/* Slab to allocate dynlocks */
struct kmem_cache *dynlock_cachep;
+/* Slab to allocate osd_it_ea */
+struct kmem_cache *osd_itea_cachep;
+
static struct lu_kmem_descr ldiskfs_caches[] = {
{
.ckd_cache = &dynlock_cachep,
.ckd_size = sizeof(struct dynlock_handle)
},
{
+ .ckd_cache = &osd_itea_cachep,
+ .ckd_name = "osd_itea_cache",
+ .ckd_size = sizeof(struct osd_it_ea)
+ },
+ {
.ckd_cache = NULL
}
};
check_oi:
if (rc != 0) {
+ struct osd_inode_id saved_id = *id;
+
LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
* 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)
+ inode->i_generation == id->oii_gen) {
rc = -ENOENT;
- else
+ } else {
+ __u32 level = D_LFSCK;
+
rc = -EREMCHG;
+ if (!thread_is_running(&dev->od_scrub.os_thread))
+ level |= D_CONSOLE;
+
+ CDEBUG(level, "%s: the OI mapping for the FID "
+ DFID" become inconsistent, the given ID "
+ "%u/%u, the ID in OI mapping %u/%u\n",
+ osd_name(dev), PFID(fid),
+ saved_id.oii_ino, saved_id.oii_gen,
+ id->oii_ino, id->oii_ino);
+ }
}
} else {
if (id->oii_gen == OSD_OII_NOGEN)
if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
fid = &lma->lma_self_fid;
rc = osd_get_idif(info, inode, dentry, fid);
- if ((rc > 0) || (rc == -ENODATA && osd->od_lma_self_repair)) {
+ if ((rc > 0) || (rc == -ENODATA && osd->od_index_in_idif)) {
/* 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. */
osd_lma_self_repair(info, osd, inode, rfid,
- fid_is_on_ost(info, osd, fid, OI_CHECK_FLD) ?
- LMAC_FID_ON_OST : 0);
+ LMAC_FID_ON_OST);
RETURN(0);
}
}
}
if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
+ __u32 level = D_LFSCK;
+
if (fid_is_idif(rfid) && fid_is_idif(fid)) {
struct ost_id *oi = &info->oti_ostid;
struct lu_fid *fid1 = &info->oti_fid3;
fid_to_ostid(fid, oi);
ostid_to_fid(fid1, oi, idx);
if (lu_fid_eq(fid1, rfid)) {
- if (osd->od_lma_self_repair)
+ if (osd->od_index_in_idif)
osd_lma_self_repair(info, osd,
inode, rfid,
LMAC_FID_ON_OST);
}
}
- CDEBUG(D_INODE, "%s: FID "DFID" != self_fid "DFID"\n",
- osd_name(osd), PFID(rfid), PFID(fid));
+
rc = -EREMCHG;
+ if (!thread_is_running(&osd->od_scrub.os_thread))
+ level |= D_CONSOLE;
+
+ CDEBUG(level, "%s: FID "DFID" != self_fid "DFID"\n",
+ osd_name(osd), PFID(rfid), PFID(fid));
}
RETURN(rc);
* whole device. */
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);
+ 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)
result = -EINPROGRESS;
else
osd_sb(osd)->s_root->d_inode,
mode);
if (!IS_ERR(inode)) {
- /* Do not update file c/mtime in ldiskfs.
- * NB: don't need any lock because no contention at this
- * early stage */
- inode->i_flags |= S_NOCMTIME;
+ /* Do not update file c/mtime in ldiskfs. */
+ inode->i_flags |= S_NOCMTIME;
/* For new created object, it must be consistent,
* and it is unnecessary to scrub against it. */
ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
+
obj->oo_inode = inode;
result = 0;
} else {
result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
th);
- if (result == 0) {
+ if (result == 0) {
osd_attr_init(info, obj, attr, dof);
osd_object_init0(obj);
- /* bz 24037 */
- if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW))
- unlock_new_inode(obj->oo_inode);
- }
+ }
+
+ if (obj->oo_inode != NULL) {
+ LASSERT(obj->oo_inode->i_state & I_NEW);
+
+ unlock_new_inode(obj->oo_inode);
+ }
/* restore previous umask value */
current->fs->umask = umask;
RETURN(local);
}
+ ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
+ unlock_new_inode(local);
+
/* Set special LMA flag for local agent inode */
rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
if (rc != 0) {
osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
result = __osd_object_create(info, obj, attr, hint, dof, th);
- if (result == 0)
- result = osd_ea_fid_set(info, obj->oo_inode, fid,
+ if (result == 0) {
+ if (fid_is_idif(fid) &&
+ !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
+ struct lu_fid *tfid = &info->oti_fid;
+ struct ost_id *oi = &info->oti_ostid;
+
+ fid_to_ostid(fid, oi);
+ ostid_to_fid(tfid, oi, 0);
+ result = osd_ea_fid_set(info, obj->oo_inode, tfid,
+ LMAC_FID_ON_OST, 0);
+ } else {
+ result = osd_ea_fid_set(info, obj->oo_inode, fid,
fid_is_on_ost(info, osd_obj2dev(obj),
fid, OI_CHECK_FLD) ?
