* GPL HEADER END
*/
/*
- * Copyright (c) 2012, Intel Corporation.
+ * Copyright (c) 2012, 2013, Intel Corporation.
*/
/*
* lustre/osd-ldiskfs/osd_scrub.c
* Author: Fan Yong <yong.fan@whamcloud.com>
*/
-#ifndef EXPORT_SYMTAB
-# define EXPORT_SYMTAB
-#endif
#define DEBUG_SUBSYSTEM S_MDS
#include <lustre/lustre_idl.h>
#include <lustre_disk.h>
#include <dt_object.h>
+#include <linux/xattr.h>
#include "osd_internal.h"
#include "osd_oi.h"
#include "osd_scrub.h"
-#define HALF_SEC (CFS_HZ >> 1)
+#define HALF_SEC (HZ >> 1)
#define OSD_OTABLE_MAX_HASH 0x00000000ffffffffULL
#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 int osd_scrub_refresh_mapping(struct osd_thread_info *info,
struct osd_device *dev,
const struct lu_fid *fid,
- const struct osd_inode_id *id, int ops)
+ const struct osd_inode_id *id,
+ int ops, enum oi_check_flags flags)
{
- struct lu_fid *oi_fid = &info->oti_fid2;
- struct osd_inode_id *oi_id = &info->oti_id2;
- struct iam_container *bag;
- struct iam_path_descr *ipd;
- handle_t *jh;
- int rc;
+ const struct lu_env *env = info->oti_env;
+ struct thandle *th;
+ struct osd_thandle *oh;
+ int rc;
ENTRY;
- fid_cpu_to_be(oi_fid, fid);
- osd_id_pack(oi_id, id);
- jh = ldiskfs_journal_start_sb(osd_sb(dev),
- osd_dto_credits_noquota[ops]);
- if (IS_ERR(jh)) {
- rc = PTR_ERR(jh);
- CERROR("%.16s: fail to start trans for scrub store: rc = %d\n",
- LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, rc);
- RETURN(rc);
- }
+ th = dt_trans_create(env, &dev->od_dt_dev);
+ if (IS_ERR(th))
+ RETURN(PTR_ERR(th));
- bag = &osd_fid2oi(dev, fid)->oi_dir.od_container;
- ipd = osd_idx_ipd_get(info->oti_env, bag);
- if (unlikely(ipd == NULL)) {
- ldiskfs_journal_stop(jh);
- CERROR("%.16s: fail to get ipd for scrub store\n",
- LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name);
- RETURN(-ENOMEM);
- }
+ oh = container_of0(th, struct osd_thandle, ot_super);
+ LASSERT(oh->ot_handle == NULL);
switch (ops) {
case DTO_INDEX_UPDATE:
- rc = iam_update(jh, bag, (const struct iam_key *)oi_fid,
- (struct iam_rec *)oi_id, ipd);
+ osd_trans_declare_op(env, oh, OSD_OT_UPDATE,
+ osd_dto_credits_noquota[DTO_INDEX_UPDATE]);
+ rc = dt_trans_start_local(env, &dev->od_dt_dev, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ rc = osd_oi_update(info, dev, fid, id, th, flags);
if (unlikely(rc == -ENOENT)) {
- /* Some unlink thread mat removed the OI mapping. */
+ /* Some unlink thread may removed the OI mapping. */
rc = 1;
}
break;
case DTO_INDEX_INSERT:
- rc = iam_insert(jh, bag, (const struct iam_key *)oi_fid,
- (struct iam_rec *)oi_id, ipd);
+ osd_trans_declare_op(env, oh, OSD_OT_INSERT,
+ osd_dto_credits_noquota[DTO_INDEX_INSERT]);
+ rc = dt_trans_start_local(env, &dev->od_dt_dev, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ rc = osd_oi_insert(info, dev, fid, id, th, flags);
if (unlikely(rc == -EEXIST)) {
rc = 1;
/* XXX: There are trouble things when adding OI
}
break;
case DTO_INDEX_DELETE:
- rc = iam_delete(jh, bag, (const struct iam_key *)oi_fid, ipd);
+ osd_trans_declare_op(env, oh, OSD_OT_DELETE,
+ osd_dto_credits_noquota[DTO_INDEX_DELETE]);
+ rc = dt_trans_start_local(env, &dev->od_dt_dev, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ rc = osd_oi_delete(info, dev, fid, th, flags);
if (rc == -ENOENT) {
/* It is normal that the unlink thread has removed the
* OI mapping already. */
LASSERTF(0, "Unexpected ops %d\n", ops);
break;
}
- osd_ipd_put(info->oti_env, bag, ipd);
- ldiskfs_journal_stop(jh);
- RETURN(rc);
+
+ GOTO(stop, rc);
+
+stop:
+ dt_trans_stop(env, &dev->od_dt_dev, th);
+ return rc;
}
/* OI_scrub file ops */
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);
}
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);
}
scrub->os_full_speed = 1;
scrub->os_in_prior = 0;
+ spin_lock(&scrub->os_lock);
scrub->os_waiting = 0;
scrub->os_paused = 0;
+ spin_unlock(&scrub->os_lock);
scrub->os_new_checked = 0;
if (sf->sf_pos_last_checkpoint != 0)
sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
}
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;
+ handle_t *jh;
+ int size = 0;
+ int rc;
+ bool removed = false;
+ bool reset = true;
+ ENTRY;
+
+ /* We want the LMA to fit into the 256-byte OST inode, so operate
+ * as following:
+ * 1) read old XATTR_NAME_FID and save the parent FID;
+ * 2) delete the old XATTR_NAME_FID;
+ * 3) make new LMA and add it;
+ * 4) generate new XATTR_NAME_FID with the saved parent FID and add it.
