static int llu_dir_do_readpage(struct inode *inode, struct page *page)
{
struct llu_inode_info *lli = llu_i2info(inode);
struct intnl_stat *st = llu_i2stat(inode);
struct llu_sb_info *sbi = llu_i2sbi(inode);
struct ptlrpc_request *request;
struct lustre_handle lockh;
struct mdt_body *body;
struct lookup_intent it = { .it_op = IT_READDIR };
struct md_op_data op_data = {{ 0 }};
ldlm_policy_data_t policy = { .l_inodebits = { MDS_INODELOCK_UPDATE } };
int rc = 0;
ENTRY;
llu_prep_md_op_data(&op_data, inode, NULL, NULL, 0, 0, LUSTRE_OPC_ANY);
rc = md_lock_match(sbi->ll_md_exp, LDLM_FL_BLOCK_GRANTED,
&lli->lli_fid, LDLM_IBITS, &policy, LCK_CR, &lockh);
if (!rc) {
struct ldlm_enqueue_info einfo = {
.ei_type = LDLM_IBITS,
.ei_mode = LCK_CR,
.ei_cb_bl = llu_md_blocking_ast,
.ei_cb_cp = ldlm_completion_ast,
.ei_cbdata = inode,
};
rc = md_enqueue(sbi->ll_md_exp, &einfo, &it,
&op_data, &lockh, NULL, 0, NULL,
LDLM_FL_CANCEL_ON_BLOCK);
request = (struct ptlrpc_request *)it.d.lustre.it_data;
if (request)
ptlrpc_req_finished(request);
if (rc < 0) {
CERROR("lock enqueue: err: %d\n", rc);
RETURN(rc);
}
}
ldlm_lock_dump_handle(D_OTHER, &lockh);
op_data.op_offset = (__u64)hash_x_index(page->index, 0);
op_data.op_npages = 1;
rc = md_readpage(sbi->ll_md_exp, &op_data, &page, &request);
if (!rc) {
body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
LASSERT(body != NULL); /* checked by md_readpage() */
if (body->valid & OBD_MD_FLSIZE)
st->st_size = body->size;
} else {
CERROR("read_dir_page(%ld) error %d\n", page->index, rc);
}
ptlrpc_req_finished(request);
EXIT;
ldlm_lock_decref(&lockh, LCK_CR);
return rc;
}
int osc_quotacheck(struct obd_device *unused, struct obd_export *exp,
struct obd_quotactl *oqctl)
{
struct client_obd *cli = &exp->exp_obd->u.cli;
struct ptlrpc_request *req;
struct obd_quotactl *body;
int rc;
ENTRY;
req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
&RQF_OST_QUOTACHECK, LUSTRE_OST_VERSION,
OST_QUOTACHECK);
if (req == NULL)
RETURN(-ENOMEM);
body = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
*body = *oqctl;
ptlrpc_request_set_replen(req);
/* the next poll will find -ENODATA, that means quotacheck is
* going on */
cli->cl_qchk_stat = -ENODATA;
rc = ptlrpc_queue_wait(req);
if (rc)
cli->cl_qchk_stat = rc;
ptlrpc_req_finished(req);
RETURN(rc);
}
/*
* Client's outgoing request callback
*/
void request_out_callback(lnet_event_t *ev)
{
struct ptlrpc_cb_id *cbid = ev->md.user_ptr;
struct ptlrpc_request *req = cbid->cbid_arg;
LASSERT(ev->type == LNET_EVENT_SEND ||
ev->type == LNET_EVENT_UNLINK);
LASSERT(ev->unlinked);
DEBUG_REQ(D_NET, req, "type %d, status %d", ev->type, ev->status);
sptlrpc_request_out_callback(req);
req->rq_real_sent = cfs_time_current_sec();
if (ev->type == LNET_EVENT_UNLINK || ev->status != 0) {
/* Failed send: make it seem like the reply timed out, just
* like failing sends in client.c does currently... */
spin_lock(&req->rq_lock);
req->rq_net_err = 1;
spin_unlock(&req->rq_lock);
ptlrpc_client_wake_req(req);
}
ptlrpc_req_finished(req);
}
static int llog_client_destroy(const struct lu_env *env,
struct llog_handle *loghandle)
{
struct obd_import *imp;
struct ptlrpc_request *req = NULL;
struct llogd_body *body;
int rc;
LLOG_CLIENT_ENTRY(loghandle->lgh_ctxt, imp);
req = ptlrpc_request_alloc_pack(imp, &RQF_LLOG_ORIGIN_HANDLE_DESTROY,
LUSTRE_LOG_VERSION,
LLOG_ORIGIN_HANDLE_DESTROY);
if (req == NULL)
GOTO(err_exit, rc =-ENOMEM);
body = req_capsule_client_get(&req->rq_pill, &RMF_LLOGD_BODY);
body->lgd_logid = loghandle->lgh_id;
body->lgd_llh_flags = loghandle->lgh_hdr->llh_flags;
if (!(body->lgd_llh_flags & LLOG_F_IS_PLAIN))
CERROR("%s: wrong llog flags %x\n", imp->imp_obd->obd_name,
body->lgd_llh_flags);
ptlrpc_request_set_replen(req);
rc = ptlrpc_queue_wait(req);
ptlrpc_req_finished(req);
err_exit:
LLOG_CLIENT_EXIT(loghandle->lgh_ctxt, imp);
return rc;
}
static int llog_client_next_block(const struct lu_env *env,
struct llog_handle *loghandle,
int *cur_idx, int next_idx,
__u64 *cur_offset, void *buf, int len)
{
struct obd_import *imp;
struct ptlrpc_request *req = NULL;
struct llogd_body *body;
void *ptr;
int rc;
LLOG_CLIENT_ENTRY(loghandle->lgh_ctxt, imp);
req = ptlrpc_request_alloc_pack(imp, &RQF_LLOG_ORIGIN_HANDLE_NEXT_BLOCK,
LUSTRE_LOG_VERSION,
LLOG_ORIGIN_HANDLE_NEXT_BLOCK);
if (!req) {
rc = -ENOMEM;
goto err_exit;
}
body = req_capsule_client_get(&req->rq_pill, &RMF_LLOGD_BODY);
body->lgd_logid = loghandle->lgh_id;
body->lgd_ctxt_idx = loghandle->lgh_ctxt->loc_idx - 1;
body->lgd_llh_flags = loghandle->lgh_hdr->llh_flags;
body->lgd_index = next_idx;
body->lgd_saved_index = *cur_idx;
body->lgd_len = len;
body->lgd_cur_offset = *cur_offset;
req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_SERVER, len);
ptlrpc_request_set_replen(req);
rc = ptlrpc_queue_wait(req);
if (rc)
goto out;
body = req_capsule_server_get(&req->rq_pill, &RMF_LLOGD_BODY);
if (!body) {
rc = -EFAULT;
goto out;
}
/* The log records are swabbed as they are processed */
ptr = req_capsule_server_get(&req->rq_pill, &RMF_EADATA);
if (!ptr) {
rc = -EFAULT;
goto out;
}
*cur_idx = body->lgd_saved_index;
*cur_offset = body->lgd_cur_offset;
memcpy(buf, ptr, len);
out:
ptlrpc_req_finished(req);
err_exit:
LLOG_CLIENT_EXIT(loghandle->lgh_ctxt, imp);
return rc;
}
static int llog_client_read_header(const struct lu_env *env,
struct llog_handle *handle)
{
struct obd_import *imp;
struct ptlrpc_request *req = NULL;
struct llogd_body *body;
struct llog_log_hdr *hdr;
struct llog_rec_hdr *llh_hdr;
int rc;
LLOG_CLIENT_ENTRY(handle->lgh_ctxt, imp);
req = ptlrpc_request_alloc_pack(imp, &RQF_LLOG_ORIGIN_HANDLE_READ_HEADER,
LUSTRE_LOG_VERSION,
LLOG_ORIGIN_HANDLE_READ_HEADER);
if (req == NULL) {
rc = -ENOMEM;
goto err_exit;
}
body = req_capsule_client_get(&req->rq_pill, &RMF_LLOGD_BODY);
body->lgd_logid = handle->lgh_id;
body->lgd_ctxt_idx = handle->lgh_ctxt->loc_idx - 1;
body->lgd_llh_flags = handle->lgh_hdr->llh_flags;
ptlrpc_request_set_replen(req);
rc = ptlrpc_queue_wait(req);
if (rc)
goto out;
hdr = req_capsule_server_get(&req->rq_pill, &RMF_LLOG_LOG_HDR);
if (hdr == NULL) {
rc = -EFAULT;
goto out;
}
memcpy(handle->lgh_hdr, hdr, sizeof(*hdr));
handle->lgh_last_idx = handle->lgh_hdr->llh_tail.lrt_index;
/* sanity checks */
llh_hdr = &handle->lgh_hdr->llh_hdr;
if (llh_hdr->lrh_type != LLOG_HDR_MAGIC) {
CERROR("bad log header magic: %#x (expecting %#x)\n",
llh_hdr->lrh_type, LLOG_HDR_MAGIC);
rc = -EIO;
} else if (llh_hdr->lrh_len != LLOG_CHUNK_SIZE) {
CERROR("incorrectly sized log header: %#x "
"(expecting %#x)\n",
llh_hdr->lrh_len, LLOG_CHUNK_SIZE);
CERROR("you may need to re-run lconf --write_conf.\n");
rc = -EIO;
}
out:
ptlrpc_req_finished(req);
err_exit:
LLOG_CLIENT_EXIT(handle->lgh_ctxt, imp);
return rc;
}
void ll_intent_release(struct lookup_intent *it)
{
ENTRY;
CDEBUG(D_INFO, "intent %p released\n", it);
ll_intent_drop_lock(it);
/* We are still holding extra reference on a request, need to free it */
if (it_disposition(it, DISP_ENQ_OPEN_REF))
ptlrpc_req_finished(it->d.lustre.it_data); /* ll_file_open */
if (it_disposition(it, DISP_ENQ_CREATE_REF)) /* create rec */
ptlrpc_req_finished(it->d.lustre.it_data);
if (it_disposition(it, DISP_ENQ_COMPLETE)) /* saved req from revalidate
* to lookup */
ptlrpc_req_finished(it->d.lustre.it_data);
it->d.lustre.it_disposition = 0;
it->d.lustre.it_data = NULL;
EXIT;
}
/*
* Send non-intent quota request to master.
