X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fosd-ldiskfs%2Fosd_handler.c;h=d6c2f86127cbfb185e9c03129ee20f04f226364f;hp=41b79fd9b3f849859c2e0971a455c8e8c9a7de43;hb=eebc3da214dfcbc01ba637f0925bfe8635b26138;hpb=bad49e39e301d4367eaead5ee566f5dcacfde8f6
diff --git a/lustre/osd-ldiskfs/osd_handler.c b/lustre/osd-ldiskfs/osd_handler.c
index 41b79fd..d6c2f86 100644
--- a/lustre/osd-ldiskfs/osd_handler.c
+++ b/lustre/osd-ldiskfs/osd_handler.c
@@ -122,6 +122,11 @@ static const struct dt_object_operations osd_obj_otable_it_ops;
static const struct dt_index_operations osd_index_iam_ops;
static const struct dt_index_operations osd_index_ea_ops;
+static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
+ const struct lu_fid *fid);
+static int osd_process_scheduled_agent_removals(const struct lu_env *env,
+ struct osd_device *osd);
+
int osd_trans_declare_op2rb[] = {
[OSD_OT_ATTR_SET] = OSD_OT_ATTR_SET,
[OSD_OT_PUNCH] = OSD_OT_MAX,
@@ -237,6 +242,11 @@ struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
{
struct inode *inode = NULL;
+ /* if we look for an inode withing a running
+ * transaction, then we risk to deadlock */
+ /* osd_dirent_check_repair() breaks this */
+ /*LASSERT(current->journal_info == NULL);*/
+
inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
if (IS_ERR(inode)) {
CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
@@ -301,6 +311,130 @@ osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
return inode;
}
+static struct inode *osd_iget_check(struct osd_thread_info *info,
+ struct osd_device *dev,
+ const struct lu_fid *fid,
+ struct osd_inode_id *id,
+ bool cached)
+{
+ struct inode *inode;
+ int rc = 0;
+ ENTRY;
+
+ /* The cached OI mapping is trustable. If we cannot locate the inode
+ * via the cached OI mapping, then return the failure to the caller
+ * directly without further OI checking. */
+
+ inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
+ if (IS_ERR(inode)) {
+ rc = PTR_ERR(inode);
+ if (cached || (rc != -ENOENT && rc != -ESTALE)) {
+ CDEBUG(D_INODE, "no inode: ino = %u, rc = %d\n",
+ id->oii_ino, rc);
+
+ GOTO(put, rc);
+ }
+
+ goto check_oi;
+ }
+
+ if (is_bad_inode(inode)) {
+ rc = -ENOENT;
+ if (cached) {
+ CDEBUG(D_INODE, "bad inode: ino = %u\n", id->oii_ino);
+
+ GOTO(put, rc);
+ }
+
+ goto check_oi;
+ }
+
+ if (id->oii_gen != OSD_OII_NOGEN &&
+ inode->i_generation != id->oii_gen) {
+ rc = -ESTALE;
+ if (cached) {
+ CDEBUG(D_INODE, "unmatched inode: ino = %u, "
+ "oii_gen = %u, i_generation = %u\n",
+ id->oii_ino, id->oii_gen, inode->i_generation);
+
+ GOTO(put, rc);
+ }
+
+ goto check_oi;
+ }
+
+ if (inode->i_nlink == 0) {
+ rc = -ENOENT;
+ if (cached) {
+ CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
+
+ GOTO(put, rc);
+ }
+
+ goto check_oi;
+ }
+
+check_oi:
+ if (rc != 0) {
+ LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
+
+ rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
+ /* XXX: There are four possible cases:
+ * 1. rc = 0.
+ * Backup/restore caused the OI invalid.
+ * 2. rc = 0.
+ * Someone unlinked the object but NOT removed
+ * the OI mapping, such as mount target device
+ * as ldiskfs, and modify something directly.
+ * 3. rc = -ENOENT.
+ * Someone just removed the object between the
+ * former oi_lookup and the iget. It is normal.
+ * 4. Other failure cases.
+ *
+ * Generally, when the device is mounted, it will
+ * auto check whether the system is restored from
+ * file-level backup or not. We trust such detect
+ * to distinguish the 1st case from the 2nd case. */
+ if (rc == 0) {
+ if (!IS_ERR(inode) && inode->i_generation != 0 &&
+ inode->i_generation == id->oii_gen)
+ /* "id->oii_gen != OSD_OII_NOGEN" is for
+ * "@cached == false" case. */
+ rc = -ENOENT;
+ else
+ rc = -EREMCHG;
+ } else {
+ /* If the OI mapping was in OI file before the
+ * osd_iget_check(), but now, it is disappear,
+ * then it must be removed by race. That is a
+ * normal race case. */
+ }
+ } else {
+ if (id->oii_gen == OSD_OII_NOGEN)
+ osd_id_gen(id, inode->i_ino, inode->i_generation);
+
+ /* Do not update file c/mtime in ldiskfs.
