETURN(rc); + } + + if (!(flags & OLF_NOT_BACKUP)) + osd_ios_index_register(info->oti_env, dev, &tfid, + inode); } - rc = osd_oi_lookup(info, dev, &tfid, id2, 0); + /* Since this called from iterate_dir() the inode lock will be taken */ + rc = osd_oi_lookup(info, dev, &tfid, id2, OI_LOCKED); if (rc != 0) { if (rc != -ENOENT) RETURN(rc); @@ -1878,10 +1968,8 @@ osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev, 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); + scrub_file_reset(scrub, dev->od_uuid, SF_INCONSISTENT); + rc = scrub_file_store(info->oti_env, scrub); if (rc != 0) RETURN(rc); } @@ -1896,14 +1984,14 @@ osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev, /** * 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. + * or filter_fid_18_23), move them back to its proper /O//d. */ #ifdef HAVE_FILLDIR_USE_CTX -static int osd_ios_lf_fill(struct dir_context *buf, const char *name, - int namelen, +static int osd_ios_lf_fill(struct dir_context *buf, #else -static int osd_ios_lf_fill(void *buf, const char *name, int namelen, +static int osd_ios_lf_fill(void *buf, #endif + const char *name, int namelen, loff_t offset, __u64 ino, unsigned d_type) { struct osd_ios_filldir_buf *fill_buf = @@ -1919,16 +2007,24 @@ static int osd_ios_lf_fill(void *buf, const char *name, int namelen, int rc; ENTRY; + fill_buf->oifb_items++; + /* skip any '.' started names */ if (name[0] == '.') RETURN(0); scrub->os_lf_scanned++; - child = osd_ios_lookup_one_len(name, parent, namelen); + child = osd_lookup_one_len(dev, name, parent, namelen); if (IS_ERR(child)) { + rc = PTR_ERR(child); CDEBUG(D_LFSCK, "%s: cannot lookup child '%.*s': rc = %d\n", - osd_name(dev), namelen, name, (int)PTR_ERR(child)); - RETURN(0); + osd_name(dev), namelen, name, rc); + RETURN(rc); + } else if (!child->d_inode) { + dput(child); + CDEBUG(D_INODE, "%s: child '%.*s' lacks inode\n", + osd_name(dev), namelen, name); + RETURN(-ENOENT); } inode = child->d_inode; @@ -1973,11 +2069,11 @@ put: } #ifdef HAVE_FILLDIR_USE_CTX -static int osd_ios_varfid_fill(struct dir_context *buf, const char *name, - int namelen, +static int osd_ios_varfid_fill(struct dir_context *buf, #else -static int osd_ios_varfid_fill(void *buf, const char *name, int namelen, +static int osd_ios_varfid_fill(void *buf, #endif + const char *name, int namelen, loff_t offset, __u64 ino, unsigned d_type) { struct osd_ios_filldir_buf *fill_buf = @@ -1987,16 +2083,19 @@ static int osd_ios_varfid_fill(void *buf, const char *name, int namelen, int rc; ENTRY; + fill_buf->oifb_items++; + /* skip any '.' started names */ if (name[0] == '.') RETURN(0); - child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen); + child = osd_lookup_one_len(dev, 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); + rc = osd_ios_scan_one(fill_buf->oifb_info, dev, + fill_buf->oifb_dentry->d_inode, child->d_inode, + NULL, name, namelen, 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); @@ -2006,11 +2105,11 @@ static int osd_ios_varfid_fill(void *buf, const char *name, int namelen, } #ifdef HAVE_FILLDIR_USE_CTX -static int osd_ios_dl_fill(struct dir_context *buf, const char *name, - int namelen, +static int osd_ios_dl_fill(struct dir_context *buf, #else -static int osd_ios_dl_fill(void *buf, const char *name, int namelen, +static int osd_ios_dl_fill(void *buf, #endif + const char *name, int namelen, loff_t offset, __u64 ino, unsigned d_type) { struct osd_ios_filldir_buf *fill_buf = @@ -2021,6 +2120,8 @@ static int osd_ios_dl_fill(void *buf, const char *name, int namelen, int rc = 0; ENTRY; + fill_buf->oifb_items++; + /* skip any '.' started names */ if (name[0] == '.') RETURN(0); @@ -2036,37 +2137,41 @@ static int osd_ios_dl_fill(void *buf, const char *name, int namelen, if (map->olm_name == NULL) RETURN(0); - child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen); + child = osd_lookup_one_len(dev, 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); + rc = osd_ios_scan_one(fill_buf->oifb_info, dev, + fill_buf->oifb_dentry->d_inode, child->d_inode, + &map->olm_fid, name, namelen, map->olm_flags); dput(child); RETURN(rc); } #ifdef HAVE_FILLDIR_USE_CTX -static int osd_ios_uld_fill(struct dir_context *buf, const char *name, - int namelen, +static int osd_ios_uld_fill(struct dir_context *buf, #else -static int osd_ios_uld_fill(void *buf, const char *name, int namelen, +static int osd_ios_uld_fill(void *buf, #endif + const char *name, int namelen, loff_t offset, __u64 ino, unsigned d_type) { struct osd_ios_filldir_buf *fill_buf = (struct osd_ios_filldir_buf *)buf; + struct osd_device *dev = fill_buf->oifb_dev; struct dentry *child; struct lu_fid tfid; int rc = 0; ENTRY; + fill_buf->oifb_items++; + /* skip any non-DFID format name */ if (name[0] != '[') RETURN(0); - child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen); + child = osd_lookup_one_len(dev, name, fill_buf->oifb_dentry, namelen); if (IS_ERR(child)) RETURN(PTR_ERR(child)); @@ -2074,7 +2179,8 @@ static int osd_ios_uld_fill(void *buf, const char *name, int namelen, sscanf(&name[1], SFID, RFID(&tfid)); if (fid_is_sane(&tfid)) rc = osd_ios_scan_one(fill_buf->oifb_info, fill_buf->oifb_dev, - child->d_inode, &tfid, 0); + fill_buf->oifb_dentry->d_inode, + child->d_inode, &tfid, name, namelen, 0); else rc = -EIO; dput(child); @@ -2083,11 +2189,11 @@ static int osd_ios_uld_fill(void *buf, const char *name, int namelen, } #ifdef HAVE_FILLDIR_USE_CTX -static int osd_ios_root_fill(struct dir_context *buf, const char *name, - int namelen, +static int osd_ios_root_fill(struct dir_context *buf, #else -static int osd_ios_root_fill(void *buf, const char *name, int namelen, +static int osd_ios_root_fill(void *buf, #endif + const char *name, int namelen, loff_t offset, __u64 ino, unsigned d_type) { struct osd_ios_filldir_buf *fill_buf = @@ -2098,6 +2204,8 @@ static int osd_ios_root_fill(void *buf, const char *name, int namelen, int rc = 0; ENTRY; + fill_buf->oifb_items++; + /* skip any '.' started names */ if (name[0] == '.') RETURN(0); @@ -2113,16 +2221,20 @@ static int osd_ios_root_fill(void *buf, const char *name, int namelen, if (map->olm_name == NULL) RETURN(0); - child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen); + child = osd_lookup_one_len(dev, name, fill_buf->oifb_dentry, namelen); if (IS_ERR(child)) RETURN(PTR_ERR(child)); + else if (!child->d_inode) + GOTO(out_put, rc = -ENOENT); 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); + rc = osd_ios_scan_one(fill_buf->oifb_info, dev, + fill_buf->oifb_dentry->d_inode, child->d_inode, + &map->olm_fid, name, namelen, 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); +out_put: dput(child); RETURN(rc); @@ -2133,9 +2245,7 @@ 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 }; @@ -2149,19 +2259,22 @@ osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev, filp->f_pos = 0; filp->f_path.dentry = dentry; - filp->f_mode = FMODE_64BITHASH; + filp->f_flags |= O_NOATIME; + filp->f_mode = FMODE_64BITHASH | FMODE_NONOTIFY; filp->f_mapping = inode->i_mapping; filp->f_op = fops; filp->private_data = NULL; + filp->f_cred = current_cred(); set_file_inode(filp, inode); + rc = osd_security_file_alloc(filp); + if (rc) + RETURN(rc); -#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 + do { + buf.oifb_items = 0; + rc = iterate_dir(filp, &buf.ctx); + } while (rc >= 0 && buf.oifb_items > 0 && + filp->f_pos != LDISKFS_HTREE_EOF_64BIT); fops->release(inode, filp); RETURN(rc); @@ -2171,10 +2284,10 @@ 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; + struct lustre_scrub *scrub = &dev->od_scrub.os_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 @@ -2191,53 +2304,60 @@ osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev, * 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)); + child = osd_lookup_one_len_unlocked(dev, 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; - } + if (PTR_ERR(child) != -ENOENT) + RETURN(PTR_ERR(child)); + goto out_scrub; + } + + /* 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). + */ + inode_lock(dentry->d_inode); + rc = osd_ios_scan_one(info, dev, dentry->d_inode, + child->d_inode, &LU_DOT_LUSTRE_FID, + dot_lustre_name, + strlen(dot_lustre_name), 0); + inode_unlock(dentry->d_inode); + if (rc == -ENOENT) { +out_scrub: + /* It is 1.8 MDT device. */ + if (!