/* return EADATA length to the caller. negative value means error */
static int mdt_getxattr_pack_reply(struct mdt_thread_info * info)
{
struct req_capsule *pill = info->mti_pill ;
struct ptlrpc_request *req = mdt_info_req(info);
char *xattr_name;
__u64 valid;
static const char user_string[] = "user.";
int size, rc;
ENTRY;
if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETXATTR_PACK))
RETURN(-ENOMEM);
valid = info->mti_body->mbo_valid & (OBD_MD_FLXATTR | OBD_MD_FLXATTRLS);
/* Determine how many bytes we need */
if (valid == OBD_MD_FLXATTR) {
xattr_name = req_capsule_client_get(pill, &RMF_NAME);
if (!xattr_name)
RETURN(-EFAULT);
if (!(exp_connect_flags(req->rq_export) & OBD_CONNECT_XATTR) &&
!strncmp(xattr_name, user_string, sizeof(user_string) - 1))
RETURN(-EOPNOTSUPP);
size = mo_xattr_get(info->mti_env,
mdt_object_child(info->mti_object),
&LU_BUF_NULL, xattr_name);
} else if (valid == OBD_MD_FLXATTRLS) {
size = mo_xattr_list(info->mti_env,
mdt_object_child(info->mti_object),
&LU_BUF_NULL);
} else if (valid == OBD_MD_FLXATTRALL) {
/* N.B. eadatasize = 0 is not valid for FLXATTRALL */
/* We could calculate accurate sizes, but this would
* introduce a lot of overhead, let's do it later... */
size = info->mti_body->mbo_eadatasize;
req_capsule_set_size(pill, &RMF_EAVALS, RCL_SERVER, size);
req_capsule_set_size(pill, &RMF_EAVALS_LENS, RCL_SERVER, size);
} else {
CDEBUG(D_INFO, "Valid bits: "LPX64"\n",
info->mti_body->mbo_valid);
RETURN(-EINVAL);
}
if (size == -ENODATA) {
size = 0;
} else if (size < 0) {
if (size != -EOPNOTSUPP)
CERROR("Error geting EA size: %d\n", size);
RETURN(size);
}
req_capsule_set_size(pill, &RMF_EADATA, RCL_SERVER,
info->mti_body->mbo_eadatasize == 0 ? 0 : size);
rc = req_capsule_server_pack(pill);
if (rc) {
LASSERT(rc < 0);
RETURN(rc);
}
RETURN(size);
}
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;
int rc;
ENTRY;
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);
range_init(in);
ptlrpc_request_set_replen(req);
if (seq->lcs_type == LUSTRE_SEQ_METADATA) {
req->rq_request_portal = SEQ_METADATA_PORTAL;
in->lsr_flags = LU_SEQ_RANGE_MDT;
} else {
LASSERTF(seq->lcs_type == LUSTRE_SEQ_DATA,
"unknown lcs_type %u\n", seq->lcs_type);
req->rq_request_portal = SEQ_DATA_PORTAL;
in->lsr_flags = LU_SEQ_RANGE_OST;
}
if (opc == SEQ_ALLOC_SUPER) {
/* Update index field of *in, it is required for
* FLD update on super sequence allocator node. */
in->lsr_index = seq->lcs_space.lsr_index;
req->rq_request_portal = SEQ_CONTROLLER_PORTAL;
} else {
LASSERTF(opc == SEQ_ALLOC_META,
"unknown opcode %u\n, opc", opc);
}
ptlrpc_at_set_req_timeout(req);
mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
rc = ptlrpc_queue_wait(req);
mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
if (rc)
GOTO(out_req, rc);
out = req_capsule_server_get(&req->rq_pill, &RMF_SEQ_RANGE);
*output = *out;
if (!range_is_sane(output)) {
CERROR("%s: Invalid range received from server: "
DRANGE"\n", seq->lcs_name, PRANGE(output));
GOTO(out_req, rc = -EINVAL);
}
if (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(D_INFO, "%s: Allocated %s-sequence "DRANGE"]\n",
seq->lcs_name, opcname, PRANGE(output));
EXIT;
out_req:
ptlrpc_req_finished(req);
return rc;
}
int mdt_getxattr(struct mdt_thread_info *info)
{
struct ptlrpc_request *req = mdt_info_req(info);
struct mdt_export_data *med = mdt_req2med(req);
struct lu_ucred *uc = lu_ucred(info->mti_env);
struct mdt_body *reqbody;
struct mdt_body *repbody = NULL;
struct md_object *next;
struct lu_buf *buf;
__u32 remote = exp_connect_rmtclient(info->mti_exp);
__u32 perm;
int easize, rc;
u64 valid;
ENTRY;
LASSERT(info->mti_object != NULL);
LASSERT(lu_object_assert_exists(&info->mti_object->mot_obj));
CDEBUG(D_INODE, "getxattr "DFID"\n", PFID(&info->mti_body->mbo_fid1));
reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
if (reqbody == NULL)
RETURN(err_serious(-EFAULT));
rc = mdt_init_ucred(info, reqbody);
if (rc)
RETURN(err_serious(rc));
next = mdt_object_child(info->mti_object);
if (info->mti_body->mbo_valid & OBD_MD_FLRMTRGETFACL) {
if (unlikely(!remote))
GOTO(out, rc = err_serious(-EINVAL));
perm = mdt_identity_get_perm(uc->uc_identity, remote,
req->rq_peer.nid);
if (!(perm & CFS_RMTACL_PERM))
GOTO(out, rc = err_serious(-EPERM));
rc = mo_permission(info->mti_env, NULL, next, NULL,
MAY_RGETFACL);
if (rc)
GOTO(out, rc = err_serious(rc));
}
easize = mdt_getxattr_pack_reply(info);
if (easize < 0)
GOTO(out, rc = err_serious(easize));
repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
LASSERT(repbody != NULL);
/* No need further getxattr. */
if (easize == 0 || reqbody->mbo_eadatasize == 0)
GOTO(out, rc = easize);
buf = &info->mti_buf;
buf->lb_buf = req_capsule_server_get(info->mti_pill, &RMF_EADATA);
buf->lb_len = easize;
valid = info->mti_body->mbo_valid & (OBD_MD_FLXATTR | OBD_MD_FLXATTRLS);
if (valid == OBD_MD_FLXATTR) {
char *xattr_name = req_capsule_client_get(info->mti_pill,
&RMF_NAME);
rc = mdt_getxattr_one(info, xattr_name, next, buf, med, uc);
} else if (valid == OBD_MD_FLXATTRLS) {
CDEBUG(D_INODE, "listxattr\n");
rc = mo_xattr_list(info->mti_env, next, buf);
if (rc < 0)
CDEBUG(D_INFO, "listxattr failed: %d\n", rc);
} else if (valid == OBD_MD_FLXATTRALL) {
rc = mdt_getxattr_all(info, reqbody, repbody,
buf, next);
} else
LBUG();
EXIT;
out:
if (rc >= 0) {
mdt_counter_incr(req, LPROC_MDT_GETXATTR);
repbody->mbo_eadatasize = rc;
rc = 0;
}
mdt_exit_ucred(info);
return rc;
}
/**
* Object updates between Targets. Because all the updates has been
* dis-assemblied into object updates at sender side, so OUT will
* call OSD API directly to execute these updates.
*
* In DNE phase I all of the updates in the request need to be executed
* in one transaction, and the transaction has to be synchronously.
*
* Please refer to lustre/include/lustre/lustre_idl.h for req/reply
* format.