*
* \param env - the environment passed by the caller
* \param exp - is the export to use to send the acquire RPC
* \param qbody - quota body to be packed in request
* \param sync - synchronous or asynchronous
* \param completion - completion callback
* \param qqi - is the qsd_qtype_info structure to pass to the completion
* function
* \param lqe - is the qid entry to be processed
*
* \retval 0 - success
* \retval -ve - appropriate errors
*/
int qsd_send_dqacq(const struct lu_env *env, struct obd_export *exp,
struct quota_body *qbody, bool sync,
qsd_req_completion_t completion, struct qsd_qtype_info *qqi,
struct lustre_handle *lockh, struct lquota_entry *lqe)
{
struct ptlrpc_request *req;
struct quota_body *req_qbody;
struct qsd_async_args *aa;
int rc;
ENTRY;
LASSERT(exp);
req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_QUOTA_DQACQ);
if (req == NULL)
GOTO(out, rc = -ENOMEM);
req->rq_no_resend = req->rq_no_delay = 1;
req->rq_no_retry_einprogress = 1;
rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, QUOTA_DQACQ);
if (rc) {
ptlrpc_request_free(req);
GOTO(out, rc);
}
req->rq_request_portal = MDS_READPAGE_PORTAL;
req_qbody = req_capsule_client_get(&req->rq_pill, &RMF_QUOTA_BODY);
*req_qbody = *qbody;
ptlrpc_request_set_replen(req);
CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
aa = ptlrpc_req_async_args(req);
aa->aa_exp = exp;
aa->aa_qqi = qqi;
aa->aa_arg = (void *)lqe;
aa->aa_completion = completion;
lustre_handle_copy(&aa->aa_lockh, lockh);
if (sync) {
rc = ptlrpc_queue_wait(req);
rc = qsd_dqacq_interpret(env, req, aa, rc);
ptlrpc_req_finished(req);
} else {
req->rq_interpret_reply = qsd_dqacq_interpret;
ptlrpcd_add_req(req);
}
RETURN(rc);
out:
completion(env, qqi, qbody, NULL, lockh, NULL, lqe, rc);
return rc;
}
static int llog_client_prev_block(const struct lu_env *env,
struct llog_handle *loghandle,
int prev_idx, void *buf, int len)
{
struct obd_import *imp;
struct ptlrpc_request *req = NULL;
struct llogd_body *body;
void *ptr;
int rc;
LLOG_CLIENT_ENTRY(loghandle->lgh_ctxt, imp);
req = ptlrpc_request_alloc_pack(imp, &RQF_LLOG_ORIGIN_HANDLE_PREV_BLOCK,
LUSTRE_LOG_VERSION,
LLOG_ORIGIN_HANDLE_PREV_BLOCK);
if (!req) {
rc = -ENOMEM;
goto err_exit;
}
body = req_capsule_client_get(&req->rq_pill, &RMF_LLOGD_BODY);
body->lgd_logid = loghandle->lgh_id;
body->lgd_ctxt_idx = loghandle->lgh_ctxt->loc_idx - 1;
body->lgd_llh_flags = loghandle->lgh_hdr->llh_flags;
body->lgd_index = prev_idx;
body->lgd_len = len;
req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_SERVER, len);
ptlrpc_request_set_replen(req);
rc = ptlrpc_queue_wait(req);
if (rc)
goto out;
body = req_capsule_server_get(&req->rq_pill, &RMF_LLOGD_BODY);
if (!body) {
rc = -EFAULT;
goto out;
}
ptr = req_capsule_server_get(&req->rq_pill, &RMF_EADATA);
if (!ptr) {
rc = -EFAULT;
goto out;
}
memcpy(buf, ptr, len);
out:
ptlrpc_req_finished(req);
err_exit:
LLOG_CLIENT_EXIT(loghandle->lgh_ctxt, imp);
return rc;
}
/**
* Cliens updates SOM attributes on MDS (including llog cookies):
* obd_getattr with no lock and md_setattr.
*/
int ll_som_update(struct inode *inode, struct md_op_data *op_data)
{
struct ll_inode_info *lli = ll_i2info(inode);
struct ptlrpc_request *request = NULL;
__u32 old_flags;
struct obdo *oa;
int rc;
ENTRY;
LASSERT(op_data != NULL);
if (lli->lli_flags & LLIF_MDS_SIZE_LOCK)
CERROR("ino %lu/%u(flags %u) som valid it just after "
"recovery\n", inode->i_ino, inode->i_generation,
lli->lli_flags);
OBDO_ALLOC(oa);
if (!oa) {
CERROR("can't allocate memory for Size-on-MDS update.\n");
RETURN(-ENOMEM);
}
old_flags = op_data->op_flags;
op_data->op_flags = MF_SOM_CHANGE;
/* If inode is already in another epoch, skip getattr from OSTs. */
if (lli->lli_ioepoch == op_data->op_ioepoch) {
rc = ll_inode_getattr(inode, oa, op_data->op_ioepoch,
old_flags & MF_GETATTR_LOCK);
if (rc) {
oa->o_valid = 0;
if (rc != -ENOENT)
CERROR("inode_getattr failed (%d): unable to "
"send a Size-on-MDS attribute update "
"for inode %lu/%u\n", rc, inode->i_ino,
inode->i_generation);
} else {
CDEBUG(D_INODE, "Size-on-MDS update on "DFID"\n",
PFID(&lli->lli_fid));
}
/* Install attributes into op_data. */
md_from_obdo(op_data, oa, oa->o_valid);
}
rc = md_setattr(ll_i2sbi(inode)->ll_md_exp, op_data,
NULL, 0, NULL, 0, &request, NULL);
ptlrpc_req_finished(request);
OBDO_FREE(oa);
RETURN(rc);
}
struct inode *search_inode_for_lustre(struct super_block *sb,
const struct lu_fid *fid)
{
struct ll_sb_info *sbi = ll_s2sbi(sb);
struct ptlrpc_request *req = NULL;
struct inode *inode = NULL;
int eadatalen = 0;
unsigned long hash = cl_fid_build_ino(fid,
ll_need_32bit_api(sbi));
struct md_op_data *op_data;
int rc;
CDEBUG(D_INFO, "searching inode for:(%lu,"DFID")\n", hash, PFID(fid));
inode = ilookup5(sb, hash, ll_nfs_test_inode, (void *)fid);
if (inode)
return inode;
rc = ll_get_default_mdsize(sbi, &eadatalen);
if (rc)
return ERR_PTR(rc);
/* Because inode is NULL, ll_prep_md_op_data can not
* be used here. So we allocate op_data ourselves */
op_data = kzalloc(sizeof(*op_data), GFP_NOFS);
if (!op_data)
return ERR_PTR(-ENOMEM);
op_data->op_fid1 = *fid;
op_data->op_mode = eadatalen;
op_data->op_valid = OBD_MD_FLEASIZE;
/* mds_fid2dentry ignores f_type */
rc = md_getattr(sbi->ll_md_exp, op_data, &req);
kfree(op_data);
if (rc) {
CERROR("can't get object attrs, fid "DFID", rc %d\n",
PFID(fid), rc);
return ERR_PTR(rc);
}
rc = ll_prep_inode(&inode, req, sb, NULL);
ptlrpc_req_finished(req);
if (rc)
return ERR_PTR(rc);
return inode;
}
/**
* Implementation of dt_object_operations::do_object_lock
*
* Enqueue a lock (by ldlm_cli_enqueue()) of remote object on the remote MDT,
* which will lock the object in the global namespace. And because the
* cross-MDT locks are relatively rare compared with normal local MDT operation,
* let's release it right away, instead of putting it into the LRU list.
*
* \param[in] env execution environment
* \param[in] dt object to be locked
* \param[out] lh lock handle
* \param[in] einfo enqueue information
* \param[in] policy lock policy
*
* \retval ELDLM_OK if locking the object succeeds.
* \retval negative errno if locking fails.