(sf->sf_flags & SF_UPGRADE)) { + scrub_file_reset(scrub, dev->od_uuid, + SF_UPGRADE); + sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID; + rc = scrub_file_store(info->oti_env, 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); + } else if (rc == 0) { + rc = osd_ios_new_item(dev, child, osd_ios_general_scan, + osd_ios_dl_fill); } + dput(child); RETURN(rc); } @@ -2246,68 +2366,76 @@ 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; + struct lustre_scrub *scrub = &dev->od_scrub.os_scrub; + struct scrub_file *sf = &scrub->os_file; + struct dentry *child; + int rc; ENTRY; if (unlikely(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID)) { sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID; - rc = osd_scrub_file_store(scrub); + rc = scrub_file_store(info->oti_env, scrub); if (rc != 0) RETURN(rc); } - child = osd_ios_lookup_one_len(ADMIN_USR, dentry, strlen(ADMIN_USR)); - if (!IS_ERR(child)) { - rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0); - dput(child); - } else { + child = osd_lookup_one_len_unlocked(dev, ADMIN_USR, dentry, + strlen(ADMIN_USR)); + if (IS_ERR(child)) { rc = PTR_ERR(child); + } else { + inode_lock(dentry->d_inode); + rc = osd_ios_scan_one(info, dev, dentry->d_inode, + child->d_inode, NULL, ADMIN_USR, + strlen(ADMIN_USR), 0); + inode_unlock(dentry->d_inode); + dput(child); } 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); - } + GOTO(out, rc); - if (rc == -ENOENT) - rc = 0; + child = osd_lookup_one_len_unlocked(dev, ADMIN_GRP, dentry, + strlen(ADMIN_GRP)); + if (IS_ERR(child)) + GOTO(out, rc = PTR_ERR(child)); - RETURN(rc); + inode_lock(dentry->d_inode); + rc = osd_ios_scan_one(info, dev, dentry->d_inode, + child->d_inode, NULL, ADMIN_GRP, + strlen(ADMIN_GRP), 0); + inode_unlock(dentry->d_inode); + dput(child); +out: + RETURN(rc == -ENOENT ? 0 : rc); } -static int osd_initial_OI_scrub(struct osd_thread_info *info, - struct osd_device *dev) +static void 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; /* Lookup IGIF in OI by force for initial OI scrub. */ dev->od_igif_inoi = 1; while (1) { - rc = scandir(info, dev, dentry, filldir); + /* Don't take inode_lock here since scandir() callbacks + * can call VFS functions which may manully take the + * inode lock itself like iterate_dir(). Since this + * is the case it is best to leave the scandir() + * callbacks to managing the inode lock. + */ + scandir(info, dev, dentry, filldir); if (item != NULL) { dput(item->oii_dentry); OBD_FREE_PTR(item); } - if (rc != 0) - break; - if (list_empty(&dev->od_ios_list)) break; @@ -2321,17 +2449,6 @@ static int osd_initial_OI_scrub(struct osd_thread_info *info, dentry = item->oii_dentry; } - 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 (rc != 0) - RETURN(rc); - /* There maybe the case that the object has been removed, but its OI * mapping is still in the OI file, such as the "CATALOGS" after MDT * file-level backup/restore. So here cleanup the stale OI mappings. */ @@ -2343,19 +2460,45 @@ static int osd_initial_OI_scrub(struct osd_thread_info *info, continue; } - child = osd_ios_lookup_one_len(map->olm_name, - osd_sb(dev)->s_root, - map->olm_namelen); - if (!IS_ERR(child)) - dput(child); - else if (PTR_ERR(child) == -ENOENT) + child = osd_lookup_one_len_unlocked(dev, map->olm_name, + osd_sb(dev)->s_root, + map->olm_namelen); + if (PTR_ERR(child) == -ENOENT || + (!IS_ERR(child) && !child->d_inode)) osd_scrub_refresh_mapping(info, dev, &map->olm_fid, NULL, DTO_INDEX_DELETE, true, 0, NULL); + if (!IS_ERR(child)) + dput(child); map++; } - RETURN(0); + if (!list_empty(&dev->od_index_restore_list)) { + char *buf; + + OBD_ALLOC_LARGE(buf, INDEX_BACKUP_BUFSIZE); + if (!buf) + CERROR("%s: not enough RAM for rebuild index\n", + osd_name(dev)); + + while (!list_empty(&dev->od_index_restore_list)) { + struct lustre_index_restore_unit *liru; + + liru = list_entry(dev->od_index_restore_list.next, + struct lustre_index_restore_unit, + liru_link); + list_del(&liru->liru_link); + if (buf) + osd_index_restore(info->oti_env, dev, liru, + buf, INDEX_BACKUP_BUFSIZE); + OBD_FREE(liru, liru->liru_len); + } + + if (buf) + OBD_FREE_LARGE(buf, INDEX_BACKUP_BUFSIZE); + } + + EXIT; } char *osd_lf_fid2name(const struct lu_fid *fid) @@ -2379,112 +2522,39 @@ char *osd_lf_fid2name(const struct lu_fid *fid) /* 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; - - /* os_lock: sync status between stop and scrub thread */ - spin_lock(&scrub->os_lock); - -again: - if (thread_is_running(thread)) { - spin_unlock(&scrub->os_lock); - if (!(scrub->os_file.sf_flags & SF_AUTO || - scrub->os_partial_scan) || - (flags & SS_AUTO_PARTIAL)) - RETURN(-EALREADY); - - osd_scrub_join(dev, flags, false); - spin_lock(&scrub->os_lock); - if (!thread_is_running(thread)) - goto again; - - spin_unlock(&scrub->os_lock); - RETURN(0); - } - - if (unlikely(thread_is_stopping(thread))) { - spin_unlock(&scrub->os_lock); - l_wait_event(thread->t_ctl_waitq, - thread_is_stopped(thread), - &lwi); - spin_lock(&scrub->os_lock); - goto again; - } - spin_unlock(&scrub->os_lock); - - if (scrub->os_file.sf_status == SS_COMPLETED) { - if (!(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); - 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); - } - - l_wait_event(thread->t_ctl_waitq, - thread_is_running(thread) || thread_is_stopped(thread), - &lwi); - - RETURN(0); -} - -int osd_scrub_start(struct osd_device *dev, __u32 flags) +int osd_scrub_start(const struct lu_env *env, struct osd_device *dev, + __u32 flags) { + struct lustre_scrub *scrub = &dev->od_scrub.os_scrub; int rc; ENTRY; + if (dev->od_dt_dev.dd_rdonly) + RETURN(-EROFS); + /* od_otable_mutex: prevent curcurrent start/stop */ mutex_lock(&dev->od_otable_mutex); - rc = do_osd_scrub_start(dev, flags); + rc = scrub_start(osd_scrub_main, scrub, dev, flags); + if (rc == -EALREADY) { + rc = 0; + if ((scrub->os_file.sf_flags & SF_AUTO || + scrub->os_partial_scan) && + !(flags & SS_AUTO_PARTIAL)) + osd_scrub_join(env, dev, flags, false); + } mutex_unlock(&dev->od_otable_mutex); - RETURN(rc == -EALREADY ? 0 : rc); + RETURN(rc); } -static void do_osd_scrub_stop(struct osd_scrub *scrub) +void osd_scrub_stop(struct osd_device *dev) { - struct ptlrpc_thread *thread = &scrub->os_thread; - struct l_wait_info lwi = { 0 }; - - /* os_lock: sync status between stop and scrub thread */ - spin_lock(&scrub->os_lock); - if (!thread_is_init(thread) && !thread_is_stopped(thread)) { - thread_set_flags(thread, SVC_STOPPING); - 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. */ - spin_lock(&scrub->os_lock); - } - spin_unlock(&scrub->os_lock); -} + struct lustre_scrub *scrub = &dev->od_scrub.os_scrub; -static void osd_scrub_stop(struct osd_device *dev) -{ /* od_otable_mutex: prevent curcurrent start/stop */ mutex_lock(&dev->od_otable_mutex); - dev->od_scrub.os_paused = 1; - do_osd_scrub_stop(&dev->od_scrub); + scrub->os_paused = 1; + scrub_stop(scrub); mutex_unlock(&dev->od_otable_mutex); } @@ -2494,57 +2564,73 @@ static const char osd_scrub_name[] = "OI_scrub"; int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev) { - struct osd_thread_info *info = osd_oti_get(env); - struct osd_scrub *scrub = &dev->od_scrub; - struct lvfs_run_ctxt *ctxt = &scrub->os_ctxt; - struct scrub_file *sf = &scrub->os_file; - struct super_block *sb = osd_sb(dev); - struct ldiskfs_super_block *es = LDISKFS_SB(sb)->s_es; - struct lvfs_run_ctxt saved; - struct file *filp; - struct inode *inode; - struct lu_fid *fid = &info->oti_fid; - bool dirty = false; - bool restored = false; - int rc = 0; + struct osd_thread_info *info = osd_oti_get(env); + struct lustre_scrub *scrub = &dev->od_scrub.os_scrub; + struct lvfs_run_ctxt *ctxt = &dev->od_scrub.os_ctxt; + struct scrub_file *sf = &scrub->os_file; + struct super_block *sb = osd_sb(dev); + struct lvfs_run_ctxt saved; + struct file *filp; + struct inode *inode; + struct lu_fid *fid = &info->oti_fid; + struct osd_inode_id *id = &info->oti_id; + struct dt_object *obj; + bool dirty = false; + bool restored = false; + int rc = 0; ENTRY; - memset(scrub, 0, sizeof(*scrub)); + memset(&dev->od_scrub, 0, sizeof(struct osd_scrub)); OBD_SET_CTXT_MAGIC(ctxt); ctxt->pwdmnt = dev->od_mnt; ctxt->pwd = dev->od_mnt->mnt_root; - ctxt->fs = get_ds(); + ctxt->fs = KERNEL_DS; 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); + scrub->os_name = osd_name(dev); push_ctxt(&saved, ctxt); - filp = filp_open(osd_scrub_name, O_RDWR | O_CREAT, 0644); + filp = filp_open(osd_scrub_name, O_RDWR | + (dev->od_dt_dev.dd_rdonly ? 0 : O_CREAT), 0644); if (IS_ERR(filp)) { pop_ctxt(&saved, ctxt); RETURN(PTR_ERR(filp)); } inode = file_inode(filp); - /* 'What the @fid is' is not imporatant, because the object - * has no OI mapping, and only is visible inside the OSD.*/ - lu_igif_build(fid, inode->i_ino, inode->i_generation); - rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0); - if (rc != 0) { - filp_close(filp, NULL); - pop_ctxt(&saved, ctxt); - RETURN(rc); + if (!dev->od_dt_dev.dd_rdonly) { + /* '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) { + filp_close(filp, NULL); + pop_ctxt(&saved, ctxt); + RETURN(rc); + } } - scrub->os_inode = igrab(inode); + osd_id_gen(id, inode->i_ino, inode->i_generation); + osd_add_oi_cache(info, dev, id, fid); 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); + obj = lu2dt(lu_object_find_slice(env, osd2lu_dev(dev), fid, NULL)); + if (IS_ERR_OR_NULL(obj)) + RETURN(obj ? PTR_ERR(obj) : -ENOENT); + +#ifndef HAVE_S_UUID_AS_UUID_T + memcpy(dev->od_uuid.b, sb->s_uuid, UUID_SIZE); +#else + uuid_copy(&dev->od_uuid, &sb->s_uuid); +#endif + scrub->os_obj = obj; + rc = scrub_file_load(env, scrub); + if (rc == -ENOENT || rc == -EFAULT) { + scrub_file_init(scrub, dev->od_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, @@ -2554,44 +2640,34 @@ int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev) * 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 + * need to check "filter_fid_18_23" 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) + if (dev->od_maybe_new && rc == -ENOENT) sf->sf_internal_flags = SIF_NO_HANDLE_OLD_FID; dirty = true; - } else if (rc != 0) { - GOTO(cleanup_inode, rc); + } else if (rc < 0) { + GOTO(cleanup_obj, rc); } else { - if (memcmp(sf->sf_uuid, es->s_uuid, 16) != 0) { - struct obd_uuid *old_uuid; - struct obd_uuid *new_uuid; - - OBD_ALLOC_PTR(old_uuid); - OBD_ALLOC_PTR(new_uuid); - if (old_uuid == NULL || new_uuid == NULL) { - CERROR("%.16s: UUID has been changed, but" - "failed to allocate RAM for report\n", - LDISKFS_SB(sb)->s_es->s_volume_name); - } else { - class_uuid_unparse(sf->sf_uuid, old_uuid); - class_uuid_unparse(es->s_uuid, new_uuid); - CERROR("%.16s: UUID has been changed from " - "%s to %s\n", - LDISKFS_SB(sb)->s_es->s_volume_name, - old_uuid->uuid, new_uuid->uuid); - } - osd_scrub_file_reset(scrub, es->s_uuid,SF_INCONSISTENT); + if (!uuid_equal(&sf->sf_uuid, &dev->od_uuid)) { + CDEBUG(D_LFSCK, + "%s: UUID has been changed from %pU to %pU\n", + osd_dev2name(dev), &sf->sf_uuid, &dev->od_uuid); + scrub_file_reset(scrub, dev->od_uuid, SF_INCONSISTENT); dirty = true; restored = true; - if (old_uuid != NULL) - OBD_FREE_PTR(old_uuid); - if (new_uuid != NULL) - OBD_FREE_PTR(new_uuid); } else if (sf->sf_status == SS_SCANNING) { sf->sf_status = SS_CRASHED; dirty = true; } + + if ((sf->sf_oi_count & (sf->sf_oi_count - 1)) != 0) { + LCONSOLE_WARN("%s: invalid oi count %d, set it to %d\n", + osd_dev2name(dev), sf->sf_oi_count, + osd_oi_count); + sf->sf_oi_count = osd_oi_count; + dirty = true; + } } if (sf->sf_pos_last_checkpoint != 0) @@ -2600,19 +2676,18 @@ int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev) scrub->os_pos_current = LDISKFS_FIRST_INO(sb) + 1; if (dirty) { - rc = osd_scrub_file_store(scrub); - if (rc != 0) - GOTO(cleanup_inode, rc); + rc = scrub_file_store(env, scrub); + if (rc) + GOTO(cleanup_obj, rc); } /* Initialize OI files. */ rc = osd_oi_init(info, dev, restored); if (rc < 0) - GOTO(cleanup_inode, rc); + GOTO(cleanup_obj, rc); - rc = osd_initial_OI_scrub(info, dev); - if (rc != 0) - GOTO(cleanup_oi, rc); + if (!dev->od_dt_dev.dd_rdonly) + osd_initial_OI_scrub(info, dev); if (sf->sf_flags & SF_UPGRADE || !