* GPL HEADER END
*/
/*
- * Copyright 2008 Sun Microsystems, Inc. All rights reserved
+ * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*/
/*
/**
* Do not return server-side uid/gid to remote client
*/
-static void ost_drop_id(struct obd_export *exp, struct obdo *oa)
+static void ost_drop_id(struct obd_export *exp, struct obdo *oa)
{
if (exp_connect_rmtclient(exp)) {
oa->o_uid = -1;
}
}
+/**
+ * Validate oa from client.
+ * 1. If the request comes from 1.8 clients, it will reset o_seq with MDT0.
+ * 2. If the request comes from 2.0 clients, currently only RSVD seq and IDIF
+ * req are valid.
+ * a. for single MDS seq = FID_SEQ_OST_MDT0,
+ * b. for CMD, seq = FID_SEQ_OST_MDT0, FID_SEQ_OST_MDT1 - FID_SEQ_OST_MAX
+ */
+static int ost_validate_obdo(struct obd_export *exp, struct obdo *oa,
+ struct obd_ioobj *ioobj)
+{
+ if (oa != NULL && (!(oa->o_valid & OBD_MD_FLGROUP) ||
+ !(exp->exp_connect_flags & OBD_CONNECT_FULL20))) {
+ oa->o_seq = FID_SEQ_OST_MDT0;
+ if (ioobj)
+ ioobj->ioo_seq = FID_SEQ_OST_MDT0;
+ /* remove fid_seq_is_rsvd() after FID-on-OST allows SEQ > 9 */
+ } else if (oa == NULL ||
+ !(fid_seq_is_rsvd(oa->o_seq) || fid_seq_is_idif(oa->o_seq))) {
+ CERROR("%s: client %s sent invalid object "POSTID"\n",
+ exp->exp_obd->obd_name, obd_export_nid2str(exp),
+ oa ? oa->o_id : -1, oa ? oa->o_seq : -1);
+ return -EPROTO;
+ }
+ obdo_from_ostid(oa, &oa->o_oi);
+ if (ioobj)
+ ioobj_from_obdo(ioobj, oa);
+ return 0;
+}
+
void oti_to_request(struct obd_trans_info *oti, struct ptlrpc_request *req)
{
struct oti_req_ack_lock *ack_lock;
if (oti == NULL)
return;
- if (req->rq_repmsg)
+ if (req->rq_repmsg) {
+ __u64 versions[PTLRPC_NUM_VERSIONS] = { 0 };
lustre_msg_set_transno(req->rq_repmsg, oti->oti_transno);
+ versions[0] = oti->oti_pre_version;
+ lustre_msg_set_versions(req->rq_repmsg, versions);
+ }
req->rq_transno = oti->oti_transno;
/* XXX 4 == entries in oti_ack_locks??? */
if (body->oa.o_id == 0)
RETURN(-EPROTO);
+ rc = ost_validate_obdo(exp, &body->oa, NULL);
+ if (rc)
+ RETURN(rc);
+
/* If there's a DLM request, cancel the locks mentioned in it*/
if (req_capsule_field_present(&req->rq_pill, &RMF_DLM_REQ, RCL_CLIENT)) {
struct ldlm_request *dlm;
RETURN(0);
}
+/**
+ * Helper function for getting server side [start, start+count] DLM lock
+ * if asked by client.