+ *
+ * Making the LMA to fit into the 256-byte OST inode can save time for
+ * normal osd_check_lma() and for other OI scrub scanning in future.
+ * So it is worth to make some slow conversion here. */
+ jh = ldiskfs_journal_start_sb(osd_sb(dev),
+ osd_dto_credits_noquota[DTO_XATTR_SET] * 3);
+ if (IS_ERR(jh)) {
+ rc = PTR_ERR(jh);
+ CERROR("%s: fail to start trans for convert ff: "DFID
+ ": rc = %d\n",
+ osd_name(dev), PFID(fid), rc);
+ RETURN(rc);
+ }
+
+ /* 1) read old XATTR_NAME_FID and save the parent FID */
+ rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
+ if (rc == sizeof(*ff)) {
+ /* 2) delete the old XATTR_NAME_FID */
+ ll_vfs_dq_init(inode);
+ rc = inode->i_op->removexattr(dentry, XATTR_NAME_FID);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ removed = true;
+ } else if (unlikely(rc == -ENODATA)) {
+ reset = false;
+ } else if (rc != sizeof(struct filter_fid)) {
+ GOTO(stop, rc = -EINVAL);
+ }
+
+ /* 3) make new LMA and add it */
+ rc = osd_ea_fid_set(info, inode, fid, LMAC_FID_ON_OST, 0);
+ if (rc == 0 && reset)
+ size = sizeof(struct filter_fid);
+ else if (rc != 0 && removed)
+ /* If failed, we should try to add the old back. */
+ size = sizeof(struct filter_fid_old);
+
+ /* 4) generate new XATTR_NAME_FID with the saved parent FID and add it*/
+ if (size > 0) {
+ int rc1;
+
+ rc1 = __osd_xattr_set(info, inode, XATTR_NAME_FID, ff, size,
+ XATTR_CREATE);
+ if (rc1 != 0 && rc != 0)
+ rc = rc1;
+ }
+
+ GOTO(stop, rc);
+
+stop:
+ ldiskfs_journal_stop(jh);
+ return rc;
+}
+
+static int
osd_scrub_check_update(struct osd_thread_info *info, struct osd_device *dev,
struct osd_idmap_cache *oic, int val)
{
int ops = DTO_INDEX_UPDATE;
int idx;
int rc;
+ bool converted = false;
ENTRY;
down_write(&scrub->os_rwsem);
if (fid_is_igif(fid))
sf->sf_items_igif++;
+ if (val == SCRUB_NEXT_OSTOBJ_OLD) {
+ inode = osd_iget(info, dev, lid);
+ if (IS_ERR(inode)) {
+ rc = PTR_ERR(inode);
+ /* Someone removed the inode. */
+ if (rc == -ENOENT || rc == -ESTALE)
+ rc = 0;
+ GOTO(out, rc);
+ }
+
+ sf->sf_flags |= SF_UPGRADE;
+ sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
+ dev->od_check_ff = 1;
+ rc = osd_scrub_convert_ff(info, dev, inode, fid);
+ rc = osd_ea_fid_set(info, inode, fid,
+ LMAC_FID_ON_OST, 0);
+ if (rc != 0)
+ GOTO(out, rc);
+
+ converted = true;
+ }
+
if ((val == SCRUB_NEXT_NOLMA) &&
- (!dev->od_handle_nolma || OBD_FAIL_CHECK(OBD_FAIL_FID_NOLMA)))
+ (!scrub->os_convert_igif || OBD_FAIL_CHECK(OBD_FAIL_FID_NOLMA)))
GOTO(out, rc = 0);
if ((oii != NULL && oii->oii_insert) || (val == SCRUB_NEXT_NOLMA))
goto iget;
- /* XXX: Currently, no FID-in-LMA for OST object, so osd_oi_lookup()
- * without checking FLD is enough.
