int mdt_capa_keys_init(const struct lu_env *env, struct mdt_device *mdt)
{
struct lustre_capa_key *keys = mdt->mdt_capa_keys;
struct mdt_thread_info *mti;
struct dt_object *obj;
struct lu_attr *la;
mdsno_t mdsnum;
unsigned long size;
int rc;
ENTRY;
mdsnum = mdt_seq_site(mdt)->ss_node_id;
mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
LASSERT(mti != NULL);
la = &mti->mti_attr.ma_attr;
obj = mdt->mdt_ck_obj;
rc = obj->do_ops->do_attr_get(env, mdt->mdt_ck_obj, la, BYPASS_CAPA);
if (rc)
RETURN(rc);
size = (unsigned long)la->la_size;
if (size == 0) {
int i;
for (i = 0; i < 2; i++) {
make_capa_key(&keys[i], mdsnum, i);
DEBUG_CAPA_KEY(D_SEC, &keys[i], "initializing");
}
rc = write_capa_keys(env, mdt, keys);
if (rc) {
CERROR("error writing MDS %s: rc %d\n", CAPA_KEYS, rc);
RETURN(rc);
}
} else {
rc = read_capa_keys(env, mdt, keys);
if (rc) {
CERROR("error reading MDS %s: rc %d\n", CAPA_KEYS, rc);
RETURN(rc);
}
}
set_capa_key_expiry(mdt);
cfs_timer_arm(&mdt->mdt_ck_timer, mdt->mdt_ck_expiry);
CDEBUG(D_SEC, "mds_ck_timer %lu\n", mdt->mdt_ck_expiry);
RETURN(0);
}
static int mdt_ck_thread_main(void *args)
{
struct mdt_device *mdt = args;
struct ptlrpc_thread *thread = &mdt->mdt_ck_thread;
struct lustre_capa_key *bkey = &mdt->mdt_capa_keys[0],
*rkey = &mdt->mdt_capa_keys[1];
struct lustre_capa_key *tmp;
struct lu_env env;
struct mdt_thread_info *info;
struct md_device *next;
struct l_wait_info lwi = { 0 };
mdsno_t mdsnum;
int rc;
ENTRY;
unshare_fs_struct();
cfs_block_allsigs();
thread_set_flags(thread, SVC_RUNNING);
wake_up(&thread->t_ctl_waitq);
rc = lu_env_init(&env, LCT_MD_THREAD|LCT_REMEMBER|LCT_NOREF);
if (rc)
RETURN(rc);
thread->t_env = &env;
env.le_ctx.lc_thread = thread;
env.le_ctx.lc_cookie = 0x1;
info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
LASSERT(info != NULL);
tmp = &info->mti_capa_key;
mdsnum = mdt_seq_site(mdt)->ss_node_id;
while (1) {
l_wait_event(thread->t_ctl_waitq,
thread_is_stopping(thread) ||
thread_is_event(thread),
&lwi);
if (thread_is_stopping(thread))
break;
thread_clear_flags(thread, SVC_EVENT);
if (cfs_time_before(cfs_time_current(), mdt->mdt_ck_expiry))
break;
*tmp = *rkey;
make_capa_key(tmp, mdsnum, rkey->lk_keyid);
next = mdt->mdt_child;
rc = next->md_ops->mdo_update_capa_key(&env, next, tmp);
if (!rc) {
spin_lock(&capa_lock);
*bkey = *rkey;
*rkey = *tmp;
spin_unlock(&capa_lock);
rc = write_capa_keys(&env, mdt, mdt->mdt_capa_keys);
if (rc) {
spin_lock(&capa_lock);
*rkey = *bkey;
memset(bkey, 0, sizeof(*bkey));
spin_unlock(&capa_lock);
} else {
set_capa_key_expiry(mdt);
DEBUG_CAPA_KEY(D_SEC, rkey, "new");
}
}
if (rc) {
DEBUG_CAPA_KEY(D_ERROR, rkey, "update failed for");
/* next retry is in 300 sec */
mdt->mdt_ck_expiry = jiffies + 300 * HZ;
}
cfs_timer_arm(&mdt->mdt_ck_timer, mdt->mdt_ck_expiry);
CDEBUG(D_SEC, "mdt_ck_timer %lu\n", mdt->mdt_ck_expiry);
}
lu_env_fini(&env);
thread_set_flags(thread, SVC_STOPPED);
wake_up(&thread->t_ctl_waitq);
RETURN(0);
}
/*
* VBR: we save three versions in reply:
* 0 - parent. Check that parent version is the same during replay.