+ * NB: we don't have any lock to protect this because we don't
+ * have reference on osd_object now, but contention with
+ * another lookup + attr_set can't happen in the tiny window
+ * between if (...) and set S_NOCMTIME. */
+ if (!(inode->i_flags & S_NOCMTIME))
+ inode->i_flags |= S_NOCMTIME;
+ }
+
+ GOTO(put, rc);
+
+put:
+ if (rc != 0) {
+ if (!IS_ERR(inode))
+ iput(inode);
+
+ inode = ERR_PTR(rc);
+ }
+
+ return inode;
+}
+
/**
* \retval +v: new filter_fid, does not contain self-fid
* \retval 0: filter_fid_old, contains self-fid
@@ -385,9 +519,6 @@ static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
}
}
- if (unlikely(rc == -ENODATA))
- RETURN(0);
-
if (rc < 0)
RETURN(rc);
@@ -450,8 +581,7 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
struct scrub_file *sf;
int result;
int saved = 0;
- bool in_oi = false;
- bool in_cache = false;
+ bool cached = true;
bool triggered = false;
ENTRY;
@@ -481,7 +611,6 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
if (lu_fid_eq(fid, &oic->oic_fid) &&
likely(oic->oic_dev == dev)) {
id = &oic->oic_lid;
- in_cache = true;
goto iget;
}
@@ -493,10 +622,12 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
goto iget;
}
+ cached = false;
/* Search order: 3. OI files. */
result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
if (result == -ENOENT) {
if (!(fid_is_norm(fid) || fid_is_igif(fid)) ||
+ fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
!ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
sf->sf_oi_bitmap))
GOTO(out, result = 0);
@@ -507,81 +638,64 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
if (result != 0)
GOTO(out, result);
- in_oi = true;
-
iget:
- inode = osd_iget(info, dev, id);
+ inode = osd_iget_check(info, dev, fid, id, cached);
if (IS_ERR(inode)) {
result = PTR_ERR(inode);
- if (result != -ENOENT && result != -ESTALE)
- GOTO(out, result);
-
- if (in_cache)
- fid_zero(&oic->oic_fid);
-
- result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
- if (result != 0)
- GOTO(out, result = (result == -ENOENT ? 0 : result));
+ if (result == -ENOENT || result == -ESTALE)
+ GOTO(out, result = -ENOENT);
- /* The OI mapping is there, but the inode is NOT there.
- * Two possible cases for that:
- *
- * 1) Backup/restore caused the OI invalid.
- * 2) Someone unlinked the object but NOT removed
- * the OI mapping, such as mount target device
- * as ldiskfs, and modify something directly.
- *
- * Generally, when the device is mounted, it will
- * auto check whether the system is restored from
- * file-level backup or not. We trust such detect
- * to distinguish the 1st case from the 2nd case. */
- if (!(scrub->os_file.sf_flags & SF_INCONSISTENT))
- GOTO(out, result = 0);
+ if (result == -EREMCHG) {
trigger:
- if (unlikely(triggered))
- GOTO(out, result = saved);
-
- triggered = true;
- if (thread_is_running(&scrub->os_thread)) {
- result = -EINPROGRESS;
- } else if (!dev->od_noscrub) {
- /* Since we do not know the right OI mapping, we have
- * to trigger OI scrub to scan the whole device. */
- result = osd_scrub_start(dev, SS_AUTO_FULL |
- SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
- CDEBUG(D_LFSCK | D_CONSOLE, "%.16s: trigger OI "
- "scrub by RPC for "DFID", rc = %d [1]\n",
- osd_name(dev), PFID(fid), result);
- if (result == 0 || result == -EALREADY)
+ if (unlikely(triggered))
+ GOTO(out, result = saved);
+
+ triggered = true;
+ if (thread_is_running(&scrub->os_thread)) {
result = -EINPROGRESS;
- else
- result = -EREMCHG;
- }
+ } else if (!dev->od_noscrub) {
+ result = osd_scrub_start(dev, SS_AUTO_FULL |
+ SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
+ LCONSOLE_WARN("%.16s: trigger OI scrub by RPC "
+ "for "DFID", rc = %d [1]\n",
+ osd_name(dev), PFID(fid), result);
+ if (result == 0 || result == -EALREADY)
+ result = -EINPROGRESS;
+ else
+ result = -EREMCHG;
+ }
- /* We still have chance to get the valid inode: for the
- * object which is referenced by remote name entry, the
- * object on the local MDT will be linked under the dir
- * of "/REMOTE_PARENT_DIR" with its FID string as name.