*/
static int osp_md_object_lock(const struct lu_env *env,
struct dt_object *dt,
struct lustre_handle *lh,
struct ldlm_enqueue_info *einfo,
union ldlm_policy_data *policy)
{
struct ldlm_res_id *res_id;
struct dt_device *dt_dev = lu2dt_dev(dt->do_lu.lo_dev);
struct osp_device *osp = dt2osp_dev(dt_dev);
struct ptlrpc_request *req;
int rc = 0;
__u64 flags = 0;
enum ldlm_mode mode;
res_id = einfo->ei_res_id;
LASSERT(res_id != NULL);
mode = ldlm_lock_match(osp->opd_obd->obd_namespace,
LDLM_FL_BLOCK_GRANTED, res_id,
einfo->ei_type, policy,
einfo->ei_mode, lh, 0);
if (mode > 0)
return ELDLM_OK;
if (einfo->ei_nonblock)
flags |= LDLM_FL_BLOCK_NOWAIT;
req = ldlm_enqueue_pack(osp->opd_exp, 0);
if (IS_ERR(req))
RETURN(PTR_ERR(req));
rc = ldlm_cli_enqueue(osp->opd_exp, &req, einfo, res_id,
(const union ldlm_policy_data *)policy,
&flags, NULL, 0, LVB_T_NONE, lh, 0);
ptlrpc_req_finished(req);
if (rc == ELDLM_OK) {
struct ldlm_lock *lock;
lock = __ldlm_handle2lock(lh, 0);
ldlm_set_cbpending(lock);
LDLM_LOCK_PUT(lock);
}
return rc == ELDLM_OK ? 0 : -EIO;
}
int ptlrpc_obd_ping(struct obd_device *obd)
{
int rc;
struct ptlrpc_request *req;
req = ptlrpc_prep_ping(obd->u.cli.cl_import);
if (!req)
return -ENOMEM;
req->rq_send_state = LUSTRE_IMP_FULL;
rc = ptlrpc_queue_wait(req);
ptlrpc_req_finished(req);
return rc;
}
static int llu_lookup_it(struct inode *parent, struct pnode *pnode,
struct lookup_intent *it, int flags)
{
struct md_op_data op_data = {{ 0 }};
struct ptlrpc_request *req = NULL;
struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
__u32 opc;
int rc;
ENTRY;
if (pnode->p_base->pb_name.len > EXT2_NAME_LEN)
RETURN(-ENAMETOOLONG);
if (!it) {
it = &lookup_it;
it->it_op_release = ll_intent_release;
}
if (it->it_op & IT_CREAT)
opc = LUSTRE_OPC_CREATE;
else
opc = LUSTRE_OPC_ANY;
llu_prep_md_op_data(&op_data, parent, NULL,
pnode->p_base->pb_name.name,
pnode->p_base->pb_name.len, flags, opc);
rc = md_intent_lock(llu_i2mdexp(parent), &op_data, NULL, 0, it,
flags, &req, llu_md_blocking_ast,
LDLM_FL_CANCEL_ON_BLOCK);
if (rc < 0)
GOTO(out, rc);
rc = lookup_it_finish(req, DLM_REPLY_REC_OFF, it, parent, pnode);
if (rc != 0) {
ll_intent_release(it);
GOTO(out, rc);
}
llu_lookup_finish_locks(it, pnode);
out:
if (req)
ptlrpc_req_finished(req);
return rc;
}
int llu_md_close(struct obd_export *md_exp, struct inode *inode)
{
struct llu_inode_info *lli = llu_i2info(inode);
struct ll_file_data *fd = lli->lli_file_data;
struct ptlrpc_request *req = NULL;
struct obd_client_handle *och = &fd->fd_mds_och;
struct intnl_stat *st = llu_i2stat(inode);
struct md_op_data op_data = { { 0 } };
int rc;
ENTRY;
/* clear group lock, if present */
if (fd->fd_flags & LL_FILE_GROUP_LOCKED)
llu_put_grouplock(inode, fd->fd_grouplock.cg_gid);
llu_prepare_close(inode, &op_data, fd);
rc = md_close(md_exp, &op_data, och->och_mod, &req);
if (rc == -EAGAIN) {
/* We are the last writer, so the MDS has instructed us to get
* the file size and any write cookies, then close again. */
LASSERT(lli->lli_open_flags & FMODE_WRITE);
rc = llu_som_update(inode, &op_data);
if (rc) {
CERROR("inode %llu mdc Size-on-MDS update failed: "
"rc = %d\n", (long long)st->st_ino, rc);
rc = 0;
}
} else if (rc) {
CERROR("inode %llu close failed: rc %d\n",
(long long)st->st_ino, rc);
} else {
rc = llu_objects_destroy(req, inode);
if (rc)
CERROR("inode %llu ll_objects destroy: rc = %d\n",
(long long)st->st_ino, rc);
}
md_clear_open_replay_data(md_exp, och);
ptlrpc_req_finished(req);
och->och_fh.cookie = DEAD_HANDLE_MAGIC;
lli->lli_file_data = NULL;
OBD_FREE(fd, sizeof(*fd));
RETURN(rc);
}
int ptlrpc_obd_ping(struct obd_device *obd)
{
int rc;
struct ptlrpc_request *req;
ENTRY;
req = ptlrpc_prep_ping(obd->u.cli.cl_import);
if (req == NULL)
RETURN(-ENOMEM);
req->rq_send_state = LUSTRE_IMP_FULL;
rc = ptlrpc_queue_wait(req);
ptlrpc_req_finished(req);
RETURN(rc);
}
/**
* Implementation of dt_object_operations::do_object_lock
*
* Enqueue a lock (by ldlm_cli_enqueue()) of remote object on the remote MDT,
* which will lock the object in the global namespace. And because the
* cross-MDT locks are relatively rare compared with normal local MDT operation,
* let's release it right away, instead of putting it into the LRU list.
*
* \param[in] env execution environment
* \param[in] dt object to be locked
* \param[out] lh lock handle
* \param[in] einfo enqueue information
* \param[in] policy lock policy
*
* \retval ELDLM_OK if locking the object succeeds.
* \retval negative errno if locking fails.
*/
static int osp_md_object_lock(const struct lu_env *env,
struct dt_object *dt,
struct lustre_handle *lh,
struct ldlm_enqueue_info *einfo,
union ldlm_policy_data *policy)
{
struct ldlm_res_id *res_id;
struct dt_device *dt_dev = lu2dt_dev(dt->do_lu.lo_dev);
struct osp_device *osp = dt2osp_dev(dt_dev);
struct lu_device *top_device;
struct ptlrpc_request *req;
int rc = 0;
__u64 flags = LDLM_FL_NO_LRU;
res_id = einfo->ei_res_id;
LASSERT(res_id != NULL);
if (einfo->ei_nonblock)
flags |= LDLM_FL_BLOCK_NOWAIT;
if (einfo->ei_mode & (LCK_EX | LCK_PW))
flags |= LDLM_FL_COS_INCOMPAT;
req = ldlm_enqueue_pack(osp->opd_exp, 0);
if (IS_ERR(req))
RETURN(PTR_ERR(req));
/* During recovery, it needs to let OSP send enqueue
* without checking recoverying status, in case the
* other target is being recovered at the same time,
* and if we wait here for the import to be recovered,
* it might cause deadlock */
top_device = dt_dev->dd_lu_dev.ld_site->ls_top_dev;
if (top_device->ld_obd->obd_recovering)
req->rq_allow_replay = 1;
rc = ldlm_cli_enqueue(osp->opd_exp, &req, einfo, res_id,
(const union ldlm_policy_data *)policy,
&flags, NULL, 0, LVB_T_NONE, lh, 0);
ptlrpc_req_finished(req);
return rc == ELDLM_OK ? 0 : -EIO;
}
int client_quota_check(struct obd_device *unused, struct obd_export *exp,
struct obd_quotactl *oqctl)
{
struct client_obd *cli = &exp->exp_obd->u.cli;
struct ptlrpc_request *req;
struct obd_quotactl *body;
const struct req_format *rf;
int ver, opc, rc;
ENTRY;
if (!strcmp(exp->exp_obd->obd_type->typ_name, LUSTRE_MDC_NAME)) {
rf = &RQF_MDS_QUOTACHECK;
ver = LUSTRE_MDS_VERSION;
opc = MDS_QUOTACHECK;
} else if (!strcmp(exp->exp_obd->obd_type->typ_name, LUSTRE_OSC_NAME)) {
rf = &RQF_OST_QUOTACHECK;
ver = LUSTRE_OST_VERSION;
opc = OST_QUOTACHECK;
} else {
RETURN(-EINVAL);
}
req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), rf, ver, opc);
if (req == NULL)
RETURN(-ENOMEM);
body = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
*body = *oqctl;
ptlrpc_request_set_replen(req);
/* the next poll will find -ENODATA, that means quotacheck is
* going on */
cli->cl_qchk_stat = -ENODATA;
rc = ptlrpc_queue_wait(req);
if (rc)
cli->cl_qchk_stat = rc;
ptlrpc_req_finished(req);
RETURN(rc);
}
static int target_quotacheck_callback(struct obd_export *exp,
struct obd_quotactl *oqctl)
{
struct ptlrpc_request *req;
struct obd_quotactl *body;
int rc;
ENTRY;
req = ptlrpc_request_alloc_pack(exp->exp_imp_reverse, &RQF_QC_CALLBACK,
LUSTRE_OBD_VERSION, OBD_QC_CALLBACK);
if (req == NULL)
RETURN(-ENOMEM);
body = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
*body = *oqctl;
ptlrpc_request_set_replen(req);
rc = ptlrpc_queue_wait(req);
ptlrpc_req_finished(req);
RETURN(rc);
}
int osc_quotactl(struct obd_device *unused, struct obd_export *exp,
struct obd_quotactl *oqctl)
{
struct ptlrpc_request *req;
struct obd_quotactl *oqc;
int rc;
ENTRY;
req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
&RQF_OST_QUOTACTL, LUSTRE_OST_VERSION,
OST_QUOTACTL);
if (req == NULL)
RETURN(-ENOMEM);
oqc = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
*oqc = *oqctl;
ptlrpc_request_set_replen(req);
ptlrpc_at_set_req_timeout(req);
req->rq_no_resend = 1;
rc = ptlrpc_queue_wait(req);
if (rc)
CERROR("ptlrpc_queue_wait failed, rc: %d\n", rc);
if (req->rq_repmsg &&
(oqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL))) {
*oqctl = *oqc;
} else if (!rc) {
CERROR ("Can't unpack obd_quotactl\n");
rc = -EPROTO;
}
ptlrpc_req_finished(req);
RETURN(rc);
}
/**
* Send request \a request.