(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID || @@ -2635,7 +2710,8 @@ int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev) * later if found that the system is upgrading. */ dev->od_igif_inoi = 1; - if (!dev->od_noscrub && + if (!dev->od_dt_dev.dd_rdonly && + dev->od_auto_scrub_interval != AS_NEVER && ((sf->sf_status == SS_PAUSED) || (sf->sf_status == SS_CRASHED && sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | @@ -2643,7 +2719,7 @@ int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev) (sf->sf_status == SS_INIT && sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE)))) - rc = osd_scrub_start(dev, SS_AUTO_FULL); + rc = osd_scrub_start(env, dev, SS_AUTO_FULL); if (rc != 0) GOTO(cleanup_oi, rc); @@ -2658,23 +2734,23 @@ int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev) RETURN(0); cleanup_oi: osd_oi_fini(info, dev); -cleanup_inode: - iput(scrub->os_inode); - scrub->os_inode = NULL; +cleanup_obj: + dt_object_put_nocache(env, scrub->os_obj); + scrub->os_obj = NULL; return rc; } void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev) { - struct osd_scrub *scrub = &dev->od_scrub; + struct lustre_scrub *scrub = &dev->od_scrub.os_scrub; LASSERT(dev->od_otable_it == NULL); - if (scrub->os_inode != NULL) { + if (scrub->os_obj != NULL) { osd_scrub_stop(dev); - iput(scrub->os_inode); - scrub->os_inode = NULL; + dt_object_put_nocache(env, scrub->os_obj); + scrub->os_obj = NULL; } if (dev->od_oi_table != NULL) osd_oi_fini(osd_oti_get(env), dev); @@ -2688,7 +2764,7 @@ static struct dt_it *osd_otable_it_init(const struct lu_env *env, enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT; enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK; struct osd_device *dev = osd_dev(dt->do_lu.lo_dev); - struct osd_scrub *scrub = &dev->od_scrub; + struct lustre_scrub *scrub = &dev->od_scrub.os_scrub; struct osd_otable_it *it; __u32 start = 0; int rc; @@ -2726,15 +2802,19 @@ static struct dt_it *osd_otable_it_init(const struct lu_env *env, start |= SS_CLEAR_DRYRUN; } - rc = do_osd_scrub_start(dev, start & ~SS_AUTO_PARTIAL); - if (rc < 0 && rc != -EALREADY) { + rc = scrub_start(osd_scrub_main, scrub, dev, start & ~SS_AUTO_PARTIAL); + if (rc == -EALREADY) { + it->ooi_cache.ooc_pos_preload = scrub->os_pos_current; + } else if (rc < 0) { dev->od_otable_it = NULL; OBD_FREE_PTR(it); - GOTO(out, it = ERR_PTR(rc)); + it = ERR_PTR(rc); + } else { + /* We have to start from the begining. */ + it->ooi_cache.ooc_pos_preload = + LDISKFS_FIRST_INO(osd_sb(dev)) + 1; } - it->ooi_cache.ooc_pos_preload = scrub->os_pos_current; - GOTO(out, it); out: @@ -2749,7 +2829,7 @@ static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di) /* od_otable_mutex: prevent curcurrent init/fini */ mutex_lock(&dev->od_otable_mutex); - do_osd_scrub_stop(&dev->od_scrub); + scrub_stop(&dev->od_scrub.os_scrub); LASSERT(dev->od_otable_it == it); dev->od_otable_it = NULL; @@ -2768,7 +2848,7 @@ static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di) } static inline int -osd_otable_it_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it) +osd_otable_it_wakeup(struct lustre_scrub *scrub, struct osd_otable_it *it) { spin_lock(&scrub->os_lock); if (it->ooi_cache.ooc_pos_preload < scrub->os_pos_current || @@ -2786,10 +2866,9 @@ static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di) { struct osd_otable_it *it = (struct osd_otable_it *)di; struct osd_device *dev = it->ooi_dev; - struct osd_scrub *scrub = &dev->od_scrub; + struct lustre_scrub *scrub = &dev->od_scrub.os_scrub; struct osd_otable_cache *ooc = &it->ooi_cache; struct ptlrpc_thread *thread = &scrub->os_thread; - struct