- *
- * We do not know whether the object for the given FID
- * is referenced by some remote name entry or not, and
- * especially for DNE II, a multiple-linked object may
- * have many name entries reside on many MDTs.
- *
- * To simplify the operation, OSD will not distinguish
- * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
- * only happened for the RPC from other MDT during the
- * OI scrub, or for the client side RPC with FID only,
- * such as FID to path, or from old connected client. */
- saved = result;
- result = osd_lookup_in_remote_parent(info, dev, fid, id);
- if (result == 0) {
- in_oi = false;
- goto iget;
+ if (fid_is_on_ost(info, dev, fid, OI_CHECK_FLD))
+ GOTO(out, result);
+
+ /* We still have chance to get the valid inode: for the
+ * object which is referenced by remote name entry, the
+ * object on the local MDT will be linked under the dir
+ * of "/REMOTE_PARENT_DIR" with its FID string as name.
+ *
+ * We do not know whether the object for the given FID
+ * is referenced by some remote name entry or not, and
+ * especially for DNE II, a multiple-linked object may
+ * have many name entries reside on many MDTs.
+ *
+ * To simplify the operation, OSD will not distinguish
+ * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
+ * only happened for the RPC from other MDT during the
+ * OI scrub, or for the client side RPC with FID only,
+ * such as FID to path, or from old connected client. */
+ saved = result;
+ result = osd_lookup_in_remote_parent(info, dev,
+ fid, id);
+ if (result == 0) {
+ cached = true;
+ goto iget;
+ }
+
+ result = saved;
}
- GOTO(out, result = saved);
+ GOTO(out, result);
}
obj->oo_inode = inode;
@@ -589,34 +703,66 @@ trigger:
result = osd_check_lma(env, obj);
if (result != 0) {
+ if (result == -ENODATA) {
+ if (cached) {
+ result = osd_oi_lookup(info, dev, fid, id,
+ OI_CHECK_FLD);
+ if (result != 0) {
+ /* result == -ENOENT means that the OI
+ * mapping has been removed by race,
+ * the target inode belongs to other
+ * object.
+ *
+ * Others error also can be returned
+ * directly. */
+ iput(inode);
+ obj->oo_inode = NULL;
+ GOTO(out, result);
+ } else {
+ /* result == 0 means the cached OI
+ * mapping is still in the OI file,
+ * the target the inode is valid. */
+ }
+ } else {
+ /* The current OI mapping is from the OI file,
+ * since the inode has been found via
+ * osd_iget_check(), no need recheck OI. */
+ }
+
+ goto found;
+ }
+
iput(inode);
obj->oo_inode = NULL;
-
if (result != -EREMCHG)
GOTO(out, result);
- if (in_cache)
- fid_zero(&oic->oic_fid);
-
- result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
- if (result == 0)
- goto trigger;
-
- if (result != -ENOENT)
- GOTO(out, result);
+ if (cached) {
+ result = osd_oi_lookup(info, dev, fid, id,
+ OI_CHECK_FLD);
+ /* result == -ENOENT means the cached OI mapping
+ * has been removed from the OI file by race,
+ * above target inode belongs to other object.