+ */
+static int ost_lock_get(struct obd_export *exp, struct obdo *oa,
+ __u64 start, __u64 count, struct lustre_handle *lh,
+ int mode, int flags)
+{
+ struct ldlm_res_id res_id;
+ ldlm_policy_data_t policy;
+ __u64 end = start + count;
+
+ ENTRY;
+
+ LASSERT(!lustre_handle_is_used(lh));
+ /* o_id and o_gr are used for localizing resource, if client miss to set
+ * them, do not trigger ASSERTION. */
+ if (unlikely((oa->o_valid & (OBD_MD_FLID | OBD_MD_FLGROUP)) !=
+ (OBD_MD_FLID | OBD_MD_FLGROUP)))
+ RETURN(-EPROTO);
+
+ if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
+ !(oa->o_flags & OBD_FL_SRVLOCK))
+ RETURN(0);
+
+ osc_build_res_name(oa->o_id, oa->o_seq, &res_id);
+ CDEBUG(D_INODE, "OST-side extent lock.\n");
+
+ policy.l_extent.start = start & CFS_PAGE_MASK;
+
+ /* If ->o_blocks is EOF it means "lock till the end of the
+ * file". Otherwise, it's size of a hole being punched (in bytes) */
+ if (count == OBD_OBJECT_EOF || end < start)
+ policy.l_extent.end = OBD_OBJECT_EOF;
+ else
+ policy.l_extent.end = end | ~CFS_PAGE_MASK;
+
+ RETURN(ldlm_cli_enqueue_local(exp->exp_obd->obd_namespace, &res_id,
+ LDLM_EXTENT, &policy, mode, &flags,
+ ldlm_blocking_ast, ldlm_completion_ast,
+ ldlm_glimpse_ast, NULL, 0, NULL, lh));
+}
+
+/* Helper function: release lock, if any. */
+static void ost_lock_put(struct obd_export *exp,
+ struct lustre_handle *lh, int mode)
+{
+ ENTRY;
+ if (lustre_handle_is_used(lh))
+ ldlm_lock_decref(lh, mode);
+ EXIT;
+}
+
static int ost_getattr(struct obd_export *exp, struct ptlrpc_request *req)
{
struct ost_body *body, *repbody;
- struct obd_info oinfo = { { { 0 } } };
+ struct obd_info *oinfo;
+ struct lustre_handle lh = { 0 };
+ struct lustre_capa *capa = NULL;
int rc;
ENTRY;
if (body == NULL)
RETURN(-EFAULT);
+ rc = ost_validate_obdo(exp, &body->oa, NULL);
+ if (rc)
+ RETURN(rc);
+
rc = req_capsule_server_pack(&req->rq_pill);
if (rc)
RETURN(rc);
- repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
- repbody->oa = body->oa;
+ rc = ost_lock_get(exp, &body->oa, 0, OBD_OBJECT_EOF, &lh, LCK_PR, 0);
+ if (rc)
+ RETURN(rc);
- oinfo.oi_oa = &repbody->oa;
- if (oinfo.oi_oa->o_valid & OBD_MD_FLOSSCAPA) {
- oinfo.oi_capa = req_capsule_client_get(&req->rq_pill,
- &RMF_CAPA1);
- if (oinfo.oi_capa == NULL) {
+ if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
+ capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
+ if (capa == NULL) {
CERROR("Missing capability for OST GETATTR");
RETURN (-EFAULT);
}
}
- req->rq_status = obd_getattr(exp, &oinfo);
+ OBD_ALLOC_PTR(oinfo);
+ if (!oinfo)
+ RETURN(-ENOMEM);
+ oinfo->oi_oa = &body->oa;
+ oinfo->oi_capa = capa;
+
+ req->rq_status = obd_getattr(exp, oinfo);
+
+ OBD_FREE_PTR(oinfo);
+
+ repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
+ repbody->oa = body->oa;
+
+ ost_lock_put(exp, &lh, LCK_PR);
+
ost_drop_id(exp, &repbody->oa);
RETURN(0);
}
osfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
req->rq_status = obd_statfs(req->rq_export->exp_obd, osfs,
- cfs_time_current_64() - HZ, 0);
- if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
- osfs->os_bfree = osfs->os_bavail = 64;
+ cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
+ 0);
if (req->rq_status != 0)
CERROR("ost: statfs failed: rc %d\n", req->rq_status);
if (body == NULL)
RETURN(-EFAULT);
+ rc = ost_validate_obdo(req->rq_export, &body->oa, NULL);
+ if (rc)
+ RETURN(rc);
+
rc = req_capsule_server_pack(&req->rq_pill);
if (rc)
RETURN(rc);
repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
- memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
- oti->oti_logcookies = &repbody->oa.o_lcookie;
+ repbody->oa = body->oa;
+ oti->oti_logcookies = &body->oa.o_lcookie;
req->rq_status = obd_create(exp, &repbody->oa, NULL, oti);
//obd_log_cancel(conn, NULL, 1, oti->oti_logcookies, 0);
RETURN(0);
}
-/**
- * Helper function for ost_punch(): if asked by client, acquire [size, EOF]
- * lock on the file being truncated.
- */
-static int ost_punch_lock_get(struct obd_export *exp, struct obdo *oa,
- struct lustre_handle *lh)
-{
- int flags;
- struct ldlm_res_id res_id;
- ldlm_policy_data_t policy;
- __u64 start;
- __u64 finis;
-
- ENTRY;
-
- osc_build_res_name(oa->o_id, oa->o_gr, &res_id);
- LASSERT(!lustre_handle_is_used(lh));
-
- if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
- !(oa->o_flags & OBD_FL_TRUNCLOCK))
- RETURN(0);
-
- CDEBUG(D_INODE, "OST-side truncate lock.\n");
-
- start = oa->o_size;
- finis = start + oa->o_blocks;
-
- /*
- * standard truncate optimization: if file body is completely
- * destroyed, don't send data back to the server.