*/
int out_handle(struct tgt_session_info *tsi)
{
const struct lu_env *env = tsi->tsi_env;
struct tgt_thread_info *tti = tgt_th_info(env);
struct thandle_exec_args *ta = &tti->tti_tea;
struct req_capsule *pill = tsi->tsi_pill;
struct dt_device *dt = tsi->tsi_tgt->lut_bottom;
struct object_update_request *ureq;
struct object_update *update;
struct object_update_reply *reply;
int bufsize;
int count;
int old_batchid = -1;
int i;
int rc = 0;
int rc1 = 0;
ENTRY;
req_capsule_set(pill, &RQF_OUT_UPDATE);
ureq = req_capsule_client_get(pill, &RMF_OUT_UPDATE);
if (ureq == NULL) {
CERROR("%s: No buf!: rc = %d\n", tgt_name(tsi->tsi_tgt),
-EPROTO);
RETURN(err_serious(-EPROTO));
}
bufsize = req_capsule_get_size(pill, &RMF_OUT_UPDATE, RCL_CLIENT);
if (bufsize != object_update_request_size(ureq)) {
CERROR("%s: invalid bufsize %d: rc = %d\n",
tgt_name(tsi->tsi_tgt), bufsize, -EPROTO);
RETURN(err_serious(-EPROTO));
}
if (ureq->ourq_magic != UPDATE_REQUEST_MAGIC) {
CERROR("%s: invalid update buffer magic %x expect %x: "
"rc = %d\n", tgt_name(tsi->tsi_tgt), ureq->ourq_magic,
UPDATE_REQUEST_MAGIC, -EPROTO);
RETURN(err_serious(-EPROTO));
}
count = ureq->ourq_count;
if (count <= 0) {
CERROR("%s: empty update: rc = %d\n", tgt_name(tsi->tsi_tgt),
-EPROTO);
RETURN(err_serious(-EPROTO));
}
req_capsule_set_size(pill, &RMF_OUT_UPDATE_REPLY, RCL_SERVER,
OUT_UPDATE_REPLY_SIZE);
rc = req_capsule_server_pack(pill);
if (rc != 0) {
CERROR("%s: Can't pack response: rc = %d\n",
tgt_name(tsi->tsi_tgt), rc);
RETURN(rc);
}
/* Prepare the update reply buffer */
reply = req_capsule_server_get(pill, &RMF_OUT_UPDATE_REPLY);
if (reply == NULL)
RETURN(err_serious(-EPROTO));
object_update_reply_init(reply, count);
tti->tti_u.update.tti_update_reply = reply;
rc = out_tx_start(env, dt, ta, tsi->tsi_exp);
if (rc != 0)
RETURN(rc);
tti->tti_mult_trans = !req_is_replay(tgt_ses_req(tsi));
/* Walk through updates in the request to execute them synchronously */
for (i = 0; i < count; i++) {
struct tgt_handler *h;
struct dt_object *dt_obj;
update = object_update_request_get(ureq, i, NULL);
if (update == NULL)
GOTO(out, rc = -EPROTO);
if (ptlrpc_req_need_swab(pill->rc_req))
lustre_swab_object_update(update);
if (old_batchid == -1) {
old_batchid = update->ou_batchid;
} else if (old_batchid != update->ou_batchid) {
/* Stop the current update transaction,
* create a new one */
rc = out_tx_end(env, ta);
if (rc < 0)
RETURN(rc);
rc = out_tx_start(env, dt, ta, tsi->tsi_exp);
if (rc != 0)
RETURN(rc);
old_batchid = update->ou_batchid;
}
if (!fid_is_sane(&update->ou_fid)) {
CERROR("%s: invalid FID "DFID": rc = %d\n",
tgt_name(tsi->tsi_tgt), PFID(&update->ou_fid),
-EPROTO);
GOTO(out, rc = err_serious(-EPROTO));
}
dt_obj = dt_locate(env, dt, &update->ou_fid);
if (IS_ERR(dt_obj))
GOTO(out, rc = PTR_ERR(dt_obj));
if (dt->dd_record_fid_accessed) {
lfsck_pack_rfa(&tti->tti_lr,
lu_object_fid(&dt_obj->do_lu));
tgt_lfsck_in_notify(env, dt, &tti->tti_lr);
}
tti->tti_u.update.tti_dt_object = dt_obj;
tti->tti_u.update.tti_update = update;
tti->tti_u.update.tti_update_reply_index = i;
h = out_handler_find(update->ou_type);
if (likely(h != NULL)) {
/* For real modification RPC, check if the update
* has been executed */
if (h->th_flags & MUTABOR) {
struct ptlrpc_request *req = tgt_ses_req(tsi);
if (out_check_resent(env, dt, dt_obj, req,
out_reconstruct, reply, i))
GOTO(next, rc);
}
rc = h->th_act(tsi);
} else {
CERROR("%s: The unsupported opc: 0x%x\n",
tgt_name(tsi->tsi_tgt), update->ou_type);
lu_object_put(env, &dt_obj->do_lu);
GOTO(out, rc = -ENOTSUPP);
}
next:
lu_object_put(env, &dt_obj->do_lu);
if (rc < 0)
GOTO(out, rc);
}
out:
rc1 = out_tx_end(env, ta);
if (rc == 0)
rc = rc1;
RETURN(rc);
}
static int mdt_create_unpack(struct mdt_thread_info *info)
{
struct lu_ucred *uc = mdt_ucred(info);
struct mdt_rec_create *rec;
struct lu_attr *attr = &info->mti_attr.ma_attr;
struct mdt_reint_record *rr = &info->mti_rr;
struct req_capsule *pill = info->mti_pill;
struct md_op_spec *sp = &info->mti_spec;
int rc;
ENTRY;
CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
rec = req_capsule_client_get(pill, &RMF_REC_REINT);
if (rec == NULL)
RETURN(-EFAULT);
/* This prior initialization is needed for old_init_ucred_reint() */
uc->uc_fsuid = rec->cr_fsuid;
uc->uc_fsgid = rec->cr_fsgid;
uc->uc_cap = rec->cr_cap;
uc->uc_suppgids[0] = rec->cr_suppgid1;
uc->uc_suppgids[1] = -1;
uc->uc_umask = rec->cr_umask;
rr->rr_fid1 = &rec->cr_fid1;
rr->rr_fid2 = &rec->cr_fid2;
attr->la_mode = rec->cr_mode;
attr->la_rdev = rec->cr_rdev;
attr->la_uid = rec->cr_fsuid;
attr->la_gid = rec->cr_fsgid;
attr->la_ctime = rec->cr_time;
attr->la_mtime = rec->cr_time;
attr->la_atime = rec->cr_time;
attr->la_valid = LA_MODE | LA_RDEV | LA_UID | LA_GID | LA_TYPE |
LA_CTIME | LA_MTIME | LA_ATIME;
memset(&sp->u, 0, sizeof(sp->u));
sp->sp_cr_flags = get_mrc_cr_flags(rec);
if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
mdt_set_capainfo(info, 0, rr->rr_fid1,
req_capsule_client_get(pill, &RMF_CAPA1));
mdt_set_capainfo(info, 1, rr->rr_fid2, BYPASS_CAPA);
rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
if (rc < 0)
RETURN(rc);
if (S_ISLNK(attr->la_mode)) {
const char *tgt = NULL;
req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_SYM);
if (req_capsule_get_size(pill, &RMF_SYMTGT, RCL_CLIENT)) {
tgt = req_capsule_client_get(pill, &RMF_SYMTGT);
sp->u.sp_symname = tgt;
}
if (tgt == NULL)
RETURN(-EFAULT);
} else {
req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_RMT_ACL);
if (S_ISDIR(attr->la_mode) &&
req_capsule_get_size(pill, &RMF_EADATA, RCL_CLIENT) > 0) {
sp->u.sp_ea.eadata =
req_capsule_client_get(pill, &RMF_EADATA);
sp->u.sp_ea.eadatalen =
req_capsule_get_size(pill, &RMF_EADATA,
RCL_CLIENT);
sp->sp_cr_flags |= MDS_OPEN_HAS_EA;
}
}
rc = mdt_dlmreq_unpack(info);
RETURN(rc);
}
/* This is a callback from the llog_* functions.
* Assumes caller has already pushed us into the kernel context.
*/
static int llog_client_open(const struct lu_env *env,
struct llog_handle *lgh, struct llog_logid *logid,
char *name, enum llog_open_param open_param)
{
struct obd_import *imp;
struct llogd_body *body;
struct llog_ctxt *ctxt = lgh->lgh_ctxt;
struct ptlrpc_request *req = NULL;
int rc;
LLOG_CLIENT_ENTRY(ctxt, imp);
/* client cannot create llog */
LASSERTF(open_param != LLOG_OPEN_NEW, "%#x\n", open_param);
LASSERT(lgh);
req = ptlrpc_request_alloc(imp, &RQF_LLOG_ORIGIN_HANDLE_CREATE);
if (!req) {
rc = -ENOMEM;
goto out;
}
if (name)
req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
strlen(name) + 1);
rc = ptlrpc_request_pack(req, LUSTRE_LOG_VERSION,
LLOG_ORIGIN_HANDLE_CREATE);
if (rc) {
ptlrpc_request_free(req);
req = NULL;
goto out;
}
ptlrpc_request_set_replen(req);
body = req_capsule_client_get(&req->rq_pill, &RMF_LLOGD_BODY);
if (logid)
body->lgd_logid = *logid;
body->lgd_ctxt_idx = ctxt->loc_idx - 1;
if (name) {
char *tmp;
tmp = req_capsule_client_sized_get(&req->rq_pill, &RMF_NAME,
strlen(name) + 1);
LASSERT(tmp);
strcpy(tmp, name);
}
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;
}
lgh->lgh_id = body->lgd_logid;
lgh->lgh_ctxt = ctxt;
out:
LLOG_CLIENT_EXIT(ctxt, imp);
ptlrpc_req_finished(req);
return rc;
}
/*
* Fetch a global or slave index from the QMT.
*
* \param env - the environment passed by the caller
* \param exp - is the export to use to issue the OBD_IDX_READ RPC
* \param ii - is the index information to be packed in the request
* on success, the index information returned by the server
* is copied there.
* \param npages - is the number of pages in the pages array
* \param pages - is an array of @npages pages
*
* \retval 0 - success
* \retval -ve - appropriate errors
*/
int qsd_fetch_index(const struct lu_env *env, struct obd_export *exp,
struct idx_info *ii, unsigned int npages,
struct page **pages, bool *need_swab)
{
struct ptlrpc_request *req;
struct idx_info *req_ii;
struct ptlrpc_bulk_desc *desc;
int rc, i;
ENTRY;
LASSERT(exp);
req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OBD_IDX_READ);
if (req == NULL)
RETURN(-ENOMEM);
rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, OBD_IDX_READ);
if (rc) {
ptlrpc_request_free(req);
RETURN(rc);
}
req->rq_request_portal = MDS_READPAGE_PORTAL;
ptlrpc_at_set_req_timeout(req);
/* allocate bulk descriptor */
desc = ptlrpc_prep_bulk_imp(req, npages, 1, BULK_PUT_SINK,
MDS_BULK_PORTAL);
if (desc == NULL) {
ptlrpc_request_free(req);
RETURN(-ENOMEM);
}
/* req now owns desc and will free it when it gets freed */
for (i = 0; i < npages; i++)
ptlrpc_prep_bulk_page_pin(desc, pages[i], 0, PAGE_CACHE_SIZE);
/* pack index information in request */
req_ii = req_capsule_client_get(&req->rq_pill, &RMF_IDX_INFO);
*req_ii = *ii;
ptlrpc_request_set_replen(req);
/* send request to master and wait for RPC to complete */
rc = ptlrpc_queue_wait(req);
if (rc)
GOTO(out, rc);
rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
req->rq_bulk->bd_nob_transferred);
if (rc < 0)
GOTO(out, rc);
else
/* sptlrpc_cli_unwrap_bulk_read() returns the number of bytes
* transferred*/
rc = 0;
req_ii = req_capsule_server_get(&req->rq_pill, &RMF_IDX_INFO);
*ii = *req_ii;
*need_swab = ptlrpc_rep_need_swab(req);
EXIT;
out:
ptlrpc_req_finished(req);
return rc;
}
/*
* Handle quota request from slave.