+ *
+ * Others error also can be returned directly. */
+ if (result != 0)
+ GOTO(out, result);
- if (!in_oi && (fid_is_norm(fid) || fid_is_igif(fid)) &&
- ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
- sf->sf_oi_bitmap))
- goto trigger;
+ /* result == 0, goto trigger */
+ } else {
+ /* The current OI mapping is from the OI file,
+ * since the inode has been found via
+ * osd_iget_check(), no need recheck OI. */
+ }
- GOTO(out, result = 0);
+ goto trigger;
}
+found:
obj->oo_compat_dot_created = 1;
obj->oo_compat_dotdot_created = 1;
- if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
+ if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
GOTO(out, result = 0);
LASSERT(obj->oo_hl_head == NULL);
@@ -629,6 +775,9 @@ trigger:
GOTO(out, result = 0);
out:
+ if (result != 0 && cached)
+ fid_zero(&oic->oic_fid);
+
LINVRNT(osd_invariant(obj));
return result;
}
@@ -1052,7 +1201,7 @@ static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
struct thandle *th)
{
- int rc = 0;
+ int rc = 0, remove_agents = 0;
struct osd_thandle *oh;
struct osd_thread_info *oti = osd_oti_get(env);
struct osd_iobuf *iobuf = &oti->oti_iobuf;
@@ -1063,6 +1212,8 @@ static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
oh = container_of0(th, struct osd_thandle, ot_super);
+ remove_agents = oh->ot_remove_agents;
+
qtrans = oh->ot_quota_trans;
oh->ot_quota_trans = NULL;
@@ -1117,6 +1268,9 @@ static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
if (!rc)
rc = iobuf->dr_error;
+ if (unlikely(remove_agents != 0))
+ osd_process_scheduled_agent_removals(env, osd);
+
RETURN(rc);
}
@@ -2283,9 +2437,11 @@ static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
RETURN(-EPERM);
result = __osd_object_create(info, obj, attr, hint, dof, th);
- if (result == 0)
+ if (result == 0) {
result = __osd_oi_insert(env, obj, fid, th);
-
+ if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
+ obj->oo_dt.do_body_ops = &osd_body_ops;
+ }
LASSERT(ergo(result == 0,
dt_object_exists(dt) && !dt_object_remote(dt)));
@@ -2556,33 +2712,86 @@ static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
}
/**
- * Delete local agent inode for remote entry
+ * when direntry is deleted, we have to take care of possible agent inode
+ * referenced by that. unfortunately we can't do this at that point:
+ * iget() within a running transaction leads to deadlock and we better do
+ * not call that every delete declaration to save performance. so we put
+ * a potention agent inode on a list and process that once the transaction
+ * is over. Notice it's not any worse in terms of real orphans as regular
+ * object destroy doesn't put inodes on the on-disk orphan list. this should
+ * be addressed separately
*/
-static int osd_delete_local_agent_inode(const struct lu_env *env,
- struct osd_device *osd,
- const struct lu_fid *fid,
- __u32 ino, struct osd_thandle *oh)
+static int osd_schedule_agent_inode_removal(const struct lu_env *env,
+ struct osd_thandle *oh,
+ __u32 ino)
{
- struct osd_thread_info *oti = osd_oti_get(env);
- struct osd_inode_id *id = &oti->oti_id;
- struct inode *inode;
- ENTRY;
+ struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev);
+ struct osd_obj_orphan *oor;
- id->oii_ino = le32_to_cpu(ino);
- id->oii_gen = OSD_OII_NOGEN;
- inode = osd_iget(oti, osd, id);
- if (IS_ERR(inode)) {
- CERROR("%s: iget error "DFID" id %u:%u\n", osd_name(osd),
- PFID(fid), id->oii_ino, id->oii_gen);
- RETURN(PTR_ERR(inode));
+ OBD_ALLOC_PTR(oor);
+ if (oor == NULL)
+ return -ENOMEM;
+
+ oor->oor_ino = ino;
+ oor->oor_env = (struct lu_env *)env;
+ spin_lock(&osd->od_osfs_lock);
+ list_add(&oor->oor_list, &osd->od_orphan_list);
+ spin_unlock(&osd->od_osfs_lock);
+
+ oh->ot_remove_agents = 1;
+
+ return 0;
+
+}
+
+static int osd_process_scheduled_agent_removals(const struct lu_env *env,
+ struct osd_device *osd)
+{
+ struct osd_thread_info *info = osd_oti_get(env);
+ struct osd_obj_orphan *oor, *tmp;
+ struct osd_inode_id id;
+ struct list_head list;
+ struct inode *inode;
+ struct lu_fid fid;
+ handle_t *jh;
+ __u32 ino;
+
+ INIT_LIST_HEAD(&list);
+
+ spin_lock(&osd->od_osfs_lock);
+ list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
+ if (oor->oor_env == env) {
+ list_del(&oor->oor_list);
+ list_add(&oor->oor_list, &list);
+ }
}
+ spin_unlock(&osd->od_osfs_lock);
- clear_nlink(inode);
- mark_inode_dirty(inode);
- CDEBUG(D_INODE, "%s: delete remote inode "DFID" %lu\n",
- osd_name(osd), PFID(fid), inode->i_ino);
- iput(inode);
- RETURN(0);
+ list_for_each_entry_safe(oor, tmp, &list, oor_list) {
+
+ ino = oor->oor_ino;
+
+ list_del(&oor->oor_list);
+ OBD_FREE_PTR(oor);
+
+ osd_id_gen(&id, ino, OSD_OII_NOGEN);
+ inode = osd_iget_fid(info, osd, &id, &fid);
+ if (IS_ERR(inode))
+ continue;
+
+ if (!osd_remote_fid(env, osd, &fid)) {
+ iput(inode);
+ continue;
+ }
+
+ jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
+ clear_nlink(inode);
+ mark_inode_dirty(inode);
+ ldiskfs_journal_stop(jh);
+ iput(inode);
+ }
+
+ return 0;
}
/**
@@ -2636,6 +2845,8 @@ static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
fid, OI_CHECK_FLD) ?