- */
- flags = (start == 0) ? LDLM_AST_DISCARD_DATA : 0;
-
- policy.l_extent.start = start & CFS_PAGE_MASK;
-
- /*
- * If ->o_blocks is EOF it means "lock till the end of the
- * file". Otherwise, it's size of a hole being punched (in bytes)
- */
- if (oa->o_blocks == OBD_OBJECT_EOF || finis < start)
- policy.l_extent.end = OBD_OBJECT_EOF;
- else
- policy.l_extent.end = finis | ~CFS_PAGE_MASK;
-
- RETURN(ldlm_cli_enqueue_local(exp->exp_obd->obd_namespace, &res_id,
- LDLM_EXTENT, &policy, LCK_PW, &flags,
- ldlm_blocking_ast, ldlm_completion_ast,
- ldlm_glimpse_ast, NULL, 0, NULL, lh));
-}
-
-/**
- * Helper function for ost_punch(): release lock acquired by
- * ost_punch_lock_get(), if any.
- */
-static void ost_punch_lock_put(struct obd_export *exp, struct obdo *oa,
- struct lustre_handle *lh)
-{
- ENTRY;
- if (lustre_handle_is_used(lh))
- ldlm_lock_decref(lh, LCK_PW);
- EXIT;
-}
-
static int ost_punch(struct obd_export *exp, struct ptlrpc_request *req,
struct obd_trans_info *oti)
{
- struct obd_info oinfo = { { { 0 } } };
struct ost_body *body, *repbody;
- int rc;
+ int rc, flags = 0;
struct lustre_handle lh = {0,};
ENTRY;
if (body == NULL)
RETURN(-EFAULT);
- oinfo.oi_oa = &body->oa;
- oinfo.oi_policy.l_extent.start = oinfo.oi_oa->o_size;
- oinfo.oi_policy.l_extent.end = oinfo.oi_oa->o_blocks;
+ rc = ost_validate_obdo(exp, &body->oa, NULL);
+ if (rc)
+ RETURN(rc);
- if ((oinfo.oi_oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
+ if ((body->oa.o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
(OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
RETURN(-EPROTO);
if (rc)
RETURN(rc);
- repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
- rc = ost_punch_lock_get(exp, oinfo.oi_oa, &lh);
+ /* standard truncate optimization: if file body is completely
+ * destroyed, don't send data back to the server. */
+ if (body->oa.o_size == 0)
+ flags |= LDLM_AST_DISCARD_DATA;
+
+ rc = ost_lock_get(exp, &body->oa, body->oa.o_size, body->oa.o_blocks,
+ &lh, LCK_PW, flags);
if (rc == 0) {
- if (oinfo.oi_oa->o_valid & OBD_MD_FLFLAGS &&
- oinfo.oi_oa->o_flags == OBD_FL_TRUNCLOCK)
+ struct obd_info *oinfo;
+ struct lustre_capa *capa = NULL;
+
+ if (body->oa.o_valid & OBD_MD_FLFLAGS &&
+ body->oa.o_flags == OBD_FL_SRVLOCK)
/*
- * If OBD_FL_TRUNCLOCK is the only bit set in
+ * If OBD_FL_SRVLOCK is the only bit set in
* ->o_flags, clear OBD_MD_FLFLAGS to avoid falling
* through filter_setattr() to filter_iocontrol().
*/
- oinfo.oi_oa->o_valid &= ~OBD_MD_FLFLAGS;
+ body->oa.o_valid &= ~OBD_MD_FLFLAGS;
- if (oinfo.oi_oa->o_valid & OBD_MD_FLOSSCAPA) {
- oinfo.oi_capa = req_capsule_client_get(&req->rq_pill,
- &RMF_CAPA1);
- if (oinfo.oi_capa == NULL) {
+ if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
+ capa = req_capsule_client_get(&req->rq_pill,
+ &RMF_CAPA1);
+ if (capa == NULL) {
CERROR("Missing capability for OST PUNCH");
RETURN (-EFAULT);
}
}
- req->rq_status = obd_punch(exp, &oinfo, oti, NULL);
- ost_punch_lock_put(exp, oinfo.oi_oa, &lh);
+