*
* \param env - is the environment passed by the caller
* \param ld - is the lu device associated with the qmt
* \param req - is the quota acquire request
*/
static int qmt_dqacq(const struct lu_env *env, struct lu_device *ld,
struct ptlrpc_request *req)
{
struct qmt_device *qmt = lu2qmt_dev(ld);
struct quota_body *qbody, *repbody;
struct obd_uuid *uuid;
struct ldlm_lock *lock;
struct lquota_entry *lqe;
int pool_id, pool_type, qtype;
int rc;
ENTRY;
qbody = req_capsule_client_get(&req->rq_pill, &RMF_QUOTA_BODY);
if (qbody == NULL)
RETURN(err_serious(-EPROTO));
repbody = req_capsule_server_get(&req->rq_pill, &RMF_QUOTA_BODY);
if (repbody == NULL)
RETURN(err_serious(-EFAULT));
/* verify if global lock is stale */
if (!lustre_handle_is_used(&qbody->qb_glb_lockh))
RETURN(-ENOLCK);
lock = ldlm_handle2lock(&qbody->qb_glb_lockh);
if (lock == NULL)
RETURN(-ENOLCK);
LDLM_LOCK_PUT(lock);
uuid = &req->rq_export->exp_client_uuid;
if (req_is_rel(qbody->qb_flags) + req_is_acq(qbody->qb_flags) +
req_is_preacq(qbody->qb_flags) > 1) {
CERROR("%s: malformed quota request with conflicting flags set "
"(%x) from slave %s\n", qmt->qmt_svname,
qbody->qb_flags, obd_uuid2str(uuid));
RETURN(-EPROTO);
}
if (req_is_acq(qbody->qb_flags) || req_is_preacq(qbody->qb_flags)) {
/* acquire and pre-acquire should use a valid ID lock */
if (!lustre_handle_is_used(&qbody->qb_lockh))
RETURN(-ENOLCK);
lock = ldlm_handle2lock(&qbody->qb_lockh);
if (lock == NULL)
/* no lock associated with this handle */
RETURN(-ENOLCK);
LDLM_DEBUG(lock, "%sacquire request",
req_is_preacq(qbody->qb_flags) ? "pre" : "");
if (!obd_uuid_equals(&lock->l_export->exp_client_uuid, uuid)) {
/* sorry, no way to cheat ... */
LDLM_LOCK_PUT(lock);
RETURN(-ENOLCK);
}
if ((lock->l_flags & LDLM_FL_AST_SENT) != 0) {
struct ptlrpc_service_part *svc;
unsigned int timeout;
svc = req->rq_rqbd->rqbd_svcpt;
timeout = at_est2timeout(at_get(&svc->scp_at_estimate));
timeout = max(timeout, ldlm_timeout);
/* lock is being cancelled, prolong timeout */
ldlm_refresh_waiting_lock(lock, timeout);
}
LDLM_LOCK_PUT(lock);
}
/* extract pool & quota information from global index FID packed in the
* request */
rc = lquota_extract_fid(&qbody->qb_fid, &pool_id, &pool_type, &qtype);
if (rc)
RETURN(-EINVAL);
/* Find the quota entry associated with the quota id */
lqe = qmt_pool_lqe_lookup(env, qmt, pool_id, pool_type, qtype,
&qbody->qb_id);
if (IS_ERR(lqe))
RETURN(PTR_ERR(lqe));
/* process quota request */
rc = qmt_dqacq0(env, lqe, qmt, uuid, qbody->qb_flags, qbody->qb_count,
qbody->qb_usage, repbody);
if (lustre_handle_is_used(&qbody->qb_lockh))
/* return current qunit value only to slaves owning an per-ID
* quota lock. For enqueue, the qunit value will be returned in
* the LVB */
repbody->qb_qunit = lqe->lqe_qunit;
lqe_putref(lqe);
RETURN(rc);
}
/* If mdc_setattr is called with an 'iattr', then it is a normal RPC that
* should take the normal semaphore and go to the normal portal.
*
* If it is called with iattr->ia_valid & ATTR_FROM_OPEN, then it is a
* magic open-path setattr that should take the setattr semaphore and
* go to the setattr portal. */
int mdc_setattr(struct obd_export *exp, struct md_op_data *op_data,
void *ea, int ealen, void *ea2, int ea2len,
struct ptlrpc_request **request, struct md_open_data **mod)
{
CFS_LIST_HEAD(cancels);
struct ptlrpc_request *req;
struct mdc_rpc_lock *rpc_lock;
struct obd_device *obd = exp->exp_obd;
int count = 0, rc;
__u64 bits;
ENTRY;
LASSERT(op_data != NULL);
bits = MDS_INODELOCK_UPDATE;
if (op_data->op_attr.ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID))
bits |= MDS_INODELOCK_LOOKUP;
if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
(fid_is_sane(&op_data->op_fid1)))
count = mdc_resource_get_unused(exp, &op_data->op_fid1,
&cancels, LCK_EX, bits);
req = ptlrpc_request_alloc(class_exp2cliimp(exp),
&RQF_MDS_REINT_SETATTR);
if (req == NULL) {
ldlm_lock_list_put(&cancels, l_bl_ast, count);
RETURN(-ENOMEM);
}
mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
if ((op_data->op_flags & (MF_SOM_CHANGE | MF_EPOCH_OPEN)) == 0)
req_capsule_set_size(&req->rq_pill, &RMF_MDT_EPOCH, RCL_CLIENT,
0);
req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT, ealen);
req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_CLIENT,
ea2len);
rc = mdc_prep_elc_req(exp, req, &cancels, count);
if (rc) {
ptlrpc_request_free(req);
RETURN(rc);
}
if (op_data->op_attr.ia_valid & ATTR_FROM_OPEN) {
req->rq_request_portal = MDS_SETATTR_PORTAL;
ptlrpc_at_set_req_timeout(req);
rpc_lock = obd->u.cli.cl_setattr_lock;
} else {
rpc_lock = obd->u.cli.cl_rpc_lock;
}
if (op_data->op_attr.ia_valid & (ATTR_MTIME | ATTR_CTIME))
CDEBUG(D_INODE, "setting mtime "CFS_TIME_T
", ctime "CFS_TIME_T"\n",
LTIME_S(op_data->op_attr.ia_mtime),
LTIME_S(op_data->op_attr.ia_ctime));
mdc_setattr_pack(req, op_data, ea, ealen, ea2, ea2len);
ptlrpc_request_set_replen(req);
if (mod && (op_data->op_flags & MF_EPOCH_OPEN) &&
req->rq_import->imp_replayable)
{
LASSERT(*mod == NULL);
*mod = obd_mod_alloc();
if (*mod == NULL) {
DEBUG_REQ(D_ERROR, req, "Can't allocate "
"md_open_data");
} else {
req->rq_replay = 1;
req->rq_cb_data = *mod;
(*mod)->mod_open_req = req;
req->rq_commit_cb = mdc_commit_open;
/**
* Take an extra reference on \var mod, it protects \var
* mod from being freed on eviction (commit callback is
* called despite rq_replay flag).
* Will be put on mdc_done_writing().
*/
obd_mod_get(*mod);
}
}
rc = mdc_reint(req, rpc_lock, LUSTRE_IMP_FULL);
/* Save the obtained info in the original RPC for the replay case. */
if (rc == 0 && (op_data->op_flags & MF_EPOCH_OPEN)) {
struct mdt_ioepoch *epoch;
struct mdt_body *body;
epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
LASSERT(epoch != NULL);
LASSERT(body != NULL);
epoch->handle = body->handle;
epoch->ioepoch = body->ioepoch;
req->rq_replay_cb = mdc_replay_open;
/** bug 3633, open may be committed and estale answer is not error */
} else if (rc == -ESTALE && (op_data->op_flags & MF_SOM_CHANGE)) {
rc = 0;
} else if (rc == -ERESTARTSYS) {
rc = 0;
}
*request = req;
if (rc && req->rq_commit_cb) {
/* Put an extra reference on \var mod on error case. */
obd_mod_put(*mod);
req->rq_commit_cb(req);
}
RETURN(rc);
}
static int osp_get_lastfid_from_ost(const struct lu_env *env,
struct osp_device *d)
{
struct ptlrpc_request *req = NULL;
struct obd_import *imp;
struct lu_fid *last_fid;
char *tmp;
int rc;
ENTRY;
imp = d->opd_obd->u.cli.cl_import;
LASSERT(imp);
req = ptlrpc_request_alloc(imp, &RQF_OST_GET_INFO_LAST_FID);
if (req == NULL)
RETURN(-ENOMEM);
req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_KEY, RCL_CLIENT,
sizeof(KEY_LAST_FID));
req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_VAL, RCL_CLIENT,
sizeof(struct lu_fid));
rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_GET_INFO);
if (rc) {
ptlrpc_request_free(req);
RETURN(rc);
}
tmp = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_KEY);
memcpy(tmp, KEY_LAST_FID, sizeof(KEY_LAST_FID));
req->rq_no_delay = req->rq_no_resend = 1;
tmp = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_VAL);
fid_cpu_to_le((struct lu_fid *)tmp, &d->opd_last_used_fid);
ptlrpc_request_set_replen(req);
rc = ptlrpc_queue_wait(req);
if (rc) {
/* bad-bad OST.. let sysadm sort this out */
if (rc == -ENOTSUPP) {
CERROR("%s: server does not support FID: rc = %d\n",
d->opd_obd->obd_name, -ENOTSUPP);
}
ptlrpc_set_import_active(imp, 0);
GOTO(out, rc);
}
last_fid = req_capsule_server_get(&req->rq_pill, &RMF_FID);
if (last_fid == NULL) {
CERROR("%s: Got last_fid failed.\n", d->opd_obd->obd_name);
GOTO(out, rc = -EPROTO);
}
if (!fid_is_sane(last_fid)) {
CERROR("%s: Got insane last_fid "DFID"\n",
d->opd_obd->obd_name, PFID(last_fid));
GOTO(out, rc = -EPROTO);
}
/* Only update the last used fid, if the OST has objects for
* this sequence, i.e. fid_oid > 0 */
if (fid_oid(last_fid) > 0)
d->opd_last_used_fid = *last_fid;
CDEBUG(D_HA, "%s: Got last_fid "DFID"\n", d->opd_obd->obd_name,
PFID(last_fid));
out:
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);
d->opd_pre_recovering = 0;
/*