LMAC_FID_ON_OST : 0, 0);
}
+ if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
+ obj->oo_dt.do_body_ops = &osd_body_ops;
}
if (result == 0)
@@ -3283,8 +3494,9 @@ static int osd_index_declare_iam_delete(const struct lu_env *env,
oh = container_of0(handle, struct osd_thandle, ot_super);
LASSERT(oh->ot_handle == NULL);
+ /* Recycle may cause additional three blocks to be changed. */
osd_trans_declare_op(env, oh, OSD_OT_DELETE,
- osd_dto_credits_noquota[DTO_INDEX_DELETE]);
+ osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
return 0;
}
@@ -3462,8 +3674,6 @@ static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
if (bh) {
- __u32 ino = 0;
-
/* If this is not the ".." entry, it might be a remote DNE
* entry and we need to check if the FID is for a remote
* MDT. If the FID is not in the directory entry (e.g.
@@ -3481,56 +3691,19 @@ static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
if (strcmp((char *)key, dotdot) != 0) {
LASSERT(de != NULL);
rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
- /* If Fid is not in dentry, try to get it from LMA */
if (rc == -ENODATA) {
- struct osd_inode_id *id;
- struct inode *inode;
-
- /* Before trying to get fid from the inode,
- * check whether the inode is valid.
- *
- * If the inode has been deleted, do not go
- * ahead to do osd_ea_fid_get, which will set
- * the inode to bad inode, which might cause
- * the inode to be deleted uncorrectly */
- inode = ldiskfs_iget(osd_sb(osd),
- le32_to_cpu(de->inode));
- if (IS_ERR(inode)) {
- CDEBUG(D_INODE, "%s: "DFID"get inode"
- "error.\n", osd_name(osd),
- PFID(fid));
- rc = PTR_ERR(inode);
- } else {
- if (likely(inode->i_nlink != 0)) {
- id = &osd_oti_get(env)->oti_id;
- rc = osd_ea_fid_get(env, obj,
- le32_to_cpu(de->inode),
- fid, id);
- } else {
- CDEBUG(D_INFO, "%s: %u "DFID
- "deleted.\n",
- osd_name(osd),
- le32_to_cpu(de->inode),
- PFID(fid));
- rc = -ESTALE;
- }
- iput(inode);
- }
+ /* can't get FID, postpone to the end of the
+ * transaction when iget() is safe */
+ osd_schedule_agent_inode_removal(env, oh,
+ le32_to_cpu(de->inode));
+ } else if (rc == 0 &&
+ unlikely(osd_remote_fid(env, osd, fid))) {
+ osd_schedule_agent_inode_removal(env, oh,
+ le32_to_cpu(de->inode));
}
- if (rc == 0 &&
- unlikely(osd_remote_fid(env, osd, fid)))
- /* Need to delete agent inode */
- ino = le32_to_cpu(de->inode);
}
rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
brelse(bh);
- if (rc == 0 && unlikely(ino != 0)) {
- rc = osd_delete_local_agent_inode(env, osd, fid, ino,
- oh);
- if (rc != 0)
- CERROR("%s: del local inode "DFID": rc = %d\n",
- osd_name(osd), PFID(fid), rc);
- }
} else {
rc = -ENOENT;
}
@@ -4959,62 +5132,44 @@ static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
return it->oie_dirent->oied_namelen;
}
-static int
-osd_dirent_update(handle_t *jh, struct super_block *sb,
- struct osd_it_ea_dirent *ent, struct lu_fid *fid,
- struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de)
+static inline bool
+osd_dot_dotdot_has_space(struct ldiskfs_dir_entry_2 *de, int dot_dotdot)
{
- struct osd_fid_pack *rec;
- int rc;
- ENTRY;
-
- LASSERT(de->file_type & LDISKFS_DIRENT_LUFID);
- LASSERT(de->rec_len >= de->name_len + sizeof(struct osd_fid_pack));
-
- rc = ldiskfs_journal_get_write_access(jh, bh);
- if (rc != 0)
- RETURN(rc);
+ LASSERTF(dot_dotdot == 1 || dot_dotdot == 2,
+ "dot_dotdot = %d\n", dot_dotdot);
- rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
- fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
- rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
+ if (LDISKFS_DIR_REC_LEN(de) >=
+ __LDISKFS_DIR_REC_LEN(dot_dotdot + 1 + sizeof(struct osd_fid_pack)))
+ return true;
- RETURN(rc);
+ return false;
}
-static inline int
-osd_dirent_has_space(__u16 reclen, __u16 namelen, unsigned blocksize)
+static inline bool
+osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