l_wait_info lwi = { 0 }; int rc; ENTRY; @@ -2807,9 +2886,8 @@ again: } if (it->ooi_all_cached) { - l_wait_event(thread->t_ctl_waitq, - !thread_is_running(thread), - &lwi); + wait_event_idle(thread->t_ctl_waitq, + !thread_is_running(thread)); RETURN(1); } @@ -2821,9 +2899,8 @@ again: } 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); + wait_event_idle(thread->t_ctl_waitq, + osd_otable_it_wakeup(scrub, it)); if (!thread_is_running(thread) && !it->ooi_used_outside) RETURN(1); @@ -2887,7 +2964,8 @@ static int osd_otable_it_load(const struct lu_env *env, struct osd_otable_it *it = (struct osd_otable_it *)di; struct osd_device *dev = it->ooi_dev; struct osd_otable_cache *ooc = &it->ooi_cache; - struct osd_scrub *scrub = &dev->od_scrub; + struct lustre_scrub *scrub = &dev->od_scrub.os_scrub; + struct osd_iit_param *param = &it->ooi_iit_param; int rc; ENTRY; @@ -2900,7 +2978,9 @@ static int osd_otable_it_load(const struct lu_env *env, if (hash > OSD_OTABLE_MAX_HASH) hash = OSD_OTABLE_MAX_HASH; - ooc->ooc_pos_preload = hash; + /* The hash is the last checkpoint position, + * we will start from the next one. */ + ooc->ooc_pos_preload = hash + 1; if (ooc->ooc_pos_preload <= LDISKFS_FIRST_INO(osd_sb(dev))) ooc->ooc_pos_preload = LDISKFS_FIRST_INO(osd_sb(dev)) + 1; @@ -2908,6 +2988,15 @@ static int osd_otable_it_load(const struct lu_env *env, if (!scrub->os_full_speed) wake_up_all(&scrub->os_thread.t_ctl_waitq); + memset(param, 0, sizeof(*param)); + param->sb = osd_sb(dev); + param->start = ooc->ooc_pos_preload; + param->bg = (ooc->ooc_pos_preload - 1) / + LDISKFS_INODES_PER_GROUP(param->sb); + param->offset = (ooc->ooc_pos_preload - 1) % + LDISKFS_INODES_PER_GROUP(param->sb); + param->gbase = 1 + param->bg * LDISKFS_INODES_PER_GROUP(param->sb); + /* Unplug OSD layer iteration by the first next() call. */ rc = osd_otable_it_next(env, (struct dt_it *)it); @@ -2944,9 +3033,10 @@ int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic, int insert) { struct osd_inconsistent_item *oii; - struct osd_scrub *scrub = &dev->od_scrub; - struct ptlrpc_thread *thread = &scrub->os_thread; - int wakeup = 0; + struct osd_scrub *oscrub = &dev->od_scrub; + struct lustre_scrub *lscrub = &oscrub->os_scrub; + struct ptlrpc_thread *thread = &lscrub->os_thread; + int wakeup = 0; ENTRY; OBD_ALLOC_PTR(oii); @@ -2957,37 +3047,37 @@ int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic, oii->oii_cache = *oic; oii->oii_insert = insert; - if (scrub->os_partial_scan) { - __u64 now = cfs_time_current_sec(); + if (lscrub->os_partial_scan) { + __u64 now = ktime_get_real_seconds(); /* 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 + + while (unlikely(oscrub->os_bad_oimap_count > 0 && + oscrub->os_bad_oimap_time + SCRUB_BAD_OIMAP_DECAY_INTERVAL < now)) { - scrub->os_bad_oimap_count >>= 1; - scrub->os_bad_oimap_time += + oscrub->os_bad_oimap_count >>= 1; + oscrub->os_bad_oimap_time += SCRUB_BAD_OIMAP_DECAY_INTERVAL; } - scrub->os_bad_oimap_time = now; - if (++scrub->os_bad_oimap_count > + oscrub->os_bad_oimap_time = now; + if (++oscrub->os_bad_oimap_count > dev->od_full_scrub_threshold_rate) - scrub->os_full_scrub = 1; + lscrub->os_full_scrub = 1; } - spin_lock(&scrub->os_lock); + spin_lock(&lscrub->os_lock); if (unlikely(!thread_is_running(thread))) { - spin_unlock(&scrub->os_lock); + spin_unlock(&lscrub->os_lock); OBD_FREE_PTR(oii); RETURN(-EAGAIN); } - if (list_empty(&scrub->os_inconsistent_items)) + if (list_empty(&lscrub->os_inconsistent_items)) wakeup = 1; - list_add_tail(&oii->oii_list, &scrub->os_inconsistent_items); - spin_unlock(&scrub->os_lock); + list_add_tail(&oii->oii_list, &lscrub->os_inconsistent_items); + spin_unlock(&lscrub->os_lock); if (wakeup != 0) wake_up_all(&thread->t_ctl_waitq); @@ -2998,7 +3088,7 @@ int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic, int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid, struct osd_inode_id *id) { - struct osd_scrub *scrub = &dev->od_scrub; + struct lustre_scrub *scrub = &dev->od_scrub.os_scrub; struct