static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
{
struct osd_thread_info *info = osd_oti_get(env);
+ struct osd_device *osd = osd_obj2dev(obj);
struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
struct inode *inode = obj->oo_inode;
struct dentry *dentry = &info->oti_obj_dentry;
rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
info->oti_mdt_attrs_old, LMA_OLD_SIZE);
if (rc == -ENODATA && !fid_is_igif(lu_object_fid(&obj->oo_dt.do_lu)) &&
- osd_obj2dev(obj)->od_check_ff) {
+ osd->od_check_ff) {
fid = &lma->lma_self_fid;
rc = osd_get_idif(info, inode, dentry, fid);
- if (rc > 0)
+ if ((rc > 0) || (rc == -ENODATA && osd->od_lma_self_repair)) {
+ handle_t *jh;
+
+ /* For the given OST-object, if it has neither LMA nor
+ * FID in XATTR_NAME_FID, then the given FID (which is
+ * contained in the @obj, from client RPC for locating
+ * the OST-object) is trusted. We use it to generate
+ * the LMA. */
+
+ LASSERT(current->journal_info == NULL);
+
+ jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
+ osd_dto_credits_noquota[DTO_XATTR_SET]);
+ if (IS_ERR(jh)) {
+ CWARN("%s: cannot start journal for "
+ "lma_self_repair: rc = %ld\n",
+ osd_name(osd), PTR_ERR(jh));
+ RETURN(0);
+ }
+
+ rc = osd_ea_fid_set(info, inode,
+ lu_object_fid(&obj->oo_dt.do_lu),
+ fid_is_on_ost(info, osd,
+ lu_object_fid(&obj->oo_dt.do_lu),
+ OI_CHECK_FLD) ?
+ LMAC_FID_ON_OST : 0, 0);
+ if (rc != 0)
+ CWARN("%s: cannot self repair the LMA: "
+ "rc = %d\n", osd_name(osd), rc);
+ ldiskfs_journal_stop(jh);
RETURN(0);
+ }
}
if (unlikely(rc == -ENODATA))
if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
CWARN("%s: unsupported incompat LMA feature(s) %#x for "
- "fid = "DFID", ino = %lu\n",
- osd_obj2dev(obj)->od_svname,
+ "fid = "DFID", ino = %lu\n", osd_name(osd),
lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
PFID(lu_object_fid(&obj->oo_dt.do_lu)),
inode->i_ino);
if (fid != NULL &&
unlikely(!lu_fid_eq(lu_object_fid(&obj->oo_dt.do_lu), fid))) {
CDEBUG(D_INODE, "%s: FID "DFID" != self_fid "DFID"\n",
- osd_obj2dev(obj)->od_svname,
- PFID(lu_object_fid(&obj->oo_dt.do_lu)),
+ osd_name(osd), PFID(lu_object_fid(&obj->oo_dt.do_lu)),
PFID(&lma->lma_self_fid));
rc = -EREMCHG;
}
* XXX temporary stuff. Some abstraction layer should
* be used.
*/
- jh = ldiskfs_journal_start_sb(osd_sb(dev), oh->ot_credits);
+ jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
osd_th_started(oh);
if (!IS_ERR(jh)) {
oh->ot_handle = jh;
RETURN(rc);
}
+static int osd_seq_exists(const struct lu_env *env,
+ struct osd_device *osd, obd_seq seq)
+{
+ struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
+ struct seq_server_site *ss = osd_seq_site(osd);
+ int rc;
+ ENTRY;
+
+ if (ss == NULL)
+ RETURN(1);
+
+ rc = osd_fld_lookup(env, osd, seq, range);
+ if (rc != 0) {
+ if (rc != -ENOENT)
+ CERROR("%s: can't lookup FLD sequence "LPX64
+ ": rc = %d\n", osd_name(osd), seq, rc);
+ RETURN(0);
+ }
+
+ RETURN(ss->ss_node_id == range->lsr_index);
+}
+
/*
* Concurrency: shouldn't matter.
*/
/*
* XXX should be taken from not-yet-existing fs abstraction layer.