LMAC_FID_ON_OST : 0, 0);
+ }
+ }
if (result == 0)
result = __osd_oi_insert(env, obj, fid, th);
if (!dev->od_noscrub && ++once == 1) {
rc = osd_scrub_start(dev, SS_AUTO_PARTIAL | SS_CLEAR_DRYRUN |
SS_CLEAR_FAILOUT);
- LCONSOLE_WARN("%.16s: trigger OI scrub by RPC for "DFID
- ", rc = %d [2]\n",
- LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
- PFID(fid), rc);
+ CDEBUG(D_LFSCK | D_CONSOLE, "%.16s: trigger OI scrub by RPC "
+ "for "DFID", rc = %d [2]\n",
+ LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
+ PFID(fid), rc);
if (rc == 0 || rc == -EALREADY)
goto again;
}
__u32 unused,
struct lustre_capa *capa)
{
- struct osd_it_iam *it;
- struct osd_thread_info *oti = osd_oti_get(env);
- struct osd_object *obj = osd_dt_obj(dt);
- struct lu_object *lo = &dt->do_lu;
- struct iam_path_descr *ipd;
- struct iam_container *bag = &obj->oo_dir->od_container;
+ struct osd_it_iam *it;
+ struct osd_object *obj = osd_dt_obj(dt);
+ struct lu_object *lo = &dt->do_lu;
+ struct iam_path_descr *ipd;
+ struct iam_container *bag = &obj->oo_dir->od_container;
if (!dt_object_exists(dt))
return ERR_PTR(-ENOENT);
- if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
- return ERR_PTR(-EACCES);
+ if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
+ return ERR_PTR(-EACCES);
- if (oti->oti_it_inline) {
- OBD_ALLOC_PTR(it);
- if (it == NULL)
- return ERR_PTR(-ENOMEM);
- } else {
- it = &oti->oti_it;
- oti->oti_it_inline = 1;
- }
+ OBD_ALLOC_PTR(it);
+ if (it == NULL)
+ return ERR_PTR(-ENOMEM);
ipd = osd_it_ipd_get(env, bag);
if (likely(ipd != NULL)) {
iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
return (struct dt_it *)it;
} else {
- if (it != &oti->oti_it)
- OBD_FREE_PTR(it);
- else
- oti->oti_it_inline = 0;
-
+ OBD_FREE_PTR(it);
return ERR_PTR(-ENOMEM);
}
}
static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
{
- struct osd_thread_info *oti = osd_oti_get(env);
struct osd_it_iam *it = (struct osd_it_iam *)di;
struct osd_object *obj = it->oi_obj;
iam_it_fini(&it->oi_it);
osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
lu_object_put(env, &obj->oo_dt.do_lu);
- if (it != &oti->oti_it)
- OBD_FREE_PTR(it);
- else
- oti->oti_it_inline = 0;
+ OBD_FREE_PTR(it);
}
/**
{
struct osd_object *obj = osd_dt_obj(dt);
struct osd_thread_info *info = osd_oti_get(env);
- struct osd_it_ea *it;
+ struct osd_it_ea *oie;
struct file *file;
struct lu_object *lo = &dt->do_lu;
- struct dentry *obj_dentry = &info->oti_it_dentry;
+ struct dentry *obj_dentry;
ENTRY;
if (!dt_object_exists(dt))
RETURN(ERR_PTR(-ENOENT));
- if (info->oti_it_inline) {
- OBD_ALLOC_PTR(it);
- if (it == NULL)
- RETURN(ERR_PTR(-ENOMEM));
- } else {
- it = &info->oti_it_ea;
- info->oti_it_inline = 1;
- }
+ OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
+ if (oie == NULL)
+ RETURN(ERR_PTR(-ENOMEM));
+ obj_dentry = &oie->oie_dentry;
obj_dentry->d_inode = obj->oo_inode;
obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
obj_dentry->d_name.hash = 0;
- it->oie_rd_dirent = 0;
- it->oie_it_dirent = 0;
- it->oie_dirent = NULL;
- it->oie_buf = info->oti_it_ea_buf;
- it->oie_obj = obj;
+ oie->oie_rd_dirent = 0;
+ oie->oie_it_dirent = 0;
+ oie->oie_dirent = NULL;