- *
- * It should be updated if FID-in-LMA for OSD object introduced
- * in the future. */
- rc = osd_oi_lookup(info, dev, fid, lid2, false);
+ rc = osd_oi_lookup(info, dev, fid, lid2,
+ (val == SCRUB_NEXT_OSTOBJ ||
+ val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0);
if (rc != 0) {
if (rc != -ENOENT)
GOTO(out, rc);
iget:
- inode = osd_iget(info, dev, lid);
- if (IS_ERR(inode)) {
- rc = PTR_ERR(inode);
- /* Someone removed the inode. */
- if (rc == -ENOENT || rc == -ESTALE)
- rc = 0;
- GOTO(out, rc);
- }
-
- /* Check whether the inode to be unlinked during OI scrub. */
- if (unlikely(inode->i_nlink == 0)) {
- iput(inode);
- GOTO(out, rc = 0);
+ if (inode == NULL) {
+ inode = osd_iget(info, dev, lid);
+ if (IS_ERR(inode)) {
+ rc = PTR_ERR(inode);
+ /* Someone removed the inode. */
+ if (rc == -ENOENT || rc == -ESTALE)
+ rc = 0;
+ GOTO(out, rc);
+ }
}
+ scrub->os_full_speed = 1;
ops = DTO_INDEX_INSERT;
idx = osd_oi_fid2idx(dev, fid);
- if (val == SCRUB_NEXT_NOLMA) {
+ switch (val) {
+ case SCRUB_NEXT_NOLMA:
sf->sf_flags |= SF_UPGRADE;
- scrub->os_full_speed = 1;
- rc = osd_ea_fid_set(info, inode, fid, 0);
+ rc = osd_ea_fid_set(info, inode, fid, 0, 0);
if (rc != 0)
GOTO(out, rc);
- } else {
- sf->sf_flags |= SF_RECREATED | SF_INCONSISTENT;
- scrub->os_full_speed = 1;
+
+ if (!(sf->sf_flags & SF_INCONSISTENT))
+ dev->od_igif_inoi = 0;
+ break;
+ case SCRUB_NEXT_OSTOBJ:
+ sf->sf_flags |= SF_INCONSISTENT;
+ case SCRUB_NEXT_OSTOBJ_OLD:
+ break;
+ default:
+ sf->sf_flags |= SF_RECREATED;
if (unlikely(!ldiskfs_test_bit(idx, sf->sf_oi_bitmap)))
ldiskfs_set_bit(idx, sf->sf_oi_bitmap);
+ break;
}
} else if (osd_id_eq(lid, lid2)) {
+ if (converted)
+ sf->sf_items_updated++;
+
GOTO(out, rc = 0);
} else {
- sf->sf_flags |= SF_INCONSISTENT;
scrub->os_full_speed = 1;
+ sf->sf_flags |= SF_INCONSISTENT;
+
+ /* XXX: If the device is restored from file-level backup, then
+ * some IGIFs may have been already in OI files, and some
+ * may be not yet. Means upgrading from 1.8 may be partly
+ * processed, but some clients may hold some immobilized
+ * IGIFs, and use them to access related objects. Under
+ * such case, OSD does not know whether an given IGIF has
+ * been processed or to be processed, and it also cannot
+ * generate local ino#/gen# directly from the immobilized
+ * IGIF because of the backup/restore. Then force OSD to
+ * lookup the given IGIF in OI files, and if no entry,
+ * then ask the client to retry after upgrading completed.
+ * No better choice. */
+ dev->od_igif_inoi = 1;
}
- rc = osd_scrub_refresh_mapping(info, dev, fid, lid, ops);
+ rc = osd_scrub_refresh_mapping(info, dev, fid, lid, ops,
+ (val == SCRUB_NEXT_OSTOBJ ||
+ val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0);
if (rc == 0) {
if (scrub->os_in_prior)
sf->sf_items_updated_prior++;
else
sf->sf_items_updated++;
+
+ /* The target has been changed, need to be re-loaded. */
+ lu_object_purge(info->oti_env, osd2lu_dev(dev), fid);
}
GOTO(out, rc);
rc = 0;
}
- if (ops == DTO_INDEX_INSERT) {
- /* There may be conflict unlink during the OI scrub,
- * if happend, then remove the new added OI mapping. */
- if (unlikely(inode->i_nlink == 0))
- osd_scrub_refresh_mapping(info, dev, fid, lid,
- DTO_INDEX_DELETE);
- iput(inode);
- }
+ /* There may be conflict unlink during the OI scrub,
+ * if happend, then remove the new added OI mapping. */
+ if (ops == DTO_INDEX_INSERT && inode != NULL && !IS_ERR(inode) &&
+ unlikely(inode->i_nlink == 0))
+ osd_scrub_refresh_mapping(info, dev, fid, lid,
+ DTO_INDEX_DELETE,
+ (val == SCRUB_NEXT_OSTOBJ ||
+ val == SCRUB_NEXT_OSTOBJ_OLD) ?
+ OI_KNOWN_ON_OST : 0);
up_write(&scrub->os_rwsem);
+ if (inode != NULL && !IS_ERR(inode))
+ iput(inode);
+
if (oii != NULL) {
LASSERT(!cfs_list_empty(&oii->oii_list));
}
sf->sf_time_last_checkpoint = cfs_time_current_sec();
if (result > 0) {
+ struct osd_device *dev =
+ container_of0(scrub, struct osd_device, od_scrub);
+
+ dev->od_igif_inoi = 1;
+ dev->od_check_ff = 0;
sf->sf_status = SS_COMPLETED;
memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
}
}
+/**
+ * \retval SCRUB_NEXT_OSTOBJ_OLD: FID-on-OST
+ * \retval 0: FID-on-MDT
+ */
+static int osd_scrub_check_local_fldb(struct osd_thread_info *info,
+ struct osd_device *dev,
+ struct lu_fid *fid)
+{
+ /* XXX: The initial OI scrub will scan the top level /O to generate
+ * a small local FLDB according to the <seq>. If the given FID
+ * is in the local FLDB, then it is FID-on-OST; otherwise it's
+ * quite possible for FID-on-MDT. */
+ return 0;
+}
+
+static int osd_scrub_get_fid(struct osd_thread_info *info,
+ struct osd_device *dev, struct inode *inode,
+ struct lu_fid *fid, bool scrub)
+{
+ struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
+ int rc;
+ bool has_lma = false;
+
+ rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
+ if (rc == 0) {
+ has_lma = true;
+ if (lma->lma_compat & LMAC_NOT_IN_OI) {
+ ldiskfs_set_inode_state(inode,
+ LDISKFS_STATE_LUSTRE_NO_OI);
+ return SCRUB_NEXT_CONTINUE;
+ }
+
+ *fid = lma->lma_self_fid;
+ if (fid_is_internal(&lma->lma_self_fid)) {
+ if (!scrub)
+ rc = SCRUB_NEXT_CONTINUE;
+ return rc;
+ }
+
+ if (!scrub)
+ return 0;
+
+ if (fid_is_namespace_visible(fid) && !fid_is_norm(fid))
+ return 0;
+
+ if (lma->lma_compat & LMAC_FID_ON_OST)
+ return SCRUB_NEXT_OSTOBJ;
+
+ if (fid_is_idif(fid) || fid_is_last_id(fid))
+ return SCRUB_NEXT_OSTOBJ_OLD;
+
+ if (lma->lma_incompat & LMAI_AGENT)
+ return SCRUB_NEXT_CONTINUE;
+
+ /* Here, it may be MDT-object, or may be 2.4 OST-object.