* 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 {
cfs_waitq_signal(&d->opd_pre_user_waitq);
}
}
RETURN(rc);
}
static int osp_precreate_send(const struct lu_env *env, struct osp_device *d)
{
struct osp_thread_info *oti = osp_env_info(env);
struct ptlrpc_request *req;
struct obd_import *imp;
struct ost_body *body;
int rc, grow, diff;
struct lu_fid *fid = &oti->osi_fid;
ENTRY;
/* don't precreate new objects till OST healthy and has free space */
if (unlikely(d->opd_pre_status)) {
CDEBUG(D_INFO, "%s: don't send new precreate: rc = %d\n",
d->opd_obd->obd_name, d->opd_pre_status);
RETURN(0);
}
/*
* if not connection/initialization is compeleted, ignore
*/
imp = d->opd_obd->u.cli.cl_import;
LASSERT(imp);
req = ptlrpc_request_alloc(imp, &RQF_OST_CREATE);
if (req == NULL)
RETURN(-ENOMEM);
req->rq_request_portal = OST_CREATE_PORTAL;
/* we should not resend create request - anyway we will have delorphan
* and kill these objects */
req->rq_no_delay = req->rq_no_resend = 1;
rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_CREATE);
if (rc) {
ptlrpc_request_free(req);
RETURN(rc);
}
spin_lock(&d->opd_pre_lock);
if (d->opd_pre_grow_count > d->opd_pre_max_grow_count / 2)
d->opd_pre_grow_count = d->opd_pre_max_grow_count / 2;
grow = d->opd_pre_grow_count;
spin_unlock(&d->opd_pre_lock);
body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
LASSERT(body);
*fid = d->opd_pre_last_created_fid;
rc = osp_precreate_fids(env, d, fid, &grow);
if (rc == 1) {
/* Current seq has been used up*/
if (!osp_is_fid_client(d)) {
osp_pre_update_status(d, -ENOSPC);
rc = -ENOSPC;
}
cfs_waitq_signal(&d->opd_pre_waitq);
GOTO(out_req, rc);
}
if (!osp_is_fid_client(d)) {
/* Non-FID client will always send seq 0 because of
* compatiblity */
LASSERTF(fid_is_idif(fid), "Invalid fid "DFID"\n", PFID(fid));
fid->f_seq = 0;
}
fid_to_ostid(fid, &body->oa.o_oi);
body->oa.o_valid = OBD_MD_FLGROUP;
ptlrpc_request_set_replen(req);
rc = ptlrpc_queue_wait(req);
if (rc) {
CERROR("%s: can't precreate: rc = %d\n", d->opd_obd->obd_name,
rc);
GOTO(out_req, rc);
}
LASSERT(req->rq_transno == 0);
body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
if (body == NULL)
GOTO(out_req, rc = -EPROTO);
ostid_to_fid(fid, &body->oa.o_oi, d->opd_index);
LASSERTF(lu_fid_diff(fid, &d->opd_pre_used_fid) > 0,
"reply fid "DFID" pre used fid "DFID"\n", PFID(fid),
PFID(&d->opd_pre_used_fid));
diff = lu_fid_diff(fid, &d->opd_pre_last_created_fid);
spin_lock(&d->opd_pre_lock);
if (diff < grow) {
/* the OST has not managed to create all the
* objects we asked for */
d->opd_pre_grow_count = max(diff, OST_MIN_PRECREATE);
d->opd_pre_grow_slow = 1;
} else {
/* the OST is able to keep up with the work,
* we could consider increasing grow_count
* next time if needed */
d->opd_pre_grow_slow = 0;
}
d->opd_pre_last_created_fid = *fid;
spin_unlock(&d->opd_pre_lock);
CDEBUG(D_HA, "%s: current precreated pool: "DFID"-"DFID"\n",
d->opd_obd->obd_name, PFID(&d->opd_pre_used_fid),
PFID(&d->opd_pre_last_created_fid));
out_req:
/* now we can wakeup all users awaiting for objects */
osp_pre_update_status(d, rc);
cfs_waitq_signal(&d->opd_pre_user_waitq);
ptlrpc_req_finished(req);
RETURN(rc);
}
int osp_object_truncate(const struct lu_env *env, struct dt_object *dt,
__u64 size)
{
struct osp_device *d = lu2osp_dev(dt->do_lu.lo_dev);
struct ptlrpc_request *req = NULL;
struct obd_import *imp;
struct ost_body *body;
struct obdo *oa = NULL;
int rc;
ENTRY;
imp = d->opd_obd->u.cli.cl_import;
LASSERT(imp);
req = ptlrpc_request_alloc(imp, &RQF_OST_PUNCH);
if (req == NULL)
RETURN(-ENOMEM);
/* XXX: capa support? */
/* osc_set_capa_size(req, &RMF_CAPA1, capa); */
rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_PUNCH);
if (rc) {
ptlrpc_request_free(req);
RETURN(rc);
}
/*
* XXX: decide how do we do here with resend
* if we don't resend, then client may see wrong file size
* if we do resend, then MDS thread can get stuck for quite long
*/
req->rq_no_resend = req->rq_no_delay = 1;
req->rq_request_portal = OST_IO_PORTAL; /* bug 7198 */
ptlrpc_at_set_req_timeout(req);
OBD_ALLOC_PTR(oa);
if (oa == NULL)
GOTO(out, rc = -ENOMEM);
rc = fid_to_ostid(lu_object_fid(&dt->do_lu), &oa->o_oi);
LASSERT(rc == 0);
oa->o_size = size;
oa->o_blocks = OBD_OBJECT_EOF;
oa->o_valid = OBD_MD_FLSIZE | OBD_MD_FLBLOCKS |
OBD_MD_FLID | OBD_MD_FLGROUP;
body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
LASSERT(body);
lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, oa);
/* XXX: capa support? */
/* osc_pack_capa(req, body, capa); */
ptlrpc_request_set_replen(req);
rc = ptlrpc_queue_wait(req);
if (rc)
CERROR("can't punch object: %d\n", rc);
out:
ptlrpc_req_finished(req);
if (oa)
OBD_FREE_PTR(oa);
RETURN(rc);
}
static int mdt_setxattr_unpack(struct mdt_thread_info *info)
{
struct mdt_reint_record *rr = &info->mti_rr;
struct lu_ucred *uc = mdt_ucred(info);
struct lu_attr *attr = &info->mti_attr.ma_attr;
struct req_capsule *pill = info->mti_pill;
struct mdt_rec_setxattr *rec;
ENTRY;
CLASSERT(sizeof(struct mdt_rec_setxattr) ==
sizeof(struct mdt_rec_reint));
rec = req_capsule_client_get(pill, &RMF_REC_REINT);
if (rec == NULL)
RETURN(-EFAULT);
/* This prior initialization is needed for old_init_ucred_reint() */
uc->uc_fsuid = rec->sx_fsuid;
uc->uc_fsgid = rec->sx_fsgid;
uc->uc_cap = rec->sx_cap;
uc->uc_suppgids[0] = rec->sx_suppgid1;
uc->uc_suppgids[1] = -1;
rr->rr_opcode = rec->sx_opcode;
rr->rr_fid1 = &rec->sx_fid;
attr->la_valid = rec->sx_valid;
attr->la_ctime = rec->sx_time;
attr->la_size = rec->sx_size;
attr->la_flags = rec->sx_flags;
if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
mdt_set_capainfo(info, 0, rr->rr_fid1,
req_capsule_client_get(pill, &RMF_CAPA1));
else
mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
if (rr->rr_name == NULL)
RETURN(-EFAULT);
rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT) - 1;
LASSERT(rr->rr_namelen > 0);
if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
RCL_CLIENT);
if (rr->rr_eadatalen > 0) {
rr->rr_eadata = req_capsule_client_get(pill,
&RMF_EADATA);
if (rr->rr_eadata == NULL)
RETURN(-EFAULT);
} else {
rr->rr_eadata = NULL;
}
} else if (!(attr->la_valid & OBD_MD_FLXATTRRM)) {
CDEBUG(D_INFO, "no xattr data supplied\n");
RETURN(-EFAULT);
}
RETURN(0);
}
/* Called whenever a target starts up. Flags indicate first connect, etc. */
static int mgs_handle_target_reg(struct ptlrpc_request *req)
{
struct obd_device *obd = req->rq_export->exp_obd;
struct mgs_target_info *mti, *rep_mti;
struct fs_db *fsdb;
int opc;
int rc = 0;
ENTRY;
mgs_counter_incr(req->rq_export, LPROC_MGS_TARGET_REG);
mti = req_capsule_client_get(&req->rq_pill, &RMF_MGS_TARGET_INFO);
opc = mti->mti_flags & LDD_F_OPC_MASK;
if (opc == LDD_F_OPC_READY) {
CDEBUG(D_MGS, "fs: %s index: %d is ready to reconnect.\n",
mti->mti_fsname, mti->mti_stripe_index);
rc = mgs_ir_update(obd, mti);
if (rc) {
LASSERT(!(mti->mti_flags & LDD_F_IR_CAPABLE));
CERROR("Update IR return with %d(ignore and IR "
"disabled)\n", rc);
}
GOTO(out_nolock, rc);
}
/* Do not support unregistering right now. */
if (opc != LDD_F_OPC_REG)
GOTO(out_nolock, rc = -EINVAL);
CDEBUG(D_MGS, "fs: %s index: %d is registered to MGS.\n",
mti->mti_fsname, mti->mti_stripe_index);
if (mti->mti_flags & LDD_F_NEED_INDEX)
mti->mti_flags |= LDD_F_WRITECONF;
if (!(mti->mti_flags & (LDD_F_WRITECONF | LDD_F_UPGRADE14 |
LDD_F_UPDATE))) {
/* We're just here as a startup ping. */
CDEBUG(D_MGS, "Server %s is running on %s\n",
mti->mti_svname, obd_export_nid2str(req->rq_export));
rc = mgs_check_target(obd, mti);
/* above will set appropriate mti flags */
if (rc <= 0)
/* Nothing wrong, or fatal error */
GOTO(out_nolock, rc);
} else {
if (!(mti->mti_flags & LDD_F_NO_PRIMNODE)
&& (rc = mgs_check_failover_reg(mti)))
GOTO(out_nolock, rc);
}
OBD_FAIL_TIMEOUT(OBD_FAIL_MGS_PAUSE_TARGET_REG, 10);
if (mti->mti_flags & LDD_F_WRITECONF) {
if (mti->mti_flags & LDD_F_SV_TYPE_MDT &&
mti->mti_stripe_index == 0) {
rc = mgs_erase_logs(obd, mti->mti_fsname);
LCONSOLE_WARN("%s: Logs for fs %s were removed by user "
"request. All servers must be restarted "
"in order to regenerate the logs."