* if \a noreply is set, don't expect any reply back and don't set up
* reply buffers.
* Returns 0 on success or error code.
*/
int ptl_send_rpc(struct ptlrpc_request *request, int noreply)
{
int rc;
int rc2;
int mpflag = 0;
struct ptlrpc_connection *connection;
lnet_handle_me_t reply_me_h;
lnet_md_t reply_md;
struct obd_device *obd = request->rq_import->imp_obd;
if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_DROP_RPC))
return 0;
LASSERT(request->rq_type == PTL_RPC_MSG_REQUEST);
LASSERT(request->rq_wait_ctx == 0);
/* If this is a re-transmit, we're required to have disengaged
* cleanly from the previous attempt */
LASSERT(!request->rq_receiving_reply);
LASSERT(!((lustre_msg_get_flags(request->rq_reqmsg) & MSG_REPLAY) &&
(request->rq_import->imp_state == LUSTRE_IMP_FULL)));
if (unlikely(obd != NULL && obd->obd_fail)) {
CDEBUG(D_HA, "muting rpc for failed imp obd %s\n",
obd->obd_name);
/* this prevents us from waiting in ptlrpc_queue_wait */
spin_lock(&request->rq_lock);
request->rq_err = 1;
spin_unlock(&request->rq_lock);
request->rq_status = -ENODEV;
return -ENODEV;
}
connection = request->rq_import->imp_connection;
lustre_msg_set_handle(request->rq_reqmsg,
&request->rq_import->imp_remote_handle);
lustre_msg_set_type(request->rq_reqmsg, PTL_RPC_MSG_REQUEST);
lustre_msg_set_conn_cnt(request->rq_reqmsg,
request->rq_import->imp_conn_cnt);
lustre_msghdr_set_flags(request->rq_reqmsg,
request->rq_import->imp_msghdr_flags);
if (request->rq_resend)
lustre_msg_add_flags(request->rq_reqmsg, MSG_RESENT);
if (request->rq_memalloc)
mpflag = cfs_memory_pressure_get_and_set();
rc = sptlrpc_cli_wrap_request(request);
if (rc)
goto out;
/* bulk register should be done after wrap_request() */
if (request->rq_bulk != NULL) {
rc = ptlrpc_register_bulk(request);
if (rc != 0)
goto out;
}
if (!noreply) {
LASSERT(request->rq_replen != 0);
if (request->rq_repbuf == NULL) {
LASSERT(request->rq_repdata == NULL);
LASSERT(request->rq_repmsg == NULL);
rc = sptlrpc_cli_alloc_repbuf(request,
request->rq_replen);
if (rc) {
/* this prevents us from looping in
* ptlrpc_queue_wait */
spin_lock(&request->rq_lock);
request->rq_err = 1;
spin_unlock(&request->rq_lock);
request->rq_status = rc;
goto cleanup_bulk;
}
} else {
request->rq_repdata = NULL;
request->rq_repmsg = NULL;
}
rc = LNetMEAttach(request->rq_reply_portal,/*XXX FIXME bug 249*/
connection->c_peer, request->rq_xid, 0,
LNET_UNLINK, LNET_INS_AFTER, &reply_me_h);
if (rc != 0) {
CERROR("LNetMEAttach failed: %d\n", rc);
LASSERT(rc == -ENOMEM);
rc = -ENOMEM;
goto cleanup_bulk;
}
}
spin_lock(&request->rq_lock);
/* If the MD attach succeeds, there _will_ be a reply_in callback */
request->rq_receiving_reply = !noreply;
request->rq_req_unlink = 1;
/* We are responsible for unlinking the reply buffer */
request->rq_reply_unlink = !noreply;
/* Clear any flags that may be present from previous sends. */
request->rq_replied = 0;
request->rq_err = 0;
request->rq_timedout = 0;
request->rq_net_err = 0;
request->rq_resend = 0;
request->rq_restart = 0;
request->rq_reply_truncate = 0;
spin_unlock(&request->rq_lock);
if (!noreply) {
reply_md.start = request->rq_repbuf;
reply_md.length = request->rq_repbuf_len;
/* Allow multiple early replies */
reply_md.threshold = LNET_MD_THRESH_INF;
/* Manage remote for early replies */
reply_md.options = PTLRPC_MD_OPTIONS | LNET_MD_OP_PUT |
LNET_MD_MANAGE_REMOTE |
LNET_MD_TRUNCATE; /* allow to make EOVERFLOW error */
reply_md.user_ptr = &request->rq_reply_cbid;
reply_md.eq_handle = ptlrpc_eq_h;
/* We must see the unlink callback to unset rq_reply_unlink,
so we can't auto-unlink */
rc = LNetMDAttach(reply_me_h, reply_md, LNET_RETAIN,
&request->rq_reply_md_h);
if (rc != 0) {
CERROR("LNetMDAttach failed: %d\n", rc);
LASSERT(rc == -ENOMEM);
spin_lock(&request->rq_lock);
/* ...but the MD attach didn't succeed... */
request->rq_receiving_reply = 0;
spin_unlock(&request->rq_lock);
rc = -ENOMEM;
goto cleanup_me;
}
CDEBUG(D_NET, "Setup reply buffer: %u bytes, xid %llu, portal %u\n",
request->rq_repbuf_len, request->rq_xid,
request->rq_reply_portal);
}
/* add references on request for request_out_callback */
ptlrpc_request_addref(request);
if (obd != NULL && obd->obd_svc_stats != NULL)
lprocfs_counter_add(obd->obd_svc_stats, PTLRPC_REQACTIVE_CNTR,
atomic_read(&request->rq_import->imp_inflight));
OBD_FAIL_TIMEOUT(OBD_FAIL_PTLRPC_DELAY_SEND, request->rq_timeout + 5);
ktime_get_real_ts64(&request->rq_arrival_time);
request->rq_sent = ktime_get_real_seconds();
/* We give the server rq_timeout secs to process the req, and
add the network latency for our local timeout. */
request->rq_deadline = request->rq_sent + request->rq_timeout +
ptlrpc_at_get_net_latency(request);
ptlrpc_pinger_sending_on_import(request->rq_import);
DEBUG_REQ(D_INFO, request, "send flg=%x",
lustre_msg_get_flags(request->rq_reqmsg));
rc = ptl_send_buf(&request->rq_req_md_h,
request->rq_reqbuf, request->rq_reqdata_len,
LNET_NOACK_REQ, &request->rq_req_cbid,
connection,
request->rq_request_portal,
request->rq_xid, 0);
if (rc == 0)
goto out;
ptlrpc_req_finished(request);
if (noreply)
goto out;
cleanup_me:
/* MEUnlink is safe; the PUT didn't even get off the ground, and
* nobody apart from the PUT's target has the right nid+XID to
* access the reply buffer. */
rc2 = LNetMEUnlink(reply_me_h);
LASSERT(rc2 == 0);
/* UNLINKED callback called synchronously */
LASSERT(!request->rq_receiving_reply);
cleanup_bulk:
/* We do sync unlink here as there was no real transfer here so
* the chance to have long unlink to sluggish net is smaller here. */
ptlrpc_unregister_bulk(request, 0);
out:
if (request->rq_memalloc)
cfs_memory_pressure_restore(mpflag);
return rc;
}
static int seq_client_rpc(struct lu_client_seq *seq,
struct lu_seq_range *output, __u32 opc,
const char *opcname)
{
struct obd_export *exp = seq->lcs_exp;
struct ptlrpc_request *req;
struct lu_seq_range *out, *in;
__u32 *op;
unsigned int debug_mask;
int rc;
ENTRY;
LASSERT(exp != NULL && !IS_ERR(exp));
req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_SEQ_QUERY,
LUSTRE_MDS_VERSION, SEQ_QUERY);
if (req == NULL)
RETURN(-ENOMEM);
/* Init operation code */
op = req_capsule_client_get(&req->rq_pill, &RMF_SEQ_OPC);
*op = opc;
/* Zero out input range, this is not recovery yet. */
in = req_capsule_client_get(&req->rq_pill, &RMF_SEQ_RANGE);
lu_seq_range_init(in);
ptlrpc_request_set_replen(req);
in->lsr_index = seq->lcs_space.lsr_index;
if (seq->lcs_type == LUSTRE_SEQ_METADATA)
fld_range_set_mdt(in);
else
fld_range_set_ost(in);
if (opc == SEQ_ALLOC_SUPER) {
req->rq_request_portal = SEQ_CONTROLLER_PORTAL;
req->rq_reply_portal = MDC_REPLY_PORTAL;
/* During allocating super sequence for data object,
* the current thread might hold the export of MDT0(MDT0
* precreating objects on this OST), and it will send the
* request to MDT0 here, so we can not keep resending the