*/
- param->ddp_mnt = osd_dt_dev(dev)->od_mnt;
param->ddp_max_name_len = LDISKFS_NAME_LEN;
param->ddp_max_nlink = LDISKFS_LINK_MAX;
param->ddp_block_shift = sb->s_blocksize_bits;
* \retval 0, on success
*/
static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
- const struct lu_fid *fid, struct thandle *th)
+ const struct lu_fid *fid, struct thandle *th)
{
- struct osd_thread_info *info = osd_oti_get(env);
- struct osd_inode_id *id = &info->oti_id;
- struct osd_device *osd = osd_obj2dev(obj);
+ struct osd_thread_info *info = osd_oti_get(env);
+ struct osd_inode_id *id = &info->oti_id;
+ struct osd_device *osd = osd_obj2dev(obj);
+ struct osd_thandle *oh;
- LASSERT(obj->oo_inode != NULL);
+ LASSERT(obj->oo_inode != NULL);
+
+ oh = container_of0(th, struct osd_thandle, ot_super);
+ LASSERT(oh->ot_handle);
osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
- return osd_oi_insert(info, osd, fid, id, th, OI_CHECK_FLD);
+ return osd_oi_insert(info, osd, fid, id, oh->ot_handle, OI_CHECK_FLD);
}
int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
obd_seq seq, struct lu_seq_range *range)
{
struct seq_server_site *ss = osd_seq_site(osd);
- int rc;
if (fid_seq_is_idif(seq)) {
fld_range_set_ost(range);
LASSERT(ss != NULL);
fld_range_set_any(range);
- rc = fld_server_lookup(env, ss->ss_server_fld, seq, range);
- if (rc != 0) {
- CERROR("%s: cannot find FLD range for "LPX64": rc = %d\n",
- osd_name(osd), seq, rc);
- }
- return rc;
+ /* OSD will only do local fld lookup */
+ return fld_local_lookup(env, ss->ss_server_fld, seq, range);
}
/*
struct dt_object_format *dof,
struct thandle *handle)
{
- struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
struct osd_thandle *oh;
int rc;
ENTRY;
if (rc != 0)
RETURN(rc);
- /* It does fld look up inside declare, and the result will be
- * added to fld cache, so the following fld lookup inside insert
- * does not need send RPC anymore, so avoid send rpc with holding
- * transaction */
- if (fid_is_norm(lu_object_fid(&dt->do_lu)) &&
- !fid_is_last_id(lu_object_fid(&dt->do_lu)))
- osd_fld_lookup(env, osd_dt_dev(handle->th_dev),
- fid_seq(lu_object_fid(&dt->do_lu)), range);
-
-
RETURN(rc);
}
RETURN(-EPERM);
if (S_ISDIR(inode->i_mode)) {
- LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1);
+ LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
+ inode->i_nlink == 2);
/* it will check/delete the inode from remote parent,
* how to optimize it? unlink performance impaction XXX */
result = osd_delete_from_remote_parent(env, osd, obj, oh);
osd_trans_exec_op(env, th, OSD_OT_DESTROY);
- result = osd_oi_delete(osd_oti_get(env), osd, fid, th, OI_CHECK_FLD);
+ result = osd_oi_delete(osd_oti_get(env), osd, fid, oh->ot_handle,
+ OI_CHECK_FLD);
/* XXX: add to ext3 orphan list */
/* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
static int osd_object_ref_add(const struct lu_env *env,
struct dt_object *dt, struct thandle *th)
{
- struct osd_object *obj = osd_dt_obj(dt);
- struct inode *inode = obj->oo_inode;
- bool need_dirty = false;
- int rc = 0;
+ struct osd_object *obj = osd_dt_obj(dt);
+ struct inode *inode = obj->oo_inode;
+ struct osd_thandle *oh;
+ int rc = 0;
LINVRNT(osd_invariant(obj));
LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
LASSERT(osd_write_locked(env, obj));
LASSERT(th != NULL);
+ oh = container_of0(th, struct osd_thandle, ot_super);
+ LASSERT(oh->ot_handle != NULL);
+
osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
- /* This based on ldiskfs_inc_count(), which is not exported.
- *
+ /*
* The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
* (65000) subdirectories by storing "1" in i_nlink if the link count
* would otherwise overflow. Directory tranversal tools understand
* in case they are being linked into the PENDING directory
*/
spin_lock(&obj->oo_guard);
- if (unlikely(!S_ISDIR(inode->i_mode) &&
- inode->i_nlink >= LDISKFS_LINK_MAX)) {
- /* MDD should have checked this, but good to be safe */
- rc = -EMLINK;
- } else if (unlikely(inode->i_nlink == 0 ||
- (S_ISDIR(inode->i_mode) &&
- inode->i_nlink >= LDISKFS_LINK_MAX))) {
- /* inc_nlink from 0 may cause WARN_ON */
- set_nlink(inode, 1);
- need_dirty = true;
- } else if (!S_ISDIR(inode->i_mode) ||
- (S_ISDIR(inode->i_mode) && inode->i_nlink >= 2)) {
- inc_nlink(inode);
- need_dirty = true;
- } /* else (S_ISDIR(inode->i_mode) && inode->i_nlink == 1) { ; } */
-
+ ldiskfs_inc_count(oh->ot_handle, inode);
LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
spin_unlock(&obj->oo_guard);
- if (need_dirty)
- ll_dirty_inode(inode, I_DIRTY_DATASYNC);
-
+ ll_dirty_inode(inode, I_DIRTY_DATASYNC);
LINVRNT(osd_invariant(obj));
return rc;
struct osd_object *obj = osd_dt_obj(dt);
struct inode *inode = obj->oo_inode;
struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
+ struct osd_thandle *oh;
LINVRNT(osd_invariant(obj));
LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
LASSERT(osd_write_locked(env, obj));
LASSERT(th != NULL);
+ oh = container_of0(th, struct osd_thandle, ot_super);
+ LASSERT(oh->ot_handle != NULL);
+
osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
spin_lock(&obj->oo_guard);
return 0;
}
- /* This based on ldiskfs_dec_count(), which is not exported.