* 1 - name. Version of 'name' if file exists with the same name or
* ENOENT_VERSION, it is needed because file may appear due to missed replays.
* 2 - child. Version of child by FID. Must be ENOENT. It is mostly sanity
* check.
*/
static int mdt_md_create(struct mdt_thread_info *info)
{
struct mdt_device *mdt = info->mti_mdt;
struct mdt_object *parent;
struct mdt_object *child;
struct mdt_lock_handle *lh;
struct mdt_body *repbody;
struct md_attr *ma = &info->mti_attr;
struct mdt_reint_record *rr = &info->mti_rr;
struct lu_name *lname;
int rc;
ENTRY;
DEBUG_REQ(D_INODE, mdt_info_req(info), "Create (%s->"DFID") in "DFID,
rr->rr_name, PFID(rr->rr_fid2), PFID(rr->rr_fid1));
repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
lh = &info->mti_lh[MDT_LH_PARENT];
mdt_lock_pdo_init(lh, LCK_PW, rr->rr_name, rr->rr_namelen);
parent = mdt_object_find_lock(info, rr->rr_fid1, lh,
MDS_INODELOCK_UPDATE);
if (IS_ERR(parent))
RETURN(PTR_ERR(parent));
if (mdt_object_obf(parent))
GOTO(out_put_parent, rc = -EPERM);
rc = mdt_version_get_check_save(info, parent, 0);
if (rc)
GOTO(out_put_parent, rc);
/*
* Check child name version during replay.
* During create replay a file may exist with same name.
*/
lname = mdt_name(info->mti_env, (char *)rr->rr_name, rr->rr_namelen);
rc = mdt_lookup_version_check(info, parent, lname,
&info->mti_tmp_fid1, 1);
if (rc == 0)
GOTO(out_put_parent, rc = -EEXIST);
/* -ENOENT is expected here */
if (rc != -ENOENT)
GOTO(out_put_parent, rc);
/* save version of file name for replay, it must be ENOENT here */
mdt_enoent_version_save(info, 1);
child = mdt_object_new(info->mti_env, mdt, rr->rr_fid2);
if (likely(!IS_ERR(child))) {
struct md_object *next = mdt_object_child(parent);
if (mdt_object_remote(child)) {
struct seq_server_site *ss;
struct lu_ucred *uc = mdt_ucred(info);
if (!md_capable(uc, CFS_CAP_SYS_ADMIN)) {
CERROR("%s: Creating remote dir is only "
"permitted for administrator: rc = %d\n",
mdt2obd_dev(mdt)->obd_name, -EPERM);
GOTO(out_put_child, rc = -EPERM);
}
ss = mdt_seq_site(mdt);
if (ss->ss_node_id != 0 &&
mdt->mdt_enable_remote_dir == 0) {
CERROR("%s: remote dir is only permitted on"
" MDT0 or set_param"
" mdt.*.enable_remote_dir=1\n",
mdt2obd_dev(mdt)->obd_name);
GOTO(out_put_child, rc = -EPERM);
}
}
ma->ma_need = MA_INODE;
ma->ma_valid = 0;
/* capa for cross-ref will be stored here */
ma->ma_capa = req_capsule_server_get(info->mti_pill,
&RMF_CAPA1);
LASSERT(ma->ma_capa);
mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
OBD_FAIL_MDS_REINT_CREATE_WRITE);
/* Version of child will be updated on disk. */
info->mti_mos = child;
rc = mdt_version_get_check_save(info, child, 2);
if (rc)
GOTO(out_put_child, rc);
/* Let lower layer know current lock mode. */
info->mti_spec.sp_cr_mode =
mdt_dlm_mode2mdl_mode(lh->mlh_pdo_mode);
/*
* Do not perform lookup sanity check. We know that name does
* not exist.
*/
info->mti_spec.sp_cr_lookup = 0;
info->mti_spec.sp_feat = &dt_directory_features;
rc = mdo_create(info->mti_env, next, lname,
mdt_object_child(child),
&info->mti_spec, ma);
if (rc == 0)
rc = mdt_attr_get_complex(info, child, ma);
if (rc == 0) {
/* Return fid & attr to client. */
if (ma->ma_valid & MA_INODE)
mdt_pack_attr2body(info, repbody, &ma->ma_attr,
mdt_object_fid(child));
}
out_put_child:
mdt_object_put(info->mti_env, child);
} else {
rc = PTR_ERR(child);
}
mdt_create_pack_capa(info, rc, child, repbody);
out_put_parent:
mdt_object_unlock_put(info, parent, lh, rc);
RETURN(rc);
}