"\n", obd->obd_name, mti->mti_fsname);
} else if (mti->mti_flags &
(LDD_F_SV_TYPE_OST | LDD_F_SV_TYPE_MDT)) {
rc = mgs_erase_log(obd, mti->mti_svname);
LCONSOLE_WARN("%s: Regenerating %s log by user "
"request.\n",
obd->obd_name, mti->mti_svname);
}
mti->mti_flags |= LDD_F_UPDATE;
/* Erased logs means start from scratch. */
mti->mti_flags &= ~LDD_F_UPGRADE14;
}
rc = mgs_find_or_make_fsdb(obd, mti->mti_fsname, &fsdb);
if (rc) {
CERROR("Can't get db for %s: %d\n", mti->mti_fsname, rc);
GOTO(out_nolock, rc);
}
/*
* Log writing contention is handled by the fsdb_mutex.
*
* It should be alright if someone was reading while we were
* updating the logs - if we revoke at the end they will just update
* from where they left off.
*/
/* COMPAT_146 */
if (mti->mti_flags & LDD_F_UPGRADE14) {
rc = mgs_upgrade_sv_14(obd, mti, fsdb);
if (rc) {
CERROR("Can't upgrade from 1.4 (%d)\n", rc);
GOTO(out, rc);
}
/* We're good to go */
mti->mti_flags |= LDD_F_UPDATE;
}
/* end COMPAT_146 */
if (mti->mti_flags & LDD_F_UPDATE) {
CDEBUG(D_MGS, "updating %s, index=%d\n", mti->mti_svname,
mti->mti_stripe_index);
/* create or update the target log
and update the client/mdt logs */
rc = mgs_write_log_target(obd, mti, fsdb);
if (rc) {
CERROR("Failed to write %s log (%d)\n",
mti->mti_svname, rc);
GOTO(out, rc);
}
mti->mti_flags &= ~(LDD_F_VIRGIN | LDD_F_UPDATE |
LDD_F_NEED_INDEX | LDD_F_WRITECONF |
LDD_F_UPGRADE14);
mti->mti_flags |= LDD_F_REWRITE_LDD;
}
out:
mgs_revoke_lock(obd, fsdb, CONFIG_T_CONFIG);
out_nolock:
CDEBUG(D_MGS, "replying with %s, index=%d, rc=%d\n", mti->mti_svname,
mti->mti_stripe_index, rc);
req->rq_status = rc;
if (rc)
/* we need an error flag to tell the target what's going on,
* instead of just doing it by error code only. */
mti->mti_flags |= LDD_F_ERROR;
rc = req_capsule_server_pack(&req->rq_pill);
if (rc)
RETURN(rc);
/* send back the whole mti in the reply */
rep_mti = req_capsule_server_get(&req->rq_pill, &RMF_MGS_TARGET_INFO);
*rep_mti = *mti;
/* Flush logs to disk */
fsfilt_sync(obd, obd->u.mgs.mgs_sb);
RETURN(rc);
}
int llog_origin_handle_prev_block(struct ptlrpc_request *req)
{
struct obd_export *exp = req->rq_export;
struct obd_device *obd = exp->exp_obd;
struct llog_handle *loghandle;
struct llogd_body *body;
struct llogd_body *repbody;
struct obd_device *disk_obd;
struct lvfs_run_ctxt saved;
struct llog_ctxt *ctxt;
__u32 flags;
__u8 *buf;
void *ptr;
int rc, rc2;
ENTRY;
body = req_capsule_client_get(&req->rq_pill, &RMF_LLOGD_BODY);
if (body == NULL)
RETURN(-EFAULT);
OBD_ALLOC(buf, LLOG_CHUNK_SIZE);
if (!buf)
RETURN(-ENOMEM);
ctxt = llog_get_context(obd, body->lgd_ctxt_idx);
if (ctxt == NULL)
GOTO(out_free, rc = -ENODEV);
disk_obd = ctxt->loc_exp->exp_obd;
push_ctxt(&saved, &disk_obd->obd_lvfs_ctxt, NULL);
rc = llog_create(ctxt, &loghandle, &body->lgd_logid, NULL);
if (rc)
GOTO(out_pop, rc);
flags = body->lgd_llh_flags;
rc = llog_init_handle(loghandle, flags, NULL);
if (rc)
GOTO(out_close, rc);
memset(buf, 0, LLOG_CHUNK_SIZE);
rc = llog_prev_block(loghandle, body->lgd_index,
buf, LLOG_CHUNK_SIZE);
if (rc)
GOTO(out_close, rc);
req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_SERVER,
LLOG_CHUNK_SIZE);
rc = req_capsule_server_pack(&req->rq_pill);
if (rc)
GOTO(out_close, rc = -ENOMEM);
repbody = req_capsule_server_get(&req->rq_pill, &RMF_LLOGD_BODY);
*repbody = *body;
ptr = req_capsule_server_get(&req->rq_pill, &RMF_EADATA);
memcpy(ptr, buf, LLOG_CHUNK_SIZE);
GOTO(out_close, rc);
out_close:
rc2 = llog_close(loghandle);
if (!rc)
rc = rc2;
out_pop:
pop_ctxt(&saved, &disk_obd->obd_lvfs_ctxt, NULL);
llog_ctxt_put(ctxt);
out_free:
OBD_FREE(buf, LLOG_CHUNK_SIZE);
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) {
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) {
rc = -EFAULT;
goto out;
}
if (handle->lgh_hdr_size < hdr->llh_hdr.lrh_len) {
rc = -EFAULT;
goto out;
}
memcpy(handle->lgh_hdr, hdr, hdr->llh_hdr.lrh_len);
handle->lgh_last_idx = LLOG_HDR_TAIL(handle->lgh_hdr)->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_HDR_TAIL(handle->lgh_hdr)->lrt_len ||
(llh_hdr->lrh_len & (llh_hdr->lrh_len - 1)) ||
llh_hdr->lrh_len < LLOG_MIN_CHUNK_SIZE ||
llh_hdr->lrh_len > handle->lgh_hdr_size) {
CERROR("incorrectly sized log header: %#x (expecting %#x) (power of two > 8192)\n",
llh_hdr->lrh_len,
LLOG_HDR_TAIL(handle->lgh_hdr)->lrt_len);
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;
}
int llog_origin_handle_cancel(struct ptlrpc_request *req)
{
struct obd_device *obd = req->rq_export->exp_obd;
int num_cookies, rc = 0, err, i, failed = 0;
struct obd_device *disk_obd;
struct llog_cookie *logcookies;
struct llog_ctxt *ctxt = NULL;
struct lvfs_run_ctxt saved;
struct llog_handle *cathandle;
struct inode *inode;
void *handle;
ENTRY;
logcookies = req_capsule_client_get(&req->rq_pill, &RMF_LOGCOOKIES);
num_cookies = req_capsule_get_size(&req->rq_pill, &RMF_LOGCOOKIES,
RCL_CLIENT) / sizeof(*logcookies);
if (logcookies == NULL || num_cookies == 0) {
DEBUG_REQ(D_HA, req, "No llog cookies sent");
RETURN(-EFAULT);
}
ctxt = llog_get_context(obd, logcookies->lgc_subsys);
if (ctxt == NULL)
RETURN(-ENODEV);
disk_obd = ctxt->loc_exp->exp_obd;
push_ctxt(&saved, &disk_obd->obd_lvfs_ctxt, NULL);
for (i = 0; i < num_cookies; i++, logcookies++) {
cathandle = ctxt->loc_handle;
LASSERT(cathandle != NULL);
inode = cathandle->lgh_file->f_dentry->d_inode;
handle = fsfilt_start_log(disk_obd, inode,
FSFILT_OP_CANCEL_UNLINK, NULL, 1);
if (IS_ERR(handle)) {
CERROR("fsfilt_start_log() failed: %ld\n",
PTR_ERR(handle));
GOTO(pop_ctxt, rc = PTR_ERR(handle));
}
rc = llog_cat_cancel_records(cathandle, 1, logcookies);
/*
* Do not raise -ENOENT errors for resent rpcs. This rec already
* might be killed.
*/
if (rc == -ENOENT &&
(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT)) {
/*
* Do not change this message, reply-single.sh test_59b
* expects to find this in log.
*/
CDEBUG(D_RPCTRACE, "RESENT cancel req %p - ignored\n",
req);
rc = 0;
} else if (rc == 0) {
CDEBUG(D_RPCTRACE, "Canceled %d llog-records\n",
num_cookies);
}
err = fsfilt_commit(disk_obd, inode, handle, 0);
if (err) {
CERROR("Error committing transaction: %d\n", err);
if (!rc)
rc = err;
failed++;
GOTO(pop_ctxt, rc);
} else if (rc)
failed++;
}
GOTO(pop_ctxt, rc);
pop_ctxt:
pop_ctxt(&saved, &disk_obd->obd_lvfs_ctxt, NULL);
if (rc)
CERROR("Cancel %d of %d llog-records failed: %d\n",
failed, num_cookies, rc);
llog_ctxt_put(ctxt);
return rc;
}
/*
* Get intent per-ID lock or global-index lock from master.