osd_inconsistent_item *oii; ENTRY; @@ -3015,153 +3105,17 @@ int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid, RETURN(-ENOENT); } -/* OI scrub dump */ - -static const char *scrub_status_names[] = { - "init", - "scanning", - "completed", - "failed", - "stopped", - "paused", - "crashed", - NULL -}; - -static const char *scrub_flags_names[] = { - "recreated", - "inconsistent", - "auto", - "upgrade", - NULL -}; - -static const char *scrub_param_names[] = { - "failout", - "dryrun", - NULL -}; - -static void scrub_bits_dump(struct seq_file *m, int bits, const char *names[], - const char *prefix) +void osd_scrub_dump(struct seq_file *m, struct osd_device *dev) { - int flag; - int i; - - seq_printf(m, "%s:%c", prefix, bits != 0 ? ' ' : '\n'); - - for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) { - if (flag & bits) { - bits &= ~flag; - seq_printf(m, "%s%c", names[i], - bits != 0 ? ',' : '\n'); - } - } -} - -static void scrub_time_dump(struct seq_file *m, __u64 time, const char *prefix) -{ - if (time != 0) - seq_printf(m, "%s: "LPU64" seconds\n", prefix, - cfs_time_current_sec() - time); - else - seq_printf(m, "%s: N/A\n", prefix); -} - -static void scrub_pos_dump(struct seq_file *m, __u64 pos, const char *prefix) -{ - if (pos != 0) - seq_printf(m, "%s: "LPU64"\n", prefix, pos); - else - seq_printf(m, "%s: N/A\n", prefix); -} + struct osd_scrub *scrub = &dev->od_scrub; -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; - - down_read(&scrub->os_rwsem); - 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]); - - scrub_bits_dump(m, sf->sf_flags, scrub_flags_names, "flags"); - - scrub_bits_dump(m, sf->sf_param, scrub_param_names, "param"); - - scrub_time_dump(m, sf->sf_time_last_complete, - "time_since_last_completed"); - - scrub_time_dump(m, sf->sf_time_latest_start, - "time_since_latest_start"); - - scrub_time_dump(m, sf->sf_time_last_checkpoint, - "time_since_last_checkpoint"); - - scrub_pos_dump(m, sf->sf_pos_latest_start, - "latest_start_position"); - - scrub_pos_dump(m, sf->sf_pos_last_checkpoint, - "last_checkpoint_position"); - - scrub_pos_dump(m, sf->sf_pos_first_inconsistent, - "first_failure_position"); - - checked = sf->sf_items_checked + scrub->os_new_checked; - seq_printf(m, "checked: "LPU64"\n" - "updated: "LPU64"\n" - "failed: "LPU64"\n" - "prior_updated: "LPU64"\n" - "noscrub: "LPU64"\n" - "igif: "LPU64"\n" - "success_count: %u\n", - checked, sf->sf_items_updated, sf->sf_items_failed, - sf->sf_items_updated_prior, sf->sf_items_noscrub, - sf->sf_items_igif, sf->sf_success_count); - - speed = checked; - if (thread_is_running(&scrub->os_thread)) { - cfs_duration_t duration = cfs_time_current() - - scrub->os_time_last_checkpoint; - __u64 new_checked = msecs_to_jiffies(scrub->os_new_checked * - MSEC_PER_SEC); - __u32 rtime = sf->sf_run_time + - cfs_duration_sec(duration + HALF_SEC); - - if (duration != 0) - do_div(new_checked, duration); - if (rtime != 0) - do_div(speed, rtime); - seq_printf(m, "run_time: %u seconds\n" - "average_speed: "LPU64" objects/sec\n" - "real-time_speed: "LPU64" objects/sec\n" - "current_position: %u\n" - "lf_scanned: "LPU64"\n" - "lf_repaired: "LPU64"\n" - "lf_failed: "LPU64"\n", - rtime, speed, new_checked, scrub->os_pos_current, - scrub->os_lf_scanned, scrub->os_lf_repaired, - scrub->os_lf_failed); - } else { - if (sf->sf_run_time != 0) - do_div(speed, sf->sf_run_time); - seq_printf(m, "run_time: %u seconds\n" - "average_speed: "LPU64" objects/sec\n" - "real-time_speed: N/A\n" - "current_position: N/A\n" - "lf_scanned: "LPU64"\n" - "lf_repaired: "LPU64"\n" - "lf_failed: "LPU64"\n", - sf->sf_run_time, speed, scrub->os_lf_scanned, - scrub->os_lf_repaired, scrub->os_lf_failed); - } - - up_read(&scrub->os_rwsem); - return 0; + scrub_dump(m, &scrub->os_scrub); + seq_printf(m, "lf_scanned: %llu\n" + "lf_%s: %llu\n" + "lf_failed: %llu\n", + scrub->os_lf_scanned, + scrub->os_scrub.os_file.sf_param & SP_DRYRUN ? + "inconsistent" : "repaired", + scrub->os_lf_repaired, + scrub->os_lf_failed); }