* request here, otherwise if MDT0 is failed(umounted),
* it can not release the export of MDT0 */
if (seq->lcs_type == LUSTRE_SEQ_DATA)
req->rq_no_delay = req->rq_no_resend = 1;
debug_mask = D_CONSOLE;
} else {
if (seq->lcs_type == LUSTRE_SEQ_METADATA) {
req->rq_reply_portal = MDC_REPLY_PORTAL;
req->rq_request_portal = SEQ_METADATA_PORTAL;
} else {
req->rq_reply_portal = OSC_REPLY_PORTAL;
req->rq_request_portal = SEQ_DATA_PORTAL;
}
debug_mask = D_INFO;
}
/* Allow seq client RPC during recovery time. */
req->rq_allow_replay = 1;
ptlrpc_at_set_req_timeout(req);
rc = ptlrpc_queue_wait(req);
if (rc)
GOTO(out_req, rc);
out = req_capsule_server_get(&req->rq_pill, &RMF_SEQ_RANGE);
*output = *out;
if (!lu_seq_range_is_sane(output)) {
CERROR("%s: Invalid range received from server: "
DRANGE"\n", seq->lcs_name, PRANGE(output));
GOTO(out_req, rc = -EINVAL);
}
if (lu_seq_range_is_exhausted(output)) {
CERROR("%s: Range received from server is exhausted: "
DRANGE"]\n", seq->lcs_name, PRANGE(output));
GOTO(out_req, rc = -EINVAL);
}
CDEBUG_LIMIT(debug_mask, "%s: Allocated %s-sequence "DRANGE"]\n",
seq->lcs_name, opcname, PRANGE(output));
EXIT;
out_req:
ptlrpc_req_finished(req);
return rc;
}
ssize_t ll_listxattr(struct dentry *dentry, char *buffer, size_t size)
{
struct inode *inode = dentry->d_inode;
int rc = 0, rc2 = 0;
struct lov_mds_md *lmm = NULL;
struct ptlrpc_request *request = NULL;
int lmmsize;
LASSERT(inode);
CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
inode->i_ino, inode->i_generation, inode);
ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LISTXATTR, 1);
rc = ll_getxattr_common(inode, NULL, buffer, size, OBD_MD_FLXATTRLS);
if (rc < 0)
GOTO(out, rc);
if (buffer != NULL) {
struct ll_sb_info *sbi = ll_i2sbi(inode);
char *xattr_name = buffer;
int xlen, rem = rc;
while (rem > 0) {
xlen = strnlen(xattr_name, rem - 1) + 1;
rem -= xlen;
if (xattr_type_filter(sbi,
get_xattr_type(xattr_name)) == 0) {
/* skip OK xattr type
* leave it in buffer
*/
xattr_name += xlen;
continue;
}
/* move up remaining xattrs in buffer
* removing the xattr that is not OK
*/
memmove(xattr_name, xattr_name + xlen, rem);
rc -= xlen;
}
}
if (S_ISREG(inode->i_mode)) {
if (!ll_i2info(inode)->lli_has_smd)
rc2 = -1;
} else if (S_ISDIR(inode->i_mode)) {
rc2 = ll_dir_getstripe(inode, &lmm, &lmmsize, &request);
}
if (rc2 < 0) {
GOTO(out, rc2 = 0);
} else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)) {
const int prefix_len = sizeof(XATTR_LUSTRE_PREFIX) - 1;
const size_t name_len = sizeof("lov") - 1;
const size_t total_len = prefix_len + name_len + 1;
if (((rc + total_len) > size) && (buffer != NULL)) {
ptlrpc_req_finished(request);
return -ERANGE;
}
if (buffer != NULL) {
buffer += rc;
memcpy(buffer, XATTR_LUSTRE_PREFIX, prefix_len);
memcpy(buffer + prefix_len, "lov", name_len);
buffer[prefix_len + name_len] = '\0';
}
rc2 = total_len;
}
out:
ptlrpc_req_finished(request);
rc = rc + rc2;
return rc;
}
ssize_t ll_getxattr(struct dentry *dentry, const char *name,
void *buffer, size_t size)
{
struct inode *inode = dentry->d_inode;
LASSERT(inode);
LASSERT(name);
CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), xattr %s\n",
inode->i_ino, inode->i_generation, inode, name);
ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_GETXATTR, 1);
if ((strncmp(name, XATTR_TRUSTED_PREFIX,
sizeof(XATTR_TRUSTED_PREFIX) - 1) == 0 &&
strcmp(name + sizeof(XATTR_TRUSTED_PREFIX) - 1, "lov") == 0) ||
(strncmp(name, XATTR_LUSTRE_PREFIX,
sizeof(XATTR_LUSTRE_PREFIX) - 1) == 0 &&
strcmp(name + sizeof(XATTR_LUSTRE_PREFIX) - 1, "lov") == 0)) {
struct lov_stripe_md *lsm;
struct lov_user_md *lump;
struct lov_mds_md *lmm = NULL;
struct ptlrpc_request *request = NULL;
int rc = 0, lmmsize = 0;
if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
return -ENODATA;
if (size == 0 && S_ISDIR(inode->i_mode)) {
/* XXX directory EA is fix for now, optimize to save
* RPC transfer */
GOTO(out, rc = sizeof(struct lov_user_md));
}
lsm = ccc_inode_lsm_get(inode);
if (lsm == NULL) {
if (S_ISDIR(inode->i_mode)) {
rc = ll_dir_getstripe(inode, &lmm,
&lmmsize, &request);
} else {
rc = -ENODATA;
}
} else {
/* LSM is present already after lookup/getattr call.
* we need to grab layout lock once it is implemented */
rc = obd_packmd(ll_i2dtexp(inode), &lmm, lsm);
lmmsize = rc;
}
ccc_inode_lsm_put(inode, lsm);
if (rc < 0)
GOTO(out, rc);
if (size == 0) {
/* used to call ll_get_max_mdsize() forward to get
* the maximum buffer size, while some apps (such as
* rsync 3.0.x) care much about the exact xattr value
* size */
rc = lmmsize;
GOTO(out, rc);
}
if (size < lmmsize) {
CERROR("server bug: replied size %d > %d for %s (%s)\n",
lmmsize, (int)size, dentry->d_name.name, name);
GOTO(out, rc = -ERANGE);
}
lump = (struct lov_user_md *)buffer;
memcpy(lump, lmm, lmmsize);
/* do not return layout gen for getxattr otherwise it would
* confuse tar --xattr by recognizing layout gen as stripe
* offset when the file is restored. See LU-2809. */
lump->lmm_layout_gen = 0;
rc = lmmsize;
out:
if (request)
ptlrpc_req_finished(request);
else if (lmm)
obd_free_diskmd(ll_i2dtexp(inode), &lmm);
return(rc);
}
return ll_getxattr_common(inode, name, buffer, size, OBD_MD_FLXATTR);
}
static
int ll_getxattr_common(struct inode *inode, const char *name,
void *buffer, size_t size, __u64 valid)
{
struct ll_sb_info *sbi = ll_i2sbi(inode);
struct ptlrpc_request *req = NULL;
struct mdt_body *body;
int xattr_type, rc;
void *xdata;
struct obd_capa *oc;
struct rmtacl_ctl_entry *rce = NULL;
CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
inode->i_ino, inode->i_generation, inode);
/* listxattr have slightly different behavior from of ext3:
* without 'user_xattr' ext3 will list all xattr names but
* filtered out "^user..*"; we list them all for simplicity.
*/
if (!name) {
xattr_type = XATTR_OTHER_T;
goto do_getxattr;
}
xattr_type = get_xattr_type(name);
rc = xattr_type_filter(sbi, xattr_type);
if (rc)
return rc;
/* b15587: ignore security.capability xattr for now */
if ((xattr_type == XATTR_SECURITY_T &&
strcmp(name, "security.capability") == 0))
return -ENODATA;
/* LU-549: Disable security.selinux when selinux is disabled */
if (xattr_type == XATTR_SECURITY_T && !selinux_is_enabled() &&
strcmp(name, "security.selinux") == 0)
return -EOPNOTSUPP;
#ifdef CONFIG_FS_POSIX_ACL
if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
(xattr_type == XATTR_ACL_ACCESS_T ||
xattr_type == XATTR_ACL_DEFAULT_T)) {
rce = rct_search(&sbi->ll_rct, current_pid());
if (rce == NULL ||
(rce->rce_ops != RMT_LSETFACL &&
rce->rce_ops != RMT_LGETFACL &&
rce->rce_ops != RMT_RSETFACL &&
rce->rce_ops != RMT_RGETFACL))
return -EOPNOTSUPP;
}
/* posix acl is under protection of LOOKUP lock. when calling to this,
* we just have path resolution to the target inode, so we have great
* chance that cached ACL is uptodate.