+ unsigned blocksize, int dot_dotdot)
{
- if (ldiskfs_rec_len_from_disk(reclen, blocksize) >=
- __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
- return 1;
- else
- return 0;
-}
+ if (dot_dotdot > 0)
+ return osd_dot_dotdot_has_space(de, dot_dotdot);
-static inline int
-osd_dot_dotdot_has_space(struct ldiskfs_dir_entry_2 *de, int dot_dotdot)
-{
- LASSERTF(dot_dotdot == 1 || dot_dotdot == 2,
- "dot_dotdot = %d\n", dot_dotdot);
+ if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
+ __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
+ return true;
- if (LDISKFS_DIR_REC_LEN(de) >=
- __LDISKFS_DIR_REC_LEN(dot_dotdot + 1 + sizeof(struct osd_fid_pack)))
- return 1;
- else
- return 0;
+ return false;
}
static int
osd_dirent_reinsert(const struct lu_env *env, handle_t *jh,
- struct inode *dir, struct inode *inode,
- struct osd_it_ea_dirent *ent, struct lu_fid *fid,
+ struct dentry *dentry, const struct lu_fid *fid,
struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de,
- struct htree_lock *hlock)
+ struct htree_lock *hlock, int dot_dotdot)
{
- struct dentry *dentry;
+ struct inode *dir = dentry->d_parent->d_inode;
+ struct inode *inode = dentry->d_inode;
struct osd_fid_pack *rec;
struct ldiskfs_dentry_param *ldp;
+ int namelen = dentry->d_name.len;
int rc;
ENTRY;
@@ -5023,29 +5178,28 @@ osd_dirent_reinsert(const struct lu_env *env, handle_t *jh,
RETURN(0);
/* There is enough space to hold the FID-in-dirent. */
- if (osd_dirent_has_space(de->rec_len, ent->oied_namelen,
- dir->i_sb->s_blocksize)) {
+ if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize,
+ dot_dotdot)) {
rc = ldiskfs_journal_get_write_access(jh, bh);
if (rc != 0)
RETURN(rc);
- de->name[de->name_len] = 0;
- rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
+ de->name[namelen] = 0;
+ rec = (struct osd_fid_pack *)(de->name + namelen + 1);
rec->fp_len = sizeof(struct lu_fid) + 1;
fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
de->file_type |= LDISKFS_DIRENT_LUFID;
-
rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
RETURN(rc);
}
+ LASSERTF(dot_dotdot == 0, "dot_dotdot = %d\n", dot_dotdot);
+
rc = ldiskfs_delete_entry(jh, dir, de, bh);
if (rc != 0)
RETURN(rc);
- dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
- ent->oied_namelen);
ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
osd_get_ldiskfs_dirent_param(ldp, fid);
dentry->d_fsdata = (void *)ldp;
@@ -5057,8 +5211,8 @@ osd_dirent_reinsert(const struct lu_env *env, handle_t *jh,
CDEBUG(D_LFSCK, "%.16s: fail to reinsert the dirent, "
"dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
- dir->i_ino, dir->i_generation,
- ent->oied_namelen, ent->oied_name, PFID(fid), rc);
+ dir->i_ino, dir->i_generation, namelen,
+ dentry->d_name.name, PFID(fid), rc);
RETURN(rc);
}
@@ -5088,6 +5242,32 @@ osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
bool dirty = false;
ENTRY;
+ osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
+ inode = osd_iget(info, dev, id);
+ if (IS_ERR(inode)) {
+ rc = PTR_ERR(inode);
+ if (rc == -ENOENT || rc == -ESTALE) {
+ *attr |= LUDA_UNKNOWN;
+ rc = 0;
+ } else {
+ CDEBUG(D_LFSCK, "%.16s: fail to iget for dirent "
+ "check_repair, dir = %lu/%u, name = %.*s: "
+ "rc = %d\n",
+ devname, dir->i_ino, dir->i_generation,
+ ent->oied_namelen, ent->oied_name, rc);
+ }
+
+ RETURN(rc);
+ }
+
+ dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
+ ent->oied_namelen);
+ rc = osd_get_lma(info, inode, dentry, lma);
+ if (rc == -ENODATA)
+ lma = NULL;
+ else if (rc != 0)
+ GOTO(out, rc);
+
if (ent->oied_name[0] == '.') {
if (ent->oied_namelen == 1)
dot_dotdot = 1;
@@ -5095,9 +5275,6 @@ osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
dot_dotdot = 2;
}
- dentry = osd_child_dentry_get(env, obj, ent->oied_name,
- ent->oied_namelen);
-
/* We need to ensure that the name entry is still valid.