+ * Fall through. */
+ }
+
+ if (rc == -ENODATA || rc == 0) {
+ rc = osd_get_idif(info, inode, &info->oti_obj_dentry, fid);
+ if (rc == 0) {
+ if (scrub)
+ /* It is old 2.x (x <= 3) or 1.8 OST-object. */
+ rc = SCRUB_NEXT_OSTOBJ_OLD;
+ return rc;
+ }
+
+ if (rc > 0) {
+ if (!has_lma)
+ /* It is FID-on-OST, but we do not know how
+ * to generate its FID, ignore it directly. */
+ rc = SCRUB_NEXT_CONTINUE;
+ else
+ /* It is 2.4 OST-object. */
+ rc = SCRUB_NEXT_OSTOBJ_OLD;
+ return rc;
+ }
+
+ if (rc != -ENODATA)
+ return rc;
+
+ if (!has_lma) {
+ if (dev->od_scrub.os_convert_igif) {
+ lu_igif_build(fid, inode->i_ino,
+ inode->i_generation);
+ if (scrub)
+ rc = SCRUB_NEXT_NOLMA;
+ else
+ rc = 0;
+ } else {
+ /* It may be FID-on-OST, or may be FID for
+ * non-MDT0, anyway, we do not know how to
+ * generate its FID, ignore it directly. */
+ rc = SCRUB_NEXT_CONTINUE;
+ }
+ return rc;
+ }
+
+ /* For OI scrub case only: the object has LMA but has no ff
+ * (or ff crashed). It may be MDT-object, may be OST-object
+ * with crashed ff. The last check is local FLDB. */
+ rc = osd_scrub_check_local_fldb(info, dev, fid);
+ }
+
+ return rc;
+}
+
static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev,
struct lu_fid *fid, struct osd_inode_id *lid, __u32 pos,
struct super_block *sb, bool scrub)
{
- struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
- struct inode *inode;
- int rc;
+ struct inode *inode;
+ int rc;
+ ENTRY;
osd_id_gen(lid, pos, OSD_OII_NOGEN);
inode = osd_iget(info, dev, lid);
/* The inode may be removed after bitmap searching, or the
* file is new created without inode initialized yet. */
if (rc == -ENOENT || rc == -ESTALE)
- return SCRUB_NEXT_CONTINUE;
+ RETURN(SCRUB_NEXT_CONTINUE);
CERROR("%.16s: fail to read inode, ino# = %u, rc = %d\n",
LDISKFS_SB(sb)->s_es->s_volume_name, pos, rc);
- return rc;
+ RETURN(rc);
}
/* If the inode has no OI mapping, then it is special locally used,
* should be invisible to OI scrub or up layer LFSCK. */
- if (ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NO_OI)) {
- iput(inode);
- return SCRUB_NEXT_CONTINUE;
- }
+ if (ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NO_OI))
+ GOTO(put, rc = SCRUB_NEXT_CONTINUE);
if (scrub &&
ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB)) {
/* Only skip it for the first OI scrub accessing. */
ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
- iput(inode);
- return SCRUB_NEXT_NOSCRUB;
+ GOTO(put, rc = SCRUB_NEXT_NOSCRUB);
}
- rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
- if (rc == 0) {
- if (fid_is_llog(&lma->lma_self_fid) ||
- (!scrub && fid_is_internal(&lma->lma_self_fid)))
- rc = SCRUB_NEXT_CONTINUE;
- else
- *fid = lma->lma_self_fid;
- } else if (rc == -ENODATA) {
- lu_igif_build(fid, inode->i_ino, inode->i_generation);
- if (scrub)
- rc = SCRUB_NEXT_NOLMA;
- else
- rc = 0;
- }
+ rc = osd_scrub_get_fid(info, dev, inode, fid, scrub);
+
+ GOTO(put, rc);
+
+put:
iput(inode);
return rc;
}
brelse(param.bitmap);
RETURN(rc);
}
-
- if (preload && dev->od_otable_it->ooi_stopping) {
- brelse(param.bitmap);
- RETURN(0);
- }
}
next_group:
brelse(param.bitmap);
-
- if (preload && dev->od_otable_it->ooi_stopping)
- RETURN(0);
}
if (*pos > limit)
ooc->ooc_pos_preload, rc);
if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
+ spin_lock(&scrub->os_lock);
scrub->os_waiting = 0;
cfs_waitq_broadcast(&scrub->os_thread.t_ctl_waitq);
+ spin_unlock(&scrub->os_lock);
}
RETURN(rc < 0 ? rc : ooc->ooc_cached_items);
int rc;
ENTRY;
- cfs_daemonize("OI_scrub");
- rc = lu_env_init(&env, LCT_DT_THREAD);
+ rc = lu_env_init(&env, LCT_LOCAL);
if (rc != 0) {
CERROR("%.16s: OI scrub, fail to init env, rc = %d\n",
LDISKFS_SB(sb)->s_es->s_volume_name, rc);
static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