*/
if (xattr_type == XATTR_ACL_ACCESS_T &&
!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
struct ll_inode_info *lli = ll_i2info(inode);
struct posix_acl *acl;
spin_lock(&lli->lli_lock);
acl = posix_acl_dup(lli->lli_posix_acl);
spin_unlock(&lli->lli_lock);
if (!acl)
return -ENODATA;
rc = posix_acl_to_xattr(&init_user_ns, acl, buffer, size);
posix_acl_release(acl);
return rc;
}
if (xattr_type == XATTR_ACL_DEFAULT_T && !S_ISDIR(inode->i_mode))
return -ENODATA;
#endif
do_getxattr:
if (sbi->ll_xattr_cache_enabled && (rce == NULL ||
rce->rce_ops == RMT_LGETFACL ||
rce->rce_ops == RMT_LSETFACL)) {
rc = ll_xattr_cache_get(inode, name, buffer, size, valid);
if (rc < 0)
GOTO(out_xattr, rc);
} else {
oc = ll_mdscapa_get(inode);
rc = md_getxattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
valid | (rce ? rce_ops2valid(rce->rce_ops) : 0),
name, NULL, 0, size, 0, &req);
capa_put(oc);
if (rc < 0)
GOTO(out_xattr, rc);
body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
LASSERT(body);
/* only detect the xattr size */
if (size == 0)
GOTO(out, rc = body->eadatasize);
if (size < body->eadatasize) {
CERROR("server bug: replied size %u > %u\n",
body->eadatasize, (int)size);
GOTO(out, rc = -ERANGE);
}
if (body->eadatasize == 0)
GOTO(out, rc = -ENODATA);
/* do not need swab xattr data */
xdata = req_capsule_server_sized_get(&req->rq_pill, &RMF_EADATA,
body->eadatasize);
if (!xdata)
GOTO(out, rc = -EFAULT);
memcpy(buffer, xdata, body->eadatasize);
rc = body->eadatasize;
}
#ifdef CONFIG_FS_POSIX_ACL
if (rce && rce->rce_ops == RMT_LSETFACL) {
ext_acl_xattr_header *acl;
acl = lustre_posix_acl_xattr_2ext(
(posix_acl_xattr_header *)buffer, rc);
if (IS_ERR(acl))
GOTO(out, rc = PTR_ERR(acl));
rc = ee_add(&sbi->ll_et, current_pid(), ll_inode2fid(inode),
xattr_type, acl);
if (unlikely(rc < 0)) {
lustre_ext_acl_xattr_free(acl);
GOTO(out, rc);
}
}
#endif
out_xattr:
if (rc == -EOPNOTSUPP && xattr_type == XATTR_USER_T) {
LCONSOLE_INFO(
"%s: disabling user_xattr feature because it is not supported on the server: rc = %d\n",
ll_get_fsname(inode->i_sb, NULL, 0), rc);
sbi->ll_flags &= ~LL_SBI_USER_XATTR;
}
out:
ptlrpc_req_finished(req);
return rc;
}
static
int ll_setxattr_common(struct inode *inode, const char *name,
const void *value, size_t size,
int flags, __u64 valid)
{
struct ll_sb_info *sbi = ll_i2sbi(inode);
struct ptlrpc_request *req = NULL;
int xattr_type, rc;
struct obd_capa *oc;
struct rmtacl_ctl_entry *rce = NULL;
#ifdef CONFIG_FS_POSIX_ACL
posix_acl_xattr_header *new_value = NULL;
ext_acl_xattr_header *acl = NULL;
#endif
const char *pv = value;
xattr_type = get_xattr_type(name);
rc = xattr_type_filter(sbi, xattr_type);
if (rc)
return rc;
/* b10667: ignore lustre special xattr for now */
if ((xattr_type == XATTR_TRUSTED_T && strcmp(name, "trusted.lov") == 0) ||
(xattr_type == XATTR_LUSTRE_T && strcmp(name, "lustre.lov") == 0))
return 0;
/* b15587: ignore security.capability xattr for now */
if ((xattr_type == XATTR_SECURITY_T &&
strcmp(name, "security.capability") == 0))
return 0;
/* LU-549: Disable security.selinux when selinux is disabled */
if (xattr_type == XATTR_SECURITY_T && !selinux_is_enabled() &&
strcmp(name, "security.selinux") == 0)
return -EOPNOTSUPP;
#ifdef CONFIG_FS_POSIX_ACL
if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
(xattr_type == XATTR_ACL_ACCESS_T ||
xattr_type == XATTR_ACL_DEFAULT_T)) {
rce = rct_search(&sbi->ll_rct, current_pid());
if (rce == NULL ||
(rce->rce_ops != RMT_LSETFACL &&
rce->rce_ops != RMT_RSETFACL))
return -EOPNOTSUPP;
if (rce->rce_ops == RMT_LSETFACL) {
struct eacl_entry *ee;
ee = et_search_del(&sbi->ll_et, current_pid(),
ll_inode2fid(inode), xattr_type);
LASSERT(ee != NULL);
if (valid & OBD_MD_FLXATTR) {
acl = lustre_acl_xattr_merge2ext(
(posix_acl_xattr_header *)value,
size, ee->ee_acl);
if (IS_ERR(acl)) {
ee_free(ee);
return PTR_ERR(acl);
}
size = CFS_ACL_XATTR_SIZE(\
le32_to_cpu(acl->a_count), \
ext_acl_xattr);
pv = (const char *)acl;
}
ee_free(ee);
} else if (rce->rce_ops == RMT_RSETFACL) {
size = lustre_posix_acl_xattr_filter(
(posix_acl_xattr_header *)value,
size, &new_value);
if (unlikely(size < 0))
return size;
pv = (const char *)new_value;
} else
return -EOPNOTSUPP;
valid |= rce_ops2valid(rce->rce_ops);
}
#endif
if (sbi->ll_xattr_cache_enabled &&
(rce == NULL || rce->rce_ops == RMT_LSETFACL)) {
rc = ll_xattr_cache_update(inode, name, pv, size, valid, flags);
} else {
oc = ll_mdscapa_get(inode);
rc = md_setxattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
valid, name, pv, size, 0, flags,
ll_i2suppgid(inode), &req);
capa_put(oc);
}
#ifdef CONFIG_FS_POSIX_ACL
if (new_value != NULL)
lustre_posix_acl_xattr_free(new_value, size);
if (acl != NULL)
lustre_ext_acl_xattr_free(acl);
#endif
if (rc) {
if (rc == -EOPNOTSUPP && xattr_type == XATTR_USER_T) {
LCONSOLE_INFO("Disabling user_xattr feature because "
"it is not supported on the server\n");
sbi->ll_flags &= ~LL_SBI_USER_XATTR;
}
return rc;
}
ptlrpc_req_finished(req);
return 0;
}
int ll_revalidate_nd(struct dentry *dentry, struct nameidata *nd)
{
int rc;
ENTRY;
if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
struct lookup_intent *it;
it = ll_convert_intent(&nd->intent.open, nd->flags);
if (IS_ERR(it))
RETURN(0);
if (it->it_op == (IT_OPEN|IT_CREAT))
if (nd->intent.open.flags & O_EXCL) {
CDEBUG(D_VFSTRACE, "create O_EXCL, returning 0\n");
rc = 0;
goto out_it;
}
rc = ll_revalidate_it(dentry, nd->flags, it);
if (rc && (nd->flags & LOOKUP_OPEN) &&
it_disposition(it, DISP_OPEN_OPEN)) {/*Open*/
#ifdef HAVE_FILE_IN_STRUCT_INTENT
// XXX Code duplication with ll_lookup_nd
if (S_ISFIFO(dentry->d_inode->i_mode)) {
// We cannot call open here as it would
// deadlock.
ptlrpc_req_finished(
(struct ptlrpc_request *)
it->d.lustre.it_data);
} else {
#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
/* 2.6.1[456] have a bug in open_namei() that forgets to check
* nd->intent.open.file for error, so we need to return it as lookup's result
* instead */
struct file *filp;
nd->intent.open.file->private_data = it;
filp = lookup_instantiate_filp(nd, dentry,NULL);
if (IS_ERR(filp)) {
rc = PTR_ERR(filp);
}
#else
nd->intent.open.file->private_data = it;
(void)lookup_instantiate_filp(nd, dentry,NULL);
#endif
}
#else
ll_release_openhandle(dentry, it);
#endif /* HAVE_FILE_IN_STRUCT_INTENT */
}
if (!rc && (nd->flags & LOOKUP_CREATE) &&
it_disposition(it, DISP_OPEN_CREATE)) {
/* We created something but we may only return
* negative dentry here, so save request in dentry,
* if lookup will be called later on, it will
* pick the request, otherwise it would be freed
* with dentry */
ll_d2d(dentry)->lld_it = it;
it = NULL; /* avoid freeing */
}
out_it:
if (it) {
ll_intent_release(it);
OBD_FREE(it, sizeof(*it));
}
} else {
rc = ll_revalidate_it(dentry, 0, NULL);
}
RETURN(rc);
}
int ll_revalidate_it(struct dentry *de, int lookup_flags,
struct lookup_intent *it)
{
struct md_op_data *op_data;
struct ptlrpc_request *req = NULL;
struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
struct obd_export *exp;
struct inode *parent = de->d_parent->d_inode;
int rc, first = 0;
ENTRY;
CDEBUG(D_VFSTRACE, "VFS Op:name=%s,intent=%s\n", de->d_name.name,
LL_IT2STR(it));
if (de->d_inode == NULL) {
/* We can only use negative dentries if this is stat or lookup,
for opens and stuff we do need to query server. */
/* If there is IT_CREAT in intent op set, then we must throw
away this negative dentry and actually do the request to
kernel to create whatever needs to be created (if possible)*/
if (it && (it->it_op & IT_CREAT))
RETURN(0);
if (de->d_flags & DCACHE_LUSTRE_INVALID)
RETURN(0);
rc = ll_have_md_lock(parent, MDS_INODELOCK_UPDATE, LCK_MINMODE);
GOTO(out_sa, rc);
}
/* Never execute intents for mount points.