*
* \param env - the environment passed by the caller
* \param exp - is the export to use to send the intent RPC
* \param qbody - quota body to be packed in request
* \param sync - synchronous or asynchronous (pre-acquire)
* \param it_op - IT_QUOTA_DQACQ or IT_QUOTA_CONN
* \param completion - completion callback
* \param qqi - is the qsd_qtype_info structure to pass to the completion
* function
* \param lvb - is the lvb associated with the lock and returned by the
* server
* \param arg - is an opaq argument passed to the completion callback
*
* \retval 0 - success
* \retval -ve - appropriate errors
*/
int qsd_intent_lock(const struct lu_env *env, struct obd_export *exp,
struct quota_body *qbody, bool sync, int it_op,
qsd_req_completion_t completion, struct qsd_qtype_info *qqi,
struct lquota_lvb *lvb, void *arg)
{
struct qsd_thread_info *qti = qsd_info(env);
struct ptlrpc_request *req;
struct qsd_async_args *aa = NULL;
struct ldlm_intent *lit;
struct quota_body *req_qbody;
__u64 flags = LDLM_FL_HAS_INTENT;
int rc;
ENTRY;
LASSERT(exp != NULL);
LASSERT(!lustre_handle_is_used(&qbody->qb_lockh));
memset(&qti->qti_lockh, 0, sizeof(qti->qti_lockh));
req = ptlrpc_request_alloc(class_exp2cliimp(exp),
&RQF_LDLM_INTENT_QUOTA);
if (req == NULL)
GOTO(out, rc = -ENOMEM);
req->rq_no_retry_einprogress = 1;
rc = ldlm_prep_enqueue_req(exp, req, NULL, 0);
if (rc) {
ptlrpc_request_free(req);
GOTO(out, rc);
}
lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
lit->opc = (__u64)it_op;
req_qbody = req_capsule_client_get(&req->rq_pill, &RMF_QUOTA_BODY);
*req_qbody = *qbody;
req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER,
sizeof(*lvb));
ptlrpc_request_set_replen(req);
switch(it_op) {
case IT_QUOTA_CONN:
/* build resource name associated with global index */
fid_build_reg_res_name(&qbody->qb_fid, &qti->qti_resid);
/* copy einfo template and fill ei_cbdata with qqi pointer */
memcpy(&qti->qti_einfo, &qsd_glb_einfo, sizeof(qti->qti_einfo));
qti->qti_einfo.ei_cbdata = qqi;
/* don't cancel global lock on memory pressure */
flags |= LDLM_FL_NO_LRU;
break;
case IT_QUOTA_DQACQ:
/* build resource name associated for per-ID quota lock */
fid_build_quota_res_name(&qbody->qb_fid, &qbody->qb_id,
&qti->qti_resid);
/* copy einfo template and fill ei_cbdata with lqe pointer */
memcpy(&qti->qti_einfo, &qsd_id_einfo, sizeof(qti->qti_einfo));
qti->qti_einfo.ei_cbdata = arg;
break;
default:
LASSERTF(0, "invalid it_op %d", it_op);
}
/* build lock enqueue request */
rc = ldlm_cli_enqueue(exp, &req, &qti->qti_einfo, &qti->qti_resid, NULL,
&flags, (void *)lvb, sizeof(*lvb), LVB_T_LQUOTA,
&qti->qti_lockh, 1);
if (rc < 0) {
ptlrpc_req_finished(req);
GOTO(out, rc);
}
/* grab reference on backend structure for the new lock */
switch(it_op) {
case IT_QUOTA_CONN:
/* grab reference on qqi for new lock */
#ifdef USE_LU_REF
{
struct ldlm_lock *lock;
lock = ldlm_handle2lock(&qti->qti_lockh);
if (lock == NULL) {
ptlrpc_req_finished(req);
GOTO(out, rc = -ENOLCK);
}
lu_ref_add(&qqi->qqi_reference, "glb_lock", lock);
LDLM_LOCK_PUT(lock);
}
#endif
qqi_getref(qqi);
break;
case IT_QUOTA_DQACQ:
/* grab reference on lqe for new lock */
lqe_getref((struct lquota_entry *)arg);
/* all acquire/release request are sent with no_resend and
* no_delay flag */
req->rq_no_resend = req->rq_no_delay = 1;
break;
default:
break;
}
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 = arg;
aa->aa_lvb = lvb;
aa->aa_completion = completion;
lustre_handle_copy(&aa->aa_lockh, &qti->qti_lockh);
if (sync) {
/* send lock enqueue request and wait for completion */
rc = ptlrpc_queue_wait(req);
rc = qsd_intent_interpret(env, req, aa, rc);
ptlrpc_req_finished(req);
} else {
/* queue lock request and return */
req->rq_interpret_reply = qsd_intent_interpret;
ptlrpcd_add_req(req, PDL_POLICY_LOCAL, -1);
}
RETURN(rc);
out:
completion(env, qqi, qbody, NULL, &qti->qti_lockh, lvb, arg, rc);
return rc;
}
size_t count = sync->resync_count;
memcpy(req_capsule_client_get(&req->rq_pill, &RMF_U32),
op_data->op_data, count * sizeof(__u32));
}
}
}
void mdc_rename_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
const char *old, size_t oldlen,
const char *new, size_t newlen)
{
struct mdt_rec_rename *rec;
CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_rename));
rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
/* XXX do something about time, uid, gid */
rec->rn_opcode = REINT_RENAME;
rec->rn_fsuid = op_data->op_fsuid;
rec->rn_fsgid = op_data->op_fsgid;
rec->rn_cap = op_data->op_cap;
rec->rn_suppgid1 = op_data->op_suppgids[0];
rec->rn_suppgid2 = op_data->op_suppgids[1];
rec->rn_fid1 = op_data->op_fid1;
rec->rn_fid2 = op_data->op_fid2;
rec->rn_time = op_data->op_mod_time;
rec->rn_mode = op_data->op_mode;
rec->rn_bias = op_data->op_bias;
mdc_pack_name(req, &RMF_NAME, old, oldlen);
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;
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);
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_request_portal = SEQ_METADATA_PORTAL;
else
req->rq_request_portal = SEQ_DATA_PORTAL;
debug_mask = D_INFO;
}
ptlrpc_at_set_req_timeout(req);
if (seq->lcs_type == LUSTRE_SEQ_METADATA)
mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
rc = ptlrpc_queue_wait(req);
if (seq->lcs_type == LUSTRE_SEQ_METADATA)
mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
if (rc)
GOTO(out_req, rc);
out = req_capsule_server_get(&req->rq_pill, &RMF_SEQ_RANGE);
*output = *out;
if (!range_is_sane(output)) {
CERROR("%s: Invalid range received from server: "
DRANGE"\n", seq->lcs_name, PRANGE(output));
GOTO(out_req, rc = -EINVAL);
}
if (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;
}
int mdt_init_sec_level(struct mdt_thread_info *info)
{
struct mdt_device *mdt = info->mti_mdt;
struct ptlrpc_request *req = mdt_info_req(info);
char *client = libcfs_nid2str(req->rq_peer.nid);
struct obd_export *exp = req->rq_export;
struct obd_device *obd = exp->exp_obd;
struct obd_connect_data *data, *reply;
int rc = 0, remote;
ENTRY;
data = req_capsule_client_get(info->mti_pill, &RMF_CONNECT_DATA);
reply = req_capsule_server_get(info->mti_pill, &RMF_CONNECT_DATA);
if (data == NULL || reply == NULL)
RETURN(-EFAULT);
/* connection from MDT is always trusted */
if (req->rq_auth_usr_mdt) {
mdt_init_sec_none(reply, exp);
RETURN(0);
}
/* no GSS support case */
if (!req->rq_auth_gss) {
if (mdt->mdt_sec_level > LUSTRE_SEC_NONE) {
CWARN("client %s -> target %s does not user GSS, "
"can not run under security level %d.\n",
client, obd->obd_name, mdt->mdt_sec_level);
RETURN(-EACCES);
} else {
mdt_init_sec_none(reply, exp);
RETURN(0);
}
}
/* old version case */
if (unlikely(!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) ||
!(data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) ||
!(data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA))) {
if (mdt->mdt_sec_level > LUSTRE_SEC_NONE) {
CWARN("client %s -> target %s uses old version, "
"can not run under security level %d.\n",
client, obd->obd_name, mdt->mdt_sec_level);
RETURN(-EACCES);
} else {
CWARN("client %s -> target %s uses old version, "
"run under security level %d.\n",
client, obd->obd_name, mdt->mdt_sec_level);
mdt_init_sec_none(reply, exp);
RETURN(0);
}
}
remote = data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT_FORCE;
if (remote) {
if (!req->rq_auth_remote)
CDEBUG(D_SEC, "client (local realm) %s -> target %s "
"asked to be remote.\n", client, obd->obd_name);
} else if (req->rq_auth_remote) {
remote = 1;
CDEBUG(D_SEC, "client (remote realm) %s -> target %s is set "
"as remote by default.\n", client, obd->obd_name);
}
if (remote) {
if (!mdt->mdt_opts.mo_oss_capa) {
CDEBUG(D_SEC, "client %s -> target %s is set as remote,"
" but OSS capabilities are not enabled: %d.\n",
client, obd->obd_name, mdt->mdt_opts.mo_oss_capa);
RETURN(-EACCES);
}
} else {
if (req->rq_auth_uid == INVALID_UID) {
CDEBUG(D_SEC, "client %s -> target %s: user is not "
"authenticated!\n", client, obd->obd_name);
RETURN(-EACCES);
}
}
switch (mdt->mdt_sec_level) {
case LUSTRE_SEC_NONE:
if (!remote) {
mdt_init_sec_none(reply, exp);
break;
} else {
CDEBUG(D_SEC, "client %s -> target %s is set as remote, "
"can not run under security level %d.\n",