loff_t offset, __u64 ino, unsigned d_type);
+static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
+ loff_t offset, __u64 ino, unsigned d_type);
static int
osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
OLF_SCAN_SUBITEMS = 0x0001,
OLF_HIDE_FID = 0x0002,
OLF_SHOW_NAME = 0x0004,
+ OLF_NO_OI = 0x0008,
};
struct osd_lf_map {
osd_ios_general_scan, osd_ios_varfid_fill },
/* PENDING */
- { "PENDING", { FID_SEQ_LOCAL_FILE, MDD_ORPHAN_OID, 0 }, 0, NULL, NULL },
+ { "PENDING", { 0, 0, 0 }, 0, NULL, NULL },
/* ROOT */
{ "ROOT", { FID_SEQ_ROOT, 1, 0 },
NULL, NULL },
/* lfsck_bookmark */
- { "lfsck_bookmark", { FID_SEQ_LOCAL_FILE, LFSCK_BOOKMARK_OID, 0 }, 0,
- NULL, NULL },
+ { "lfsck_bookmark", { 0, 0, 0 }, 0, NULL, NULL },
/* lov_objid */
{ LOV_OBJID, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OID, 0 }, OLF_SHOW_NAME,
{ QSD_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
osd_ios_general_scan, osd_ios_varfid_fill },
- /* seq-200000003-lastid */
- { "seq-200000003-lastid", { FID_SEQ_LOCAL_NAME, 1, 0 }, 0,
- NULL, NULL },
-
/* seq_ctl */
{ "seq_ctl", { FID_SEQ_LOCAL_FILE, FID_SEQ_CTL_OID, 0 },
OLF_SHOW_NAME, NULL, NULL },
{ "seq_srv", { FID_SEQ_LOCAL_FILE, FID_SEQ_SRV_OID, 0 },
OLF_SHOW_NAME, NULL, NULL },
- /* LAST_GROUP */
- { "LAST_GROUP", { FID_SEQ_LOCAL_FILE, OFD_LAST_GROUP_OID, 0 },
- OLF_SHOW_NAME, NULL, NULL },
-
/* health_check */
{ HEALTH_CHECK, { FID_SEQ_LOCAL_FILE, OFD_HEALTH_CHECK_OID, 0 },
OLF_SHOW_NAME, NULL, NULL },
/* lfsck_namespace */
- { "lfsck_namespace", { FID_SEQ_LOCAL_FILE, LFSCK_BOOKMARK_OID, 0 }, 0,
- NULL, NULL },
+ { "lfsck_namespace", { 0, 0, 0 }, 0, NULL, NULL },
/* OBJECTS, upgrade from old device */
{ OBJECTS, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, osd_ios_OBJECTS_scan, NULL },
/* lquota_v2.group, upgrade from old device */
{ "lquota_v2.group", { 0, 0, 0 }, 0, NULL, NULL },
+ /* LAST_GROUP, upgrade from old device */
+ { "LAST_GROUP", { FID_SEQ_LOCAL_FILE, OFD_LAST_GROUP_OID, 0 },
+ OLF_SHOW_NAME, NULL, NULL },
+
+ /* lost+found */
+ { "lost+found", { 0, 0, 0 }, OLF_SCAN_SUBITEMS | OLF_NO_OI,
+ osd_ios_general_scan, osd_ios_lf_fill },
+
{ NULL, { 0, 0, 0 }, 0, NULL, NULL }
};
return dentry;
}
-static inline void
-osd_ios_llogname2fid(struct lu_fid *fid, const char *name, int namelen)
-{
- obd_id id = 0;
- int i = 0;
-
- fid->f_seq = FID_SEQ_LLOG;
- while (i < namelen)
- id = id * 10 + name[i++] - '0';
-
- fid->f_oid = id & 0x00000000ffffffffULL;
- fid->f_ver = id >> 32;
-}
-
-static inline void
-osd_ios_Oname2fid(struct lu_fid *fid, const char *name, int namelen)
-{
- __u64 seq = 0;
- int i = 0;
-
- while (i < namelen)
- seq = seq * 10 + name[i++] - '0';
-
- lu_last_id_fid(fid, seq);
-}
-
static int
osd_ios_new_item(struct osd_device *dev, struct dentry *dentry,
scandir_t scandir, filldir_t filldir)
lu_igif_build(&tfid, inode->i_ino, inode->i_generation);
else
tfid = *fid;
- rc = osd_ea_fid_set(info, inode, &tfid, 0);
+ rc = osd_ea_fid_set(info, inode, &tfid, 0, 0);
if (rc != 0)
RETURN(rc);
} else {
+ if (lma->lma_compat & LMAC_NOT_IN_OI)
+ RETURN(0);
+
tfid = lma->lma_self_fid;
}
- rc = __osd_oi_lookup(info, dev, &tfid, id2);
+ rc = osd_oi_lookup(info, dev, &tfid, id2, 0);
if (rc != 0) {
if (rc != -ENOENT)
RETURN(rc);
rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
- DTO_INDEX_INSERT);
+ DTO_INDEX_INSERT, 0);
+ if (rc > 0)
+ rc = 0;
+
RETURN(rc);
}
RETURN(rc);
}
- rc = osd_scrub_refresh_mapping(info, dev, &tfid, id, DTO_INDEX_UPDATE);
+ rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
+ DTO_INDEX_UPDATE, 0);
+ if (rc > 0)
+ rc = 0;
RETURN(rc);
}
+/**
+ * It scans the /lost+found, and for the OST-object (with filter_fid
+ * or filter_fid_old), move them back to its proper /O/<seq>/d<x>.