* Because it may be removed or renamed by other already.
*
@@ -5114,8 +5291,9 @@ osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
-again:
if (dev->od_dirent_journal != 0) {
+
+again:
jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
if (IS_ERR(jh)) {
rc = PTR_ERR(jh);
@@ -5124,7 +5302,8 @@ again:
"name = %.*s: rc = %d\n",
devname, dir->i_ino, dir->i_generation, credits,
ent->oied_namelen, ent->oied_name, rc);
- RETURN(rc);
+
+ GOTO(out_inode, rc);
}
if (obj->oo_hl_head != NULL) {
@@ -5154,58 +5333,64 @@ again:
* For the whole directory, only dot/dotdot entry have no FID-in-dirent
* and needs to get FID from LMA when readdir, it will not affect the
* performance much. */
- if ((bh == NULL) || (le32_to_cpu(de->inode) != ent->oied_ino) ||
+ if ((bh == NULL) || (le32_to_cpu(de->inode) != inode->i_ino) ||
(dot_dotdot != 0 && !osd_dot_dotdot_has_space(de, dot_dotdot))) {
*attr |= LUDA_IGNORE;
- GOTO(out_journal, rc = 0);
- }
- osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
- inode = osd_iget(info, dev, id);
- if (IS_ERR(inode)) {
- rc = PTR_ERR(inode);
- if (rc == -ENOENT || rc == -ESTALE)
- rc = 1;
- else
- CDEBUG(D_LFSCK, "%.16s: fail to iget for dirent "
- "check_repair, dir = %lu/%u, name = %.*s: "
- "rc = %d\n",
- devname, dir->i_ino, dir->i_generation,
- ent->oied_namelen, ent->oied_name, rc);
-
- GOTO(out_journal, rc);
+ GOTO(out, rc = 0);
}
- rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
- if (rc == 0) {
- LASSERT(!(lma->lma_compat & LMAC_NOT_IN_OI));
+ if (lma != NULL) {
+ if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
+ struct lu_fid *tfid = &lma->lma_self_fid;
+
+ *attr |= LUDA_IGNORE;
+ /* It must be REMOTE_PARENT_DIR and as the
+ * dotdot entry of remote directory */
+ if (unlikely(dot_dotdot != 2 ||
+ fid_seq(tfid) != FID_SEQ_LOCAL_FILE ||
+ fid_oid(tfid) != REMOTE_PARENT_DIR_OID)) {
+ CDEBUG(D_LFSCK, "%.16s: expect remote agent "
+ "parent directory, but got %.*s under "
+ "dir = %lu/%u with the FID "DFID"\n",
+ devname, ent->oied_namelen,
+ ent->oied_name, dir->i_ino,
+ dir->i_generation, PFID(tfid));
+
+ GOTO(out, rc = -EIO);
+ }
+
+ GOTO(out, rc = 0);
+ }
if (fid_is_sane(fid)) {
/* FID-in-dirent is valid. */
if (lu_fid_eq(fid, &lma->lma_self_fid))
- GOTO(out_inode, rc = 0);
+ GOTO(out, rc = 0);
/* Do not repair under dryrun mode. */
if (*attr & LUDA_VERIFY_DRYRUN) {
*attr |= LUDA_REPAIR;
- GOTO(out_inode, rc = 0);
+
+ GOTO(out, rc = 0);
}
- if (dev->od_dirent_journal == 0) {
- iput(inode);
+ if (jh == NULL) {
brelse(bh);
if (hlock != NULL)
ldiskfs_htree_unlock(hlock);
else
up_read(&obj->oo_ext_idx_sem);
dev->od_dirent_journal = 1;
+
goto again;
}
*fid = lma->lma_self_fid;
dirty = true;
/* Update the FID-in-dirent. */
- rc = osd_dirent_update(jh, sb, ent, fid, bh, de);
+ rc = osd_dirent_reinsert(env, jh, dentry, fid, bh, de,
+ hlock, dot_dotdot);
if (rc == 0)
*attr |= LUDA_REPAIR;
else
@@ -5220,25 +5405,26 @@ again:
if (*attr & LUDA_VERIFY_DRYRUN) {
*fid = lma->lma_self_fid;
*attr |= LUDA_REPAIR;
- GOTO(out_inode, rc = 0);
+
+ GOTO(out, rc = 0);
}
- if (dev->od_dirent_journal == 0) {
- iput(inode);
+ if (jh == NULL) {
brelse(bh);
if (hlock != NULL)