- *
- * If a directory already has nlink == 1, then do not drop the nlink
- * count to 0, even temporarily, to avoid race conditions with other
- * threads not holding oo_guard seeing i_nlink == 0 in rare cases.
- *
- * nlink == 1 means the directory has/had > EXT4_LINK_MAX subdirs.
- */
- if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 1) {
- drop_nlink(inode);
+ ldiskfs_dec_count(oh->ot_handle, inode);
+ spin_unlock(&obj->oo_guard);
- spin_unlock(&obj->oo_guard);
- ll_dirty_inode(inode, I_DIRTY_DATASYNC);
- LINVRNT(osd_invariant(obj));
- } else {
- spin_unlock(&obj->oo_guard);
- }
+ ll_dirty_inode(inode, I_DIRTY_DATASYNC);
+ LINVRNT(osd_invariant(obj));
return 0;
}
file->f_dentry = dentry;
file->f_mapping = inode->i_mapping;
file->f_op = inode->i_fop;
+ set_file_inode(file, inode);
#ifndef HAVE_FILE_FSYNC_4ARGS
mutex_lock(&inode->i_mutex);
#endif
return rc;
}
-static int osd_mdt_seq_exists(const struct lu_env *env,
- struct osd_device *osd, obd_seq seq)
-{
- struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
- struct seq_server_site *ss = osd_seq_site(osd);
- int rc;
- ENTRY;
-
- if (ss == NULL)
- RETURN(1);
-
- /* XXX: currently, each MDT only store avaible sequence on disk, and no
- * allocated sequences information on disk, so we have to lookup FLDB,
- * but it probably makes more sense also store allocated sequence
- * locally, so we do not need do remote FLDB lookup in OSD */
- rc = osd_fld_lookup(env, osd, seq, range);
- if (rc != 0) {
- CERROR("%s: Can not lookup fld for "LPX64"\n",
- osd_name(osd), seq);
- RETURN(0);
- }
-
- RETURN(ss->ss_node_id == range->lsr_index);
-}
-
static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
struct lu_fid *fid)
{
if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
RETURN(0);
- /* Currently only check this for FID on MDT */
- if (osd_mdt_seq_exists(env, osd, fid_seq(fid)))
+ if (osd_seq_exists(env, osd, fid_seq(fid)))
RETURN(0);
RETURN(1);
down_write(&obj->oo_ext_idx_sem);
}
- bh = ldiskfs_find_entry(dir, &dentry->d_name, &de, hlock);
+ bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
if (bh) {
__u32 ino = 0;
down_read(&obj->oo_ext_idx_sem);
}
- bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
+ bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
if (bh) {
struct osd_thread_info *oti = osd_oti_get(env);
struct osd_inode_id *id = &oti->oti_id;
file->f_dentry = obj_dentry;
file->f_mapping = obj->oo_inode->i_mapping;
file->f_op = obj->oo_inode->i_fop;
+ set_file_inode(file, obj->oo_inode);
+
lu_object_get(lo);
RETURN((struct dt_it *) it);
}
again:
if (dev->od_dirent_journal) {
- jh = ldiskfs_journal_start_sb(sb, credits);
+ jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
if (IS_ERR(jh)) {
rc = PTR_ERR(jh);
CERROR("%.16s: fail to start trans for dirent "
}
}
- bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
+ bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
/* For dot/dotdot entry, if there is not enough space to hold the
* FID-in-dirent, just keep them there. It only happens when the
* device upgraded from 1.8 or restored from MDT file-level backup.
struct osd_device *o = osd_dev(d);
ENTRY;
- osd_procfs_fini(o);
osd_shutdown(env, o);
+ osd_procfs_fini(o);
osd_scrub_cleanup(env, o);
osd_obj_map_fini(o);
osd_umount(env, o);
if (rc != 0)
GOTO(out_site, rc);
+ /* self-repair LMA by default */
+ o->od_lma_self_repair = 1;
+
CFS_INIT_LIST_HEAD(&o->od_ios_list);
/* setup scrub, including OI files initialization */
rc = osd_scrub_setup(env, o);
if (rc)
return rc;
- rc = class_register_type(&osd_obd_device_ops, NULL, lvars.module_vars,
- LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
+ rc = class_register_type(&osd_obd_device_ops, NULL, NULL,
+#ifndef HAVE_ONLY_PROCFS_SEQ
+ lvars.module_vars,
+#endif
+ LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
if (rc)
lu_kmem_fini(ldiskfs_caches);
return rc;
git://git.whamcloud.com / fs / lustre-release.git / blobdiff