+ */
+static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
+ loff_t offset, __u64 ino, unsigned d_type)
+{
+ struct osd_ios_filldir_buf *fill_buf = buf;
+ struct osd_thread_info *info = fill_buf->oifb_info;
+ struct osd_device *dev = fill_buf->oifb_dev;
+ struct lu_fid *fid = &info->oti_fid;
+ struct osd_scrub *scrub = &dev->od_scrub;
+ struct dentry *parent = fill_buf->oifb_dentry;
+ struct dentry *child;
+ struct inode *dir = parent->d_inode;
+ struct inode *inode;
+ int rc;
+ ENTRY;
+
+ /* skip any '.' started names */
+ if (name[0] == '.')
+ RETURN(0);
+
+ scrub->os_lf_scanned++;
+ child = osd_ios_lookup_one_len(name, parent, namelen);
+ if (IS_ERR(child)) {
+ CWARN("%s: cannot lookup child '%.*s': rc = %d\n",
+ osd_name(dev), namelen, name, (int)PTR_ERR(child));
+ RETURN(0);
+ }
+
+ inode = child->d_inode;
+ if (S_ISDIR(inode->i_mode)) {
+ rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
+ osd_ios_lf_fill);
+ if (rc != 0)
+ CWARN("%s: cannot add child '%.*s': rc = %d\n",
+ osd_name(dev), namelen, name, rc);
+ GOTO(put, rc);
+ }
+
+ if (!S_ISREG(inode->i_mode))
+ GOTO(put, rc = 0);
+
+ rc = osd_scrub_get_fid(info, dev, inode, fid, true);
+ if (rc == SCRUB_NEXT_OSTOBJ || rc == SCRUB_NEXT_OSTOBJ_OLD) {
+ rc = osd_obj_map_recover(info, dev, dir, child, fid);
+ if (rc == 0) {
+ CDEBUG(D_LFSCK, "recovered '%.*s' ["DFID"] from "
+ "/lost+found.\n", namelen, name, PFID(fid));
+ scrub->os_lf_repaired++;
+ } else {
+ CWARN("%s: cannot rename for '%.*s' "DFID": rc = %d\n",
+ osd_name(dev), namelen, name, PFID(fid), rc);
+ }
+ }
+
+ /* XXX: For MDT-objects, we can move them from /lost+found to namespace
+ * visible place, such as the /ROOT/.lustre/lost+found, then LFSCK
+ * can process them in furtuer. */
+
+ GOTO(put, rc);
+
+put:
+ if (rc < 0)
+ scrub->os_lf_failed++;
+ dput(child);
+ /* skip the failure to make the scanning to continue. */
+ return 0;
+}
+
static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
loff_t offset, __u64 ino, unsigned d_type)
{
if (IS_ERR(child))
RETURN(PTR_ERR(child));
- rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
- &map->olm_fid, map->olm_flags);
+ if (!(map->olm_flags & OLF_NO_OI))
+ rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
+ &map->olm_fid, map->olm_flags);
if (rc == 0 && map->olm_flags & OLF_SCAN_SUBITEMS)
rc = osd_ios_new_item(dev, child, map->olm_scandir,
map->olm_filldir);
int rc;
ENTRY;
- /* It is existing MDT device. */
- dev->od_handle_nolma = 1;
+ /* It is existing MDT0 device. We only allow the case of object without
+ * LMA to happen on the MDT0, which is usually for old 1.8 MDT. Then we
+ * can generate IGIF mode FID for the object and related OI mapping. If
+ * it is on other MDTs, then becuase file-level backup/restore, related
+ * OI mapping may be invalid already, we do not know which is the right
+ * FID for the object. We only allow IGIF objects to reside on the MDT0.
+ *
+ * XXX: For the case of object on non-MDT0 device with neither LMA nor
+ * "fid" xattr, then something crashed. We cannot re-generate the
+ * FID directly, instead, the OI scrub will scan the OI structure
+ * and try to re-generate the LMA from the OI mapping. But if the
+ * OI mapping crashed or lost also, then we have to give up under
+ * double failure cases. */
+ scrub->os_convert_igif = 1;
child = osd_ios_lookup_one_len(dot_lustre_name, dentry,
strlen(dot_lustre_name));
if (IS_ERR(child)) {
osd_scrub_file_reset(scrub,
LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
SF_UPGRADE);
+ sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
rc = osd_scrub_file_store(scrub);
} else {
rc = 0;
}
}
} else {
- rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
+ /* For lustre-2.x (x <= 3), the ".lustre" has NO FID-in-LMA,
+ * so the client will get IGIF for the ".lustre" object when
+ * the MDT restart.
+ *
+ * From the OI scrub view, when the MDT upgrade to Lustre-2.4,
+ * it does not know whether there are some old clients cached
+ * the ".lustre" IGIF during the upgrading. Two choices:
+ *
+ * 1) Generate IGIF-in-LMA and IGIF-in-OI for the ".lustre".
+ * It will allow the old connected clients to access the
+ * ".lustre" with cached IGIF. But it will cause others
+ * on the MDT failed to check "fid_is_dot_lustre()".
+ *
+ * 2) Use fixed FID {FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE, 0}
+ * for ".lustre" in spite of whether there are some clients
+ * cached the ".lustre" IGIF or not. It enables the check
+ * "fid_is_dot_lustre()" on the MDT, although it will cause
+ * that the old connected clients cannot access the ".lustre"
+ * with the cached IGIF.