ldiskfs_htree_unlock(hlock);
else
up_read(&obj->oo_ext_idx_sem);
dev->od_dirent_journal = 1;
+
goto again;
}
*fid = lma->lma_self_fid;
dirty = true;
/* Append the FID-in-dirent. */
- rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
- fid, bh, de, hlock);
+ rc = osd_dirent_reinsert(env, jh, dentry, fid, bh, de,
+ hlock, dot_dotdot);
if (rc == 0)
*attr |= LUDA_REPAIR;
else
@@ -5249,7 +5435,7 @@ again:
ent->oied_namelen, ent->oied_name,
PFID(fid), rc);
}
- } else if (rc == -ENODATA) {
+ } else {
/* Do not repair under dryrun mode. */
if (*attr & LUDA_VERIFY_DRYRUN) {
if (fid_is_sane(fid)) {
@@ -5259,17 +5445,18 @@ again:
inode->i_generation);
*attr |= LUDA_UPGRADE;
}
- GOTO(out_inode, rc = 0);
+
+ GOTO(out, rc = 0);
}
- if (dev->od_dirent_journal == 0) {
- iput(inode);
+ if (jh == NULL) {
brelse(bh);
if (hlock != NULL)
ldiskfs_htree_unlock(hlock);
else
up_read(&obj->oo_ext_idx_sem);
dev->od_dirent_journal = 1;
+
goto again;
}
@@ -5291,8 +5478,8 @@ again:
lu_igif_build(fid, inode->i_ino, inode->i_generation);
/* It is probably IGIF object. Only aappend the
* FID-in-dirent. OI scrub will process FID-in-LMA. */
- rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
- fid, bh, de, hlock);
+ rc = osd_dirent_reinsert(env, jh, dentry, fid, bh, de,
+ hlock, dot_dotdot);
if (rc == 0)
*attr |= LUDA_UPGRADE;
else
@@ -5305,12 +5492,9 @@ again:
}
}
- GOTO(out_inode, rc);
-
-out_inode:
- iput(inode);
+ GOTO(out, rc);
-out_journal:
+out:
brelse(bh);
if (hlock != NULL) {
ldiskfs_htree_unlock(hlock);
@@ -5320,10 +5504,15 @@ out_journal:
else
up_read(&obj->oo_ext_idx_sem);
}
+
if (jh != NULL)
ldiskfs_journal_stop(jh);
+
+out_inode:
+ iput(inode);
if (rc >= 0 && !dirty)
dev->od_dirent_journal = 0;
+
return rc;
}
@@ -5369,6 +5558,11 @@ static inline int osd_it_ea_rec(const struct lu_env *env,
rc = osd_dirent_check_repair(env, obj, it, fid, id,
&attr);
}
+
+ if (!fid_is_sane(fid)) {
+ attr &= ~LUDA_IGNORE;
+ attr |= LUDA_UNKNOWN;
+ }
} else {
attr &= ~LU_DIRENT_ATTRS_MASK;
if (!fid_is_sane(fid)) {
@@ -5406,7 +5600,7 @@ static inline int osd_it_ea_rec(const struct lu_env *env,
if (osd_remote_fid(env, dev, fid))
RETURN(0);
- if (likely(!(attr & LUDA_IGNORE) && rc == 0))
+ if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0))
osd_add_oi_cache(oti, dev, id, fid);
RETURN(rc > 0 ? 0 : rc);
@@ -5875,6 +6069,7 @@ static int osd_device_init0(const struct lu_env *env,
spin_lock_init(&o->od_osfs_lock);
mutex_init(&o->od_otable_mutex);
+ INIT_LIST_HEAD(&o->od_orphan_list);
o->od_read_cache = 1;
o->od_writethrough_cache = 1;
@@ -6201,6 +6396,7 @@ static int __init osd_mod_init(void)
{
int rc;
+ LASSERT(BH_DXLock < sizeof(((struct buffer_head *)0)->b_state) * 8);
#if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
/* please, try to keep osd_thread_info smaller than a page */
CLASSERT(sizeof(struct osd_thread_info) <= PAGE_SIZE);
@@ -6226,7 +6422,7 @@ static void __exit osd_mod_exit(void)
lu_kmem_fini(ldiskfs_caches);
}
-MODULE_AUTHOR("Sun Microsystems, Inc. ");
+MODULE_AUTHOR("OpenSFS, Inc. ");
MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
MODULE_VERSION(LUSTRE_VERSION_STRING);
MODULE_LICENSE("GPL");