client, obd->obd_name, mdt->mdt_sec_level);
RETURN(-EACCES);
}
case LUSTRE_SEC_REMOTE:
if (!remote)
mdt_init_sec_none(reply, exp);
break;
case LUSTRE_SEC_ALL:
if (!remote) {
reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
OBD_CONNECT_RMT_CLIENT_FORCE);
if (!mdt->mdt_opts.mo_mds_capa)
reply->ocd_connect_flags &= ~OBD_CONNECT_MDS_CAPA;
if (!mdt->mdt_opts.mo_oss_capa)
reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
cfs_spin_lock(&exp->exp_lock);
exp->exp_connect_flags = reply->ocd_connect_flags;
cfs_spin_unlock(&exp->exp_lock);
}
break;
default:
RETURN(-EINVAL);
}
RETURN(rc);
}
static int osc_object_fiemap(const struct lu_env *env, struct cl_object *obj,
struct ll_fiemap_info_key *fmkey,
struct fiemap *fiemap, size_t *buflen)
{
struct obd_export *exp = osc_export(cl2osc(obj));
struct ldlm_res_id resid;
union ldlm_policy_data policy;
struct lustre_handle lockh;
enum ldlm_mode mode = LCK_MINMODE;
struct ptlrpc_request *req;
struct fiemap *reply;
char *tmp;
int rc;
ENTRY;
fmkey->lfik_oa.o_oi = cl2osc(obj)->oo_oinfo->loi_oi;
if (!(fmkey->lfik_fiemap.fm_flags & FIEMAP_FLAG_SYNC))
goto skip_locking;
policy.l_extent.start = fmkey->lfik_fiemap.fm_start & PAGE_CACHE_MASK;
if (OBD_OBJECT_EOF - fmkey->lfik_fiemap.fm_length <=
fmkey->lfik_fiemap.fm_start + PAGE_CACHE_SIZE - 1)
policy.l_extent.end = OBD_OBJECT_EOF;
else
policy.l_extent.end = (fmkey->lfik_fiemap.fm_start +
fmkey->lfik_fiemap.fm_length +
PAGE_CACHE_SIZE - 1) & PAGE_CACHE_MASK;
ostid_build_res_name(&fmkey->lfik_oa.o_oi, &resid);
mode = ldlm_lock_match(exp->exp_obd->obd_namespace,
LDLM_FL_BLOCK_GRANTED | LDLM_FL_LVB_READY,
&resid, LDLM_EXTENT, &policy,
LCK_PR | LCK_PW, &lockh, 0);
if (mode) { /* lock is cached on client */
if (mode != LCK_PR) {
ldlm_lock_addref(&lockh, LCK_PR);
ldlm_lock_decref(&lockh, LCK_PW);
}
} else { /* no cached lock, needs acquire lock on server side */
fmkey->lfik_oa.o_valid |= OBD_MD_FLFLAGS;
fmkey->lfik_oa.o_flags |= OBD_FL_SRVLOCK;
}
skip_locking:
req = ptlrpc_request_alloc(class_exp2cliimp(exp),
&RQF_OST_GET_INFO_FIEMAP);
if (req == NULL)
GOTO(drop_lock, rc = -ENOMEM);
req_capsule_set_size(&req->rq_pill, &RMF_FIEMAP_KEY, RCL_CLIENT,
sizeof(*fmkey));
req_capsule_set_size(&req->rq_pill, &RMF_FIEMAP_VAL, RCL_CLIENT,
*buflen);
req_capsule_set_size(&req->rq_pill, &RMF_FIEMAP_VAL, RCL_SERVER,
*buflen);
rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_GET_INFO);
if (rc != 0) {
ptlrpc_request_free(req);
GOTO(drop_lock, rc);
}
tmp = req_capsule_client_get(&req->rq_pill, &RMF_FIEMAP_KEY);
memcpy(tmp, fmkey, sizeof(*fmkey));
tmp = req_capsule_client_get(&req->rq_pill, &RMF_FIEMAP_VAL);
memcpy(tmp, fiemap, *buflen);
ptlrpc_request_set_replen(req);
rc = ptlrpc_queue_wait(req);
if (rc != 0)
GOTO(fini_req, rc);
reply = req_capsule_server_get(&req->rq_pill, &RMF_FIEMAP_VAL);
if (reply == NULL)
GOTO(fini_req, rc = -EPROTO);
memcpy(fiemap, reply, *buflen);
fini_req:
ptlrpc_req_finished(req);
drop_lock:
if (mode)
ldlm_lock_decref(&lockh, LCK_PR);
RETURN(rc);
}
/**
* Object updates between Targets. Because all the updates has been
* dis-assemblied into object updates in master MDD layer, so out
* will skip MDD layer, and call OSD API directly to execute these
* updates.
*
* In phase I, all of the updates in the request need to be executed
* in one transaction, and the transaction has to be synchronously.
*
* Please refer to lustre/include/lustre/lustre_idl.h for req/reply
* format.
*/
int out_handle(struct mdt_thread_info *info)
{
struct req_capsule *pill = info->mti_pill;
struct update_buf *ubuf;
struct update *update;
struct thandle_exec_args *th = &info->mti_handle;
int bufsize;
int count;
unsigned off;
int i;
int rc = 0;
int rc1 = 0;
ENTRY;
req_capsule_set(pill, &RQF_UPDATE_OBJ);
bufsize = req_capsule_get_size(pill, &RMF_UPDATE, RCL_CLIENT);
if (bufsize != UPDATE_BUFFER_SIZE) {
CERROR("%s: invalid bufsize %d: rc = %d\n",
mdt_obd_name(info->mti_mdt), bufsize, -EPROTO);
RETURN(err_serious(-EPROTO));
}
ubuf = req_capsule_client_get(pill, &RMF_UPDATE);
if (ubuf == NULL) {
CERROR("%s: No buf!: rc = %d\n", mdt_obd_name(info->mti_mdt),
-EPROTO);
RETURN(err_serious(-EPROTO));
}
if (le32_to_cpu(ubuf->ub_magic) != UPDATE_BUFFER_MAGIC) {
CERROR("%s: invalid magic %x expect %x: rc = %d\n",
mdt_obd_name(info->mti_mdt), le32_to_cpu(ubuf->ub_magic),
UPDATE_BUFFER_MAGIC, -EPROTO);
RETURN(err_serious(-EPROTO));
}
count = le32_to_cpu(ubuf->ub_count);
if (count <= 0) {
CERROR("%s: No update!: rc = %d\n",
mdt_obd_name(info->mti_mdt), -EPROTO);
RETURN(err_serious(-EPROTO));
}
req_capsule_set_size(pill, &RMF_UPDATE_REPLY, RCL_SERVER,
UPDATE_BUFFER_SIZE);
rc = req_capsule_server_pack(pill);
if (rc != 0) {
CERROR("%s: Can't pack response: rc = %d\n",
mdt_obd_name(info->mti_mdt), rc);
RETURN(rc);
}
/* Prepare the update reply buffer */
info->mti_u.update.mti_update_reply =
req_capsule_server_get(pill, &RMF_UPDATE_REPLY);
update_init_reply_buf(info->mti_u.update.mti_update_reply, count);
rc = out_tx_start(info->mti_env, info->mti_mdt, th);
if (rc != 0)
RETURN(rc);
/* Walk through updates in the request to execute them synchronously */
off = cfs_size_round(offsetof(struct update_buf, ub_bufs[0]));
for (i = 0; i < count; i++) {
struct out_handler *h;
struct dt_object *dt_obj;
update = (struct update *)((char *)ubuf + off);
fid_le_to_cpu(&update->u_fid, &update->u_fid);
if (!fid_is_sane(&update->u_fid)) {
CERROR("%s: invalid FID "DFID": rc = %d\n",
mdt_obd_name(info->mti_mdt),
PFID(&update->u_fid), -EPROTO);
GOTO(out, rc = err_serious(-EPROTO));
}
dt_obj = out_object_find(info, &update->u_fid);
if (IS_ERR(dt_obj))
GOTO(out, rc = PTR_ERR(dt_obj));
info->mti_u.update.mti_dt_object = dt_obj;
info->mti_u.update.mti_update = update;
info->mti_u.update.mti_update_reply_index = i;
h = mdt_handler_find(update->u_type, out_handlers);
if (likely(h != NULL)) {
rc = h->mh_act(info);
} else {
CERROR("%s: The unsupported opc: 0x%x\n",
mdt_obd_name(info->mti_mdt), update->u_type);
lu_object_put(info->mti_env, &dt_obj->do_lu);
GOTO(out, rc = -ENOTSUPP);
}
lu_object_put(info->mti_env, &dt_obj->do_lu);
if (rc < 0)
GOTO(out, rc);
off += cfs_size_round(update_size(update));
}
out:
rc1 = out_tx_end(info, th);
rc = rc == 0 ? rc1 : rc;
info->mti_fail_id = OBD_FAIL_UPDATE_OBJ_NET;
RETURN(rc);
}
static int mdt_open_unpack(struct mdt_thread_info *info)
{
struct lu_ucred *uc = mdt_ucred(info);
struct mdt_rec_create *rec;
struct lu_attr *attr = &info->mti_attr.ma_attr;
struct req_capsule *pill = info->mti_pill;
struct mdt_reint_record *rr = &info->mti_rr;
struct ptlrpc_request *req = mdt_info_req(info);
struct md_op_spec *sp = &info->mti_spec;
ENTRY;
CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
rec = req_capsule_client_get(pill, &RMF_REC_REINT);
if (rec == NULL)
RETURN(-EFAULT);
/* This prior initialization is needed for old_init_ucred_reint() */
uc->uc_fsuid = rec->cr_fsuid;
uc->uc_fsgid = rec->cr_fsgid;
uc->uc_cap = rec->cr_cap;
uc->uc_suppgids[0] = rec->cr_suppgid1;
uc->uc_suppgids[1] = rec->cr_suppgid2;
uc->uc_umask = rec->cr_umask;
rr->rr_fid1 = &rec->cr_fid1;
rr->rr_fid2 = &rec->cr_fid2;
rr->rr_handle = &rec->cr_old_handle;
attr->la_mode = rec->cr_mode;
attr->la_rdev = rec->cr_rdev;
attr->la_uid = rec->cr_fsuid;
attr->la_gid = rec->cr_fsgid;
attr->la_ctime = rec->cr_time;
attr->la_mtime = rec->cr_time;
attr->la_atime = rec->cr_time;
attr->la_valid = LA_MODE | LA_RDEV | LA_UID | LA_GID |
LA_CTIME | LA_MTIME | LA_ATIME;
memset(&info->mti_spec.u, 0, sizeof(info->mti_spec.u));
info->mti_spec.sp_cr_flags = get_mrc_cr_flags(rec);
/* Do not trigger ASSERTION if client miss to set such flags. */
if (unlikely(info->mti_spec.sp_cr_flags == 0))
RETURN(-EPROTO);
info->mti_replayepoch = rec->cr_ioepoch;
info->mti_cross_ref = !!(rec->cr_bias & MDS_CROSS_REF);
if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
mdt_set_capainfo(info, 0, rr->rr_fid1,
req_capsule_client_get(pill, &RMF_CAPA1));
if (req_is_replay(req) &&
req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT)) {
#if 0
mdt_set_capainfo(info, 1, rr->rr_fid2,
req_capsule_client_get(pill, &RMF_CAPA2));
#else
/*
* FIXME: capa in replay open request might have expired,
* bypass capa check. Security hole?