+ *
+ * Usually, it is rare case for the old connected clients
+ * to access the ".lustre" with cached IGIF. So we prefer
+ * to the solution 2). */
+ rc = osd_ios_scan_one(info, dev, child->d_inode,
+ &LU_DOT_LUSTRE_FID, 0);
dput(child);
}
osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
struct dentry *dentry, filldir_t filldir)
{
- struct dentry *child;
- int rc;
+ struct osd_scrub *scrub = &dev->od_scrub;
+ struct scrub_file *sf = &scrub->os_file;
+ struct dentry *child;
+ int rc;
ENTRY;
+ if (unlikely(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID)) {
+ sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
+ rc = osd_scrub_file_store(scrub);
+ if (rc != 0)
+ RETURN(rc);
+ }
+
child = osd_ios_lookup_one_len(ADMIN_USR, dentry, strlen(ADMIN_USR));
if (!IS_ERR(child)) {
rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
static int osd_initial_OI_scrub(struct osd_thread_info *info,
struct osd_device *dev)
{
- struct osd_ios_item *item = NULL;
- scandir_t scandir = osd_ios_general_scan;
- filldir_t filldir = osd_ios_root_fill;
- struct dentry *dentry = osd_sb(dev)->s_root;
- int rc;
+ struct osd_ios_item *item = NULL;
+ scandir_t scandir = osd_ios_general_scan;
+ filldir_t filldir = osd_ios_root_fill;
+ struct dentry *dentry = osd_sb(dev)->s_root;
+ const struct osd_lf_map *map = osd_lf_maps;
+ int rc;
ENTRY;
while (1) {
OBD_FREE_PTR(item);
}
- RETURN(rc);
+ if (rc != 0)
+ RETURN(rc);
+
+ /* There maybe the case that the object has been removed, but its OI
+ * mapping is still in the OI file, such as the "CATALOGS" after MDT
+ * file-level backup/restore. So here cleanup the stale OI mappings. */
+ while (map->olm_name != NULL) {
+ struct dentry *child;
+
+ if (fid_is_zero(&map->olm_fid)) {
+ map++;
+ continue;
+ }
+
+ child = osd_ios_lookup_one_len(map->olm_name,
+ osd_sb(dev)->s_root,
+ strlen(map->olm_name));
+ if (!IS_ERR(child))
+ dput(child);
+ else if (PTR_ERR(child) == -ENOENT)
+ osd_scrub_refresh_mapping(info, dev, &map->olm_fid,
+ NULL, DTO_INDEX_DELETE, 0);
+ map++;
+ }
+
+ RETURN(0);
}
char *osd_lf_fid2name(const struct lu_fid *fid)
scrub->os_start_flags = flags;
thread_set_flags(thread, 0);
- rc = cfs_create_thread(osd_scrub_main, dev, 0);
- if (rc < 0) {
+ rc = PTR_ERR(kthread_run(osd_scrub_main, dev, "OI_scrub"));
+ if (IS_ERR_VALUE(rc)) {
CERROR("%.16s: cannot start iteration thread, rc = %d\n",
LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, rc);
RETURN(rc);
struct ldiskfs_super_block *es = LDISKFS_SB(sb)->s_es;
struct lvfs_run_ctxt saved;
struct file *filp;
+ struct inode *inode;
+ struct lu_fid *fid = &info->oti_fid;
int dirty = 0;
int rc = 0;
ENTRY;
push_ctxt(&saved, ctxt, NULL);
filp = filp_open(osd_scrub_name, O_RDWR | O_CREAT, 0644);
- if (IS_ERR(filp))
+ if (IS_ERR(filp)) {
+ pop_ctxt(&saved, ctxt, NULL);
RETURN(PTR_ERR(filp));
+ }
- scrub->os_inode = igrab(filp->f_dentry->d_inode);
+ inode = filp->f_dentry->d_inode;
+ ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NO_OI);
+ /* 'What the @fid is' is not imporatant, because the object
+ * has no OI mapping, and only is visible inside the OSD.*/
+ lu_igif_build(fid, inode->i_ino, inode->i_generation);
+ rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
+ if (rc != 0) {
+ filp_close(filp, 0);
+ pop_ctxt(&saved, ctxt, NULL);
+ RETURN(rc);
+ }
+
+ scrub->os_inode = igrab(inode);
filp_close(filp, 0);
pop_ctxt(&saved, ctxt, NULL);
- ldiskfs_set_inode_state(scrub->os_inode,
- LDISKFS_STATE_LUSTRE_NO_OI);
rc = osd_scrub_file_load(scrub);
if (rc == -ENOENT) {
osd_scrub_file_init(scrub, es->s_uuid);
+ /* If the "/O" dir does not exist when mount (indicated by
+ * osd_device::od_maybe_new), neither for the "/OI_scrub",
+ * then it is quite probably that the device is a new one,
+ * under such case, mark it as SIF_NO_HANDLE_OLD_FID.
+ *
+ * For the rare case that "/O" and "OI_scrub" both lost on
+ * an old device, it can be found and cleared later.
+ *
+ * For the system with "SIF_NO_HANDLE_OLD_FID", we do not
+ * need to check "filter_fid_old" and to convert it to
+ * "filter_fid" for each object, and all the IGIF should
+ * have their FID mapping in OI files already. */
+ if (dev->od_maybe_new)
+ sf->sf_internal_flags = SIF_NO_HANDLE_OLD_FID;
dirty = 1;
} else if (rc != 0) {
RETURN(rc);
RETURN(rc);
rc = osd_initial_OI_scrub(info, dev);
- if (rc == 0 && !dev->od_noscrub &&
- ((sf->sf_status == SS_PAUSED) ||
- (sf->sf_status == SS_CRASHED &&
- sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE |
- SF_AUTO)) ||
- (sf->sf_status == SS_INIT &&
- sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE))))
- rc = osd_scrub_start(dev);
+ if (rc == 0) {
+ if (sf->sf_flags & SF_UPGRADE ||
+ !(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID ||
+ sf->sf_success_count > 0)) {
+ dev->od_igif_inoi = 0;
+ dev->od_check_ff = 1;
+ } else {
+ dev->od_igif_inoi = 1;
+ dev->od_check_ff = 0;
+ }
+
+ if (sf->sf_flags & SF_INCONSISTENT)
+ /* The 'od_igif_inoi' will be set under the
+ * following cases:
+ * 1) new created system, or
+ * 2) restored from file-level backup, or
+ * 3) the upgrading completed.