+ if (unlikely(!info->oti_it_ea_buf_used)) {
+ oie->oie_buf = info->oti_it_ea_buf;
+ info->oti_it_ea_buf_used = 1;
+ } else {
+ OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
+ if (oie->oie_buf == NULL)
+ RETURN(ERR_PTR(-ENOMEM));
+ }
+ oie->oie_obj = obj;
+
+ file = &oie->oie_file;
- file = &it->oie_file;
- /* Reset the "file" totally to avoid to reuse any old value from
- * former readdir handling, the "file->f_pos" should be zero. */
- memset(file, 0, sizeof(*file));
/* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
if (attr & LUDA_64BITHASH)
file->f_mode = FMODE_64BITHASH;
set_file_inode(file, obj->oo_inode);
lu_object_get(lo);
- RETURN((struct dt_it *) it);
+ RETURN((struct dt_it *) oie);
}
/**
*/
static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
{
- struct osd_thread_info *info = osd_oti_get(env);
- struct osd_it_ea *it = (struct osd_it_ea *)di;
- struct osd_object *obj = it->oie_obj;
- struct inode *inode = obj->oo_inode;
+ struct osd_thread_info *info = osd_oti_get(env);
+ struct osd_it_ea *oie = (struct osd_it_ea *)di;
+ struct osd_object *obj = oie->oie_obj;
+ struct inode *inode = obj->oo_inode;
- ENTRY;
- it->oie_file.f_op->release(inode, &it->oie_file);
- lu_object_put(env, &obj->oo_dt.do_lu);
- if (it != &info->oti_it_ea)
- OBD_FREE_PTR(it);
+ ENTRY;
+ oie->oie_file.f_op->release(inode, &oie->oie_file);
+ lu_object_put(env, &obj->oo_dt.do_lu);
+ if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
+ OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
else
- info->oti_it_inline = 0;
+ info->oti_it_ea_buf_used = 0;
+ OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
EXIT;
}
if (rc != 0)
GOTO(out_site, rc);
- /* self-repair LMA by default */
- o->od_lma_self_repair = 1;
-
INIT_LIST_HEAD(&o->od_ios_list);
/* setup scrub, including OI files initialization */
rc = osd_scrub_setup(env, o);
}
static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
- struct lu_device *dev)
+ struct lu_device *dev)
{
- struct osd_device *osd = osd_dev(dev);
- int result = 0;
+ struct osd_device *osd = osd_dev(dev);
+ struct lr_server_data *lsd =
+ &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
+ int result = 0;
ENTRY;
if (osd->od_quota_slave != NULL) {
RETURN(result);
}
+ if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
+#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
+ if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
+ osd->od_index_in_idif = 1;
+ } else {
+ osd->od_index_in_idif = 0;
+ result = osd_register_proc_index_in_idif(osd);
+ if (result != 0)
+ RETURN(result);
+ }
+#else
+ osd->od_index_in_idif = 1;
+#endif
+ }
+
result = osd_fid_init(env, osd);
RETURN(result);
.ldt_ctx_tags = LCT_LOCAL,
};
+static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
+{
+ struct osd_device *osd = osd_dev(obd->obd_lu_dev);
+ struct super_block *sb = osd_sb(osd);
+
+ return (osd->od_mnt == NULL || sb->s_flags & MS_RDONLY);
+}
+
/*
* lprocfs legacy support.
*/
.o_connect = osd_obd_connect,
.o_disconnect = osd_obd_disconnect,
.o_fid_alloc = osd_fid_alloc,
+ .o_health_check = osd_health_check,
};
static int __init osd_mod_init(void)
git://git.whamcloud.com / fs / lustre-release.git / blobdiff