* Attributes will be fixed up in ll_inode_revalidate_it */
if (d_mountpoint(de))
GOTO(out_sa, rc = 1);
/* need to get attributes in case root got changed from other client */
if (de == de->d_sb->s_root) {
rc = __ll_inode_revalidate_it(de, it, MDS_INODELOCK_LOOKUP);
if (rc == 0)
rc = 1;
GOTO(out_sa, rc);
}
exp = ll_i2mdexp(de->d_inode);
OBD_FAIL_TIMEOUT(OBD_FAIL_MDC_REVALIDATE_PAUSE, 5);
ll_frob_intent(&it, &lookup_it);
LASSERT(it);
if (it->it_op == IT_LOOKUP && !(de->d_flags & DCACHE_LUSTRE_INVALID))
GOTO(out_sa, rc = 1);
op_data = ll_prep_md_op_data(NULL, parent, de->d_inode,
de->d_name.name, de->d_name.len,
0, LUSTRE_OPC_ANY, NULL);
if (IS_ERR(op_data))
RETURN(PTR_ERR(op_data));
if ((it->it_op == IT_OPEN) && de->d_inode) {
struct inode *inode = de->d_inode;
struct ll_inode_info *lli = ll_i2info(inode);
struct obd_client_handle **och_p;
__u64 *och_usecount;
/*
* We used to check for MDS_INODELOCK_OPEN here, but in fact
* just having LOOKUP lock is enough to justify inode is the
* same. And if inode is the same and we have suitable
* openhandle, then there is no point in doing another OPEN RPC
* just to throw away newly received openhandle. There are no
* security implications too, if file owner or access mode is
* change, LOOKUP lock is revoked.
*/
if (it->it_flags & FMODE_WRITE) {
och_p = &lli->lli_mds_write_och;
och_usecount = &lli->lli_open_fd_write_count;
} else if (it->it_flags & FMODE_EXEC) {
och_p = &lli->lli_mds_exec_och;
och_usecount = &lli->lli_open_fd_exec_count;
} else {
och_p = &lli->lli_mds_read_och;
och_usecount = &lli->lli_open_fd_read_count;
}
/* Check for the proper lock. */
if (!ll_have_md_lock(inode, MDS_INODELOCK_LOOKUP, LCK_MINMODE))
goto do_lock;
cfs_down(&lli->lli_och_sem);
if (*och_p) { /* Everything is open already, do nothing */
/*(*och_usecount)++; Do not let them steal our open
handle from under us */
/* XXX The code above was my original idea, but in case
we have the handle, but we cannot use it due to later
checks (e.g. O_CREAT|O_EXCL flags set), nobody
would decrement counter increased here. So we just
hope the lock won't be invalidated in between. But
if it would be, we'll reopen the open request to
MDS later during file open path */
cfs_up(&lli->lli_och_sem);
ll_finish_md_op_data(op_data);
RETURN(1);
} else {
cfs_up(&lli->lli_och_sem);
}
}
if (it->it_op == IT_GETATTR) {
first = ll_statahead_enter(parent, &de, 0);
if (first == 1) {
ll_statahead_exit(parent, de, 1);
ll_finish_md_op_data(op_data);
GOTO(out, rc = 1);
}
}
do_lock:
it->it_create_mode &= ~current->fs->umask;
it->it_create_mode |= M_CHECK_STALE;
rc = md_intent_lock(exp, op_data, NULL, 0, it,
lookup_flags,
&req, ll_md_blocking_ast, 0);
it->it_create_mode &= ~M_CHECK_STALE;
ll_finish_md_op_data(op_data);
if (it->it_op == IT_GETATTR && !first)
/* If there are too many locks on client-side, then some
* locks taken by statahead maybe dropped automatically
* before the real "revalidate" using them. */
ll_statahead_exit(parent, de, req == NULL ? rc : 0);
else if (first == -EEXIST)
ll_statahead_mark(parent, de);
/* If req is NULL, then md_intent_lock only tried to do a lock match;
* if all was well, it will return 1 if it found locks, 0 otherwise. */
if (req == NULL && rc >= 0) {
if (!rc)
goto do_lookup;
GOTO(out, rc);
}
if (rc < 0) {
if (rc != -ESTALE) {
CDEBUG(D_INFO, "ll_intent_lock: rc %d : it->it_status "
"%d\n", rc, it->d.lustre.it_status);
}
GOTO(out, rc = 0);
}
revalidate_finish:
rc = ll_revalidate_it_finish(req, it, de);
if (rc != 0) {
if (rc != -ESTALE && rc != -ENOENT)
ll_intent_release(it);
GOTO(out, rc = 0);
}
if ((it->it_op & IT_OPEN) && de->d_inode &&
!S_ISREG(de->d_inode->i_mode) &&
!S_ISDIR(de->d_inode->i_mode)) {
ll_release_openhandle(de, it);
}
rc = 1;
/* unfortunately ll_intent_lock may cause a callback and revoke our
* dentry */
cfs_spin_lock(&ll_lookup_lock);
spin_lock(&dcache_lock);
lock_dentry(de);
__d_drop(de);
unlock_dentry(de);
d_rehash_cond(de, 0);
spin_unlock(&dcache_lock);
cfs_spin_unlock(&ll_lookup_lock);
out:
/* We do not free request as it may be reused during following lookup
* (see comment in mdc/mdc_locks.c::mdc_intent_lock()), request will
* be freed in ll_lookup_it or in ll_intent_release. But if
* request was not completed, we need to free it. (bug 5154, 9903) */
if (req != NULL && !it_disposition(it, DISP_ENQ_COMPLETE))
ptlrpc_req_finished(req);
if (rc == 0) {
ll_unhash_aliases(de->d_inode);
/* done in ll_unhash_aliases()
dentry->d_flags |= DCACHE_LUSTRE_INVALID; */
} else {
CDEBUG(D_DENTRY, "revalidated dentry %.*s (%p) parent %p "
"inode %p refc %d\n", de->d_name.len,
de->d_name.name, de, de->d_parent, de->d_inode,
atomic_read(&de->d_count));
if (first != 1) {
if (de->d_flags & DCACHE_LUSTRE_INVALID) {
lock_dentry(de);
de->d_flags &= ~DCACHE_LUSTRE_INVALID;
unlock_dentry(de);
}
ll_lookup_finish_locks(it, de);
}
}
RETURN(rc);
/*
* This part is here to combat evil-evil race in real_lookup on 2.6
* kernels. The race details are: We enter do_lookup() looking for some
* name, there is nothing in dcache for this name yet and d_lookup()
* returns NULL. We proceed to real_lookup(), and while we do this,
* another process does open on the same file we looking up (most simple
* reproducer), open succeeds and the dentry is added. Now back to
* us. In real_lookup() we do d_lookup() again and suddenly find the
* dentry, so we call d_revalidate on it, but there is no lock, so
* without this code we would return 0, but unpatched real_lookup just
* returns -ENOENT in such a case instead of retrying the lookup. Once
* this is dealt with in real_lookup(), all of this ugly mess can go and
* we can just check locks in ->d_revalidate without doing any RPCs
* ever.
*/
do_lookup:
if (it != &lookup_it) {
/* MDS_INODELOCK_UPDATE needed for IT_GETATTR case. */
if (it->it_op == IT_GETATTR)
lookup_it.it_op = IT_GETATTR;
ll_lookup_finish_locks(it, de);
it = &lookup_it;
}
/* Do real lookup here. */
op_data = ll_prep_md_op_data(NULL, parent, NULL, de->d_name.name,
de->d_name.len, 0, (it->it_op & IT_CREAT ?
LUSTRE_OPC_CREATE :
LUSTRE_OPC_ANY), NULL);
if (IS_ERR(op_data))
RETURN(PTR_ERR(op_data));
rc = md_intent_lock(exp, op_data, NULL, 0, it, 0, &req,
ll_md_blocking_ast, 0);
if (rc >= 0) {
struct mdt_body *mdt_body;
struct lu_fid fid = {.f_seq = 0, .f_oid = 0, .f_ver = 0};
mdt_body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
if (de->d_inode)
fid = *ll_inode2fid(de->d_inode);
/* see if we got same inode, if not - return error */
if (lu_fid_eq(&fid, &mdt_body->fid1)) {
ll_finish_md_op_data(op_data);
op_data = NULL;
goto revalidate_finish;
}
ll_intent_release(it);
}
ll_finish_md_op_data(op_data);
GOTO(out, rc = 0);
out_sa:
/*
* For rc == 1 case, should not return directly to prevent losing
* statahead windows; for rc == 0 case, the "lookup" will be done later.
*/
if (it && it->it_op == IT_GETATTR && rc == 1) {
first = ll_statahead_enter(parent, &de, 0);
if (first >= 0)
ll_statahead_exit(parent, de, 1);
else if (first == -EEXIST)
ll_statahead_mark(parent, de);
}
return rc;
}
#if 0
static void ll_pin(struct dentry *de, struct vfsmount *mnt, int flag)
{
struct inode *inode= de->d_inode;
struct ll_sb_info *sbi = ll_i2sbi(inode);
struct ll_dentry_data *ldd = ll_d2d(de);
struct obd_client_handle *handle;
struct obd_capa *oc;
int rc = 0;
ENTRY;
LASSERT(ldd);
cfs_lock_kernel();
/* Strictly speaking this introduces an additional race: the
* increments should wait until the rpc has returned.