+ *
+ * The 'od_igif_inoi' may be cleared by OI scrub
+ * later if found that the system is upgrading. */
+ dev->od_igif_inoi = 1;
+
+ if (!dev->od_noscrub &&
+ ((sf->sf_status == SS_PAUSED) ||
+ (sf->sf_status == SS_CRASHED &&
+ sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
+ SF_UPGRADE | SF_AUTO)) ||
+ (sf->sf_status == SS_INIT &&
+ sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
+ SF_UPGRADE))))
+ rc = osd_scrub_start(dev);
+ }
+
+ /* it is possible that dcache entries may keep objects after they are
+ * deleted by OSD. While it looks safe this can cause object data to
+ * stay until umount causing failures in tests calculating free space,
+ * e.g. replay-ost-single. Since those dcache entries are not used
+ * anymore let's just free them after use here */
+ shrink_dcache_sb(sb);
RETURN(rc);
}
dev->od_otable_it = it;
it->ooi_dev = dev;
- it->ooi_pid = cfs_curproc_pid();
it->ooi_cache.ooc_consumer_idx = -1;
if (flags & DOIF_OUTUSED)
it->ooi_used_outside = 1;
return 0;
}
-/**
- * It is hack here:
- *
- * Sometimes the otable-based iteration driver (LFSCK) may be blocked in OSD
- * layer when someone wants to stop/pause the iteration. Under such case, we
- * need some mechanism to notify the event and wakeup the blocker.
- */
static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di)
{
- struct osd_otable_it *it = (struct osd_otable_it *)di;
- struct osd_device *dev = it->ooi_dev;
-
- /* od_otable_mutex: prevent curcurrent init/fini */
- mutex_lock(&dev->od_otable_mutex);
- if (it->ooi_pid == cfs_curproc_pid()) {
- dev->od_scrub.os_paused = 1;
- } else {
- struct ptlrpc_thread *thread = &dev->od_scrub.os_thread;
-
- it->ooi_stopping = 1;
- if (it->ooi_waiting)
- cfs_waitq_broadcast(&thread->t_ctl_waitq);
- }
- mutex_unlock(&dev->od_otable_mutex);
}
static inline int
{
spin_lock(&scrub->os_lock);
if (it->ooi_cache.ooc_pos_preload < scrub->os_pos_current ||
- scrub->os_waiting || it->ooi_stopping ||
+ scrub->os_waiting ||
!thread_is_running(&scrub->os_thread))
it->ooi_waiting = 0;
else
if (!thread_is_running(thread) && !it->ooi_used_outside)
RETURN(1);
- if (it->ooi_stopping)
- RETURN(0);
-
rc = osd_otable_it_preload(env, it);
if (rc >= 0)
goto again;
struct osd_otable_cache *ooc = &it->ooi_cache;
__u64 hash;
- if (!it->ooi_user_ready)
- hash = ooc->ooc_pos_preload;
- else if (likely(ooc->ooc_consumer_idx != -1))
+ if (it->ooi_user_ready && ooc->ooc_consumer_idx != -1)
hash = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino;
else
- hash = 0;
+ hash = ooc->ooc_pos_preload;
return hash;
}
if (thread_is_running(&scrub->os_thread)) {
cfs_duration_t duration = cfs_time_current() -
scrub->os_time_last_checkpoint;
- __u64 new_checked = scrub->os_new_checked * CFS_HZ;
+ __u64 new_checked = scrub->os_new_checked * HZ;
__u32 rtime = sf->sf_run_time +
cfs_duration_sec(duration + HALF_SEC);
"run_time: %u seconds\n"
"average_speed: "LPU64" objects/sec\n"
"real-time_speed: "LPU64" objects/sec\n"
- "current_position: %u\n",
- rtime, speed, new_checked, scrub->os_pos_current);
+ "current_position: %u\n"
+ "lf_scanned: "LPU64"\n"
+ "lf_reparied: "LPU64"\n"
+ "lf_failed: "LPU64"\n",
+ rtime, speed, new_checked, scrub->os_pos_current,
+ scrub->os_lf_scanned, scrub->os_lf_repaired,
+ scrub->os_lf_failed);
} else {
if (sf->sf_run_time != 0)
do_div(speed, sf->sf_run_time);
"run_time: %u seconds\n"
"average_speed: "LPU64" objects/sec\n"
"real-time_speed: N/A\n"
- "current_position: N/A\n",
- sf->sf_run_time, speed);
+ "current_position: N/A\n"
+ "lf_scanned: "LPU64"\n"
+ "lf_reparied: "LPU64"\n"
+ "lf_failed: "LPU64"\n",
+ sf->sf_run_time, speed, scrub->os_lf_scanned,
+ scrub->os_lf_repaired, scrub->os_lf_failed);
}
if (rc <= 0)
goto out;
git://git.whamcloud.com / fs / lustre-release.git / blobdiff