*/
mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
mdt_set_capainfo(info, 1, rr->rr_fid2, BYPASS_CAPA);
#endif
}
mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, MNF_FIX_ANON);
if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
RCL_CLIENT);
if (rr->rr_eadatalen > 0) {
rr->rr_eadata = req_capsule_client_get(pill,
&RMF_EADATA);
sp->u.sp_ea.eadatalen = rr->rr_eadatalen;
sp->u.sp_ea.eadata = rr->rr_eadata;
sp->no_create = !!req_is_replay(req);
mdt_fix_lov_magic(info);
}
/*
* Client default md_size may be 0 right after client start,
* until all osc are connected, set here just some reasonable
* value to prevent misbehavior.
*/
if (rr->rr_eadatalen == 0 &&
!(info->mti_spec.sp_cr_flags & MDS_OPEN_DELAY_CREATE))
rr->rr_eadatalen = MIN_MD_SIZE;
}
RETURN(0);
}
/* TODO: handle requests in a similar way as MDT: see mdt_handle_common() */
static int ldlm_callback_handler(struct ptlrpc_request *req)
{
struct ldlm_namespace *ns;
struct ldlm_request *dlm_req;
struct ldlm_lock *lock;
int rc;
/* Requests arrive in sender's byte order. The ptlrpc service
* handler has already checked and, if necessary, byte-swapped the
* incoming request message body, but I am responsible for the
* message buffers. */
/* do nothing for sec context finalize */
if (lustre_msg_get_opc(req->rq_reqmsg) == SEC_CTX_FINI)
return 0;
req_capsule_init(&req->rq_pill, req, RCL_SERVER);
if (req->rq_export == NULL) {
rc = ldlm_callback_reply(req, -ENOTCONN);
ldlm_callback_errmsg(req, "Operate on unconnected server",
rc, NULL);
return 0;
}
LASSERT(req->rq_export != NULL);
LASSERT(req->rq_export->exp_obd != NULL);
switch (lustre_msg_get_opc(req->rq_reqmsg)) {
case LDLM_BL_CALLBACK:
if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_BL_CALLBACK_NET))
return 0;
break;
case LDLM_CP_CALLBACK:
if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CP_CALLBACK_NET))
return 0;
break;
case LDLM_GL_CALLBACK:
if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_GL_CALLBACK_NET))
return 0;
break;
case LDLM_SET_INFO:
rc = ldlm_handle_setinfo(req);
ldlm_callback_reply(req, rc);
return 0;
case OBD_LOG_CANCEL: /* remove this eventually - for 1.4.0 compat */
CERROR("shouldn't be handling OBD_LOG_CANCEL on DLM thread\n");
req_capsule_set(&req->rq_pill, &RQF_LOG_CANCEL);
if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOG_CANCEL_NET))
return 0;
rc = llog_origin_handle_cancel(req);
if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOG_CANCEL_REP))
return 0;
ldlm_callback_reply(req, rc);
return 0;
case LLOG_ORIGIN_HANDLE_CREATE:
req_capsule_set(&req->rq_pill, &RQF_LLOG_ORIGIN_HANDLE_CREATE);
if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOGD_NET))
return 0;
rc = llog_origin_handle_open(req);
ldlm_callback_reply(req, rc);
return 0;
case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
req_capsule_set(&req->rq_pill,
&RQF_LLOG_ORIGIN_HANDLE_NEXT_BLOCK);
if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOGD_NET))
return 0;
rc = llog_origin_handle_next_block(req);
ldlm_callback_reply(req, rc);
return 0;
case LLOG_ORIGIN_HANDLE_READ_HEADER:
req_capsule_set(&req->rq_pill,
&RQF_LLOG_ORIGIN_HANDLE_READ_HEADER);
if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOGD_NET))
return 0;
rc = llog_origin_handle_read_header(req);
ldlm_callback_reply(req, rc);
return 0;
case LLOG_ORIGIN_HANDLE_CLOSE:
if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOGD_NET))
return 0;
rc = llog_origin_handle_close(req);
ldlm_callback_reply(req, rc);
return 0;
case OBD_QC_CALLBACK:
req_capsule_set(&req->rq_pill, &RQF_QC_CALLBACK);
if (OBD_FAIL_CHECK(OBD_FAIL_OBD_QC_CALLBACK_NET))
return 0;
rc = ldlm_handle_qc_callback(req);
ldlm_callback_reply(req, rc);
return 0;
default:
CERROR("unknown opcode %u\n",
lustre_msg_get_opc(req->rq_reqmsg));
ldlm_callback_reply(req, -EPROTO);
return 0;
}
ns = req->rq_export->exp_obd->obd_namespace;
LASSERT(ns != NULL);
req_capsule_set(&req->rq_pill, &RQF_LDLM_CALLBACK);
dlm_req = req_capsule_client_get(&req->rq_pill, &RMF_DLM_REQ);
if (dlm_req == NULL) {
rc = ldlm_callback_reply(req, -EPROTO);
ldlm_callback_errmsg(req, "Operate without parameter", rc,
NULL);
return 0;
}
/* Force a known safe race, send a cancel to the server for a lock
* which the server has already started a blocking callback on. */
if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CANCEL_BL_CB_RACE) &&
lustre_msg_get_opc(req->rq_reqmsg) == LDLM_BL_CALLBACK) {
rc = ldlm_cli_cancel(&dlm_req->lock_handle[0], 0);
if (rc < 0)
CERROR("ldlm_cli_cancel: %d\n", rc);
}
lock = ldlm_handle2lock_long(&dlm_req->lock_handle[0], 0);
if (!lock) {
CDEBUG(D_DLMTRACE, "callback on lock "LPX64" - lock "
"disappeared\n", dlm_req->lock_handle[0].cookie);
rc = ldlm_callback_reply(req, -EINVAL);
ldlm_callback_errmsg(req, "Operate with invalid parameter", rc,
&dlm_req->lock_handle[0]);
return 0;
}
if ((lock->l_flags & LDLM_FL_FAIL_LOC) &&
lustre_msg_get_opc(req->rq_reqmsg) == LDLM_BL_CALLBACK)
OBD_RACE(OBD_FAIL_LDLM_CP_BL_RACE);
/* Copy hints/flags (e.g. LDLM_FL_DISCARD_DATA) from AST. */
lock_res_and_lock(lock);
lock->l_flags |= ldlm_flags_from_wire(dlm_req->lock_flags &
LDLM_AST_FLAGS);
if (lustre_msg_get_opc(req->rq_reqmsg) == LDLM_BL_CALLBACK) {
/* If somebody cancels lock and cache is already dropped,
* or lock is failed before cp_ast received on client,
* we can tell the server we have no lock. Otherwise, we
* should send cancel after dropping the cache. */
if (((lock->l_flags & LDLM_FL_CANCELING) &&
(lock->l_flags & LDLM_FL_BL_DONE)) ||
(lock->l_flags & LDLM_FL_FAILED)) {
LDLM_DEBUG(lock, "callback on lock "
LPX64" - lock disappeared\n",
dlm_req->lock_handle[0].cookie);
unlock_res_and_lock(lock);
LDLM_LOCK_RELEASE(lock);
rc = ldlm_callback_reply(req, -EINVAL);
ldlm_callback_errmsg(req, "Operate on stale lock", rc,
&dlm_req->lock_handle[0]);
return 0;
}
/* BL_AST locks are not needed in LRU.
* Let ldlm_cancel_lru() be fast. */
ldlm_lock_remove_from_lru(lock);
lock->l_flags |= LDLM_FL_BL_AST;
}
unlock_res_and_lock(lock);
/* We want the ost thread to get this reply so that it can respond
* to ost requests (write cache writeback) that might be triggered
* in the callback.
*
* But we'd also like to be able to indicate in the reply that we're
* cancelling right now, because it's unused, or have an intent result
* in the reply, so we might have to push the responsibility for sending
* the reply down into the AST handlers, alas. */
switch (lustre_msg_get_opc(req->rq_reqmsg)) {
case LDLM_BL_CALLBACK:
CDEBUG(D_INODE, "blocking ast\n");
req_capsule_extend(&req->rq_pill, &RQF_LDLM_BL_CALLBACK);
if (!(lock->l_flags & LDLM_FL_CANCEL_ON_BLOCK)) {
rc = ldlm_callback_reply(req, 0);
if (req->rq_no_reply || rc)
ldlm_callback_errmsg(req, "Normal process", rc,
&dlm_req->lock_handle[0]);
}
if (ldlm_bl_to_thread_lock(ns, &dlm_req->lock_desc, lock))
ldlm_handle_bl_callback(ns, &dlm_req->lock_desc, lock);
break;
case LDLM_CP_CALLBACK:
CDEBUG(D_INODE, "completion ast\n");
req_capsule_extend(&req->rq_pill, &RQF_LDLM_CP_CALLBACK);
ldlm_callback_reply(req, 0);
ldlm_handle_cp_callback(req, ns, dlm_req, lock);
break;
case LDLM_GL_CALLBACK:
CDEBUG(D_INODE, "glimpse ast\n");
req_capsule_extend(&req->rq_pill, &RQF_LDLM_GL_CALLBACK);
ldlm_handle_gl_callback(req, ns, dlm_req, lock);
break;
default:
LBUG(); /* checked above */
}
return 0;
}