* However, given that at present the function is void, this
* issue is moot. */
if (flag == 1 && (++ldd->lld_mnt_count) > 1) {
cfs_unlock_kernel();
EXIT;
return;
}
if (flag == 0 && (++ldd->lld_cwd_count) > 1) {
cfs_unlock_kernel();
EXIT;
return;
}
cfs_unlock_kernel();
handle = (flag) ? &ldd->lld_mnt_och : &ldd->lld_cwd_och;
oc = ll_mdscapa_get(inode);
rc = obd_pin(sbi->ll_md_exp, ll_inode2fid(inode), oc, handle, flag);
capa_put(oc);
if (rc) {
cfs_lock_kernel();
memset(handle, 0, sizeof(*handle));
if (flag == 0)
ldd->lld_cwd_count--;
else
ldd->lld_mnt_count--;
cfs_unlock_kernel();
}
EXIT;
return;
}
static void ll_unpin(struct dentry *de, struct vfsmount *mnt, int flag)
{
struct ll_sb_info *sbi = ll_i2sbi(de->d_inode);
struct ll_dentry_data *ldd = ll_d2d(de);
struct obd_client_handle handle;
int count, rc = 0;
ENTRY;
LASSERT(ldd);
cfs_lock_kernel();
/* Strictly speaking this introduces an additional race: the
* increments should wait until the rpc has returned.
* However, given that at present the function is void, this
* issue is moot. */
handle = (flag) ? ldd->lld_mnt_och : ldd->lld_cwd_och;
if (handle.och_magic != OBD_CLIENT_HANDLE_MAGIC) {
/* the "pin" failed */
cfs_unlock_kernel();
EXIT;
return;
}
if (flag)
count = --ldd->lld_mnt_count;
else
count = --ldd->lld_cwd_count;
cfs_unlock_kernel();
if (count != 0) {
EXIT;
return;
}
rc = obd_unpin(sbi->ll_md_exp, &handle, flag);
EXIT;
return;
}
#endif
#ifdef HAVE_VFS_INTENT_PATCHES
int ll_revalidate_nd(struct dentry *dentry, struct nameidata *nd)
{
int rc;
ENTRY;
if (nd && nd->flags & LOOKUP_LAST && !(nd->flags & LOOKUP_LINK_NOTLAST))
rc = ll_revalidate_it(dentry, nd->flags, &nd->intent);
else
rc = ll_revalidate_it(dentry, 0, NULL);
RETURN(rc);
}
int lustre_check_remote_perm(struct inode *inode, int mask)
{
struct ll_inode_info *lli = ll_i2info(inode);
struct ll_sb_info *sbi = ll_i2sbi(inode);
struct ptlrpc_request *req = NULL;
struct mdt_remote_perm *perm;
struct obd_capa *oc;
unsigned long save;
int i = 0, rc;
do {
save = lli->lli_rmtperm_time;
rc = do_check_remote_perm(lli, mask);
if (!rc || (rc != -ENOENT && i))
break;
might_sleep();
mutex_lock(&lli->lli_rmtperm_mutex);
/* check again */
if (save != lli->lli_rmtperm_time) {
rc = do_check_remote_perm(lli, mask);
if (!rc || (rc != -ENOENT && i)) {
mutex_unlock(&lli->lli_rmtperm_mutex);
break;
}
}
if (i++ > 5) {
CERROR("check remote perm falls in dead loop!\n");
LBUG();
}
oc = ll_mdscapa_get(inode);
rc = md_get_remote_perm(sbi->ll_md_exp, ll_inode2fid(inode), oc,
ll_i2suppgid(inode), &req);
capa_put(oc);
if (rc) {
mutex_unlock(&lli->lli_rmtperm_mutex);
break;
}
perm = req_capsule_server_swab_get(&req->rq_pill, &RMF_ACL,
lustre_swab_mdt_remote_perm);
if (unlikely(perm == NULL)) {
mutex_unlock(&lli->lli_rmtperm_mutex);
rc = -EPROTO;
break;
}
rc = ll_update_remote_perm(inode, perm);
mutex_unlock(&lli->lli_rmtperm_mutex);
if (rc == -ENOMEM)
break;
ptlrpc_req_finished(req);
req = NULL;
} while (1);
ptlrpc_req_finished(req);
return rc;
}
/**
* asks OST to clean precreate orphans
* and gets next id for new objects
*/
static int osp_precreate_cleanup_orphans(struct lu_env *env,
struct osp_device *d)
{
struct osp_thread_info *osi = osp_env_info(env);
struct lu_fid *last_fid = &osi->osi_fid;
struct ptlrpc_request *req = NULL;
struct obd_import *imp;
struct ost_body *body;
struct l_wait_info lwi = { 0 };
int update_status = 0;
int rc;
int diff;
ENTRY;
/*
* wait for local recovery to finish, so we can cleanup orphans
* orphans are all objects since "last used" (assigned), but
* there might be objects reserved and in some cases they won't
* be used. we can't cleanup them till we're sure they won't be
* used. also can't we allow new reservations because they may
* end up getting orphans being cleaned up below. so we block
* new reservations and wait till all reserved objects either
* user or released.
*/
spin_lock(&d->opd_pre_lock);
d->opd_pre_recovering = 1;
spin_unlock(&d->opd_pre_lock);
/*
* The locking above makes sure the opd_pre_reserved check below will
* catch all osp_precreate_reserve() calls who find
* "!opd_pre_recovering".
*/
l_wait_event(d->opd_pre_waitq,
(!d->opd_pre_reserved && d->opd_recovery_completed) ||
!osp_precreate_running(d) || d->opd_got_disconnected,
&lwi);
if (!osp_precreate_running(d) || d->opd_got_disconnected)
GOTO(out, rc = -EAGAIN);
CDEBUG(D_HA, "%s: going to cleanup orphans since "DFID"\n",
d->opd_obd->obd_name, PFID(&d->opd_last_used_fid));
*last_fid = d->opd_last_used_fid;
/* The OSP should already get the valid seq now */
LASSERT(!fid_is_zero(last_fid));
if (fid_oid(&d->opd_last_used_fid) < 2) {
/* lastfid looks strange... ask OST */
rc = osp_get_lastfid_from_ost(env, d);
if (rc)
GOTO(out, rc);
}
imp = d->opd_obd->u.cli.cl_import;
LASSERT(imp);
req = ptlrpc_request_alloc(imp, &RQF_OST_CREATE);
if (req == NULL)
GOTO(out, rc = -ENOMEM);
rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_CREATE);
if (rc) {
ptlrpc_request_free(req);
req = NULL;
GOTO(out, rc);
}
body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
if (body == NULL)
GOTO(out, rc = -EPROTO);
body->oa.o_flags = OBD_FL_DELORPHAN;
body->oa.o_valid = OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
fid_to_ostid(&d->opd_last_used_fid, &body->oa.o_oi);
ptlrpc_request_set_replen(req);
/* Don't resend the delorphan req */
req->rq_no_resend = req->rq_no_delay = 1;
rc = ptlrpc_queue_wait(req);
if (rc) {
update_status = 1;
GOTO(out, rc);
}
body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
if (body == NULL)
GOTO(out, rc = -EPROTO);
/*
* OST provides us with id new pool starts from in body->oa.o_id
*/
ostid_to_fid(last_fid, &body->oa.o_oi, d->opd_index);
spin_lock(&d->opd_pre_lock);
diff = lu_fid_diff(&d->opd_last_used_fid, last_fid);
if (diff > 0) {
d->opd_pre_grow_count = OST_MIN_PRECREATE + diff;
d->opd_pre_last_created_fid = d->opd_last_used_fid;
} else {
d->opd_pre_grow_count = OST_MIN_PRECREATE;
d->opd_pre_last_created_fid = *last_fid;
}
/*
* This empties the pre-creation pool and effectively blocks any new
* reservations.
*/
LASSERT(fid_oid(&d->opd_pre_last_created_fid) <=
LUSTRE_DATA_SEQ_MAX_WIDTH);
d->opd_pre_used_fid = d->opd_pre_last_created_fid;
d->opd_pre_grow_slow = 0;
spin_unlock(&d->opd_pre_lock);
CDEBUG(D_HA, "%s: Got last_id "DFID" from OST, last_created "DFID
"last_used is "DFID"\n", d->opd_obd->obd_name, PFID(last_fid),
PFID(&d->opd_pre_last_created_fid), PFID(&d->opd_last_used_fid));
out:
if (req)
ptlrpc_req_finished(req);
spin_lock(&d->opd_pre_lock);
d->opd_pre_recovering = 0;
spin_unlock(&d->opd_pre_lock);
/*
* If rc is zero, the pre-creation window should have been emptied.
* Since waking up the herd would be useless without pre-created
* objects, we defer the signal to osp_precreate_send() in that case.
*/
if (rc != 0) {
if (update_status) {
CERROR("%s: cannot cleanup orphans: rc = %d\n",
d->opd_obd->obd_name, rc);
/* we can't proceed from here, OST seem to
* be in a bad shape, better to wait for
* a new instance of the server and repeat
* from the beginning. notify possible waiters
* this OSP isn't quite functional yet */
osp_pre_update_status(d, rc);
} else {
wake_up(&d->opd_pre_user_waitq);
}
}
RETURN(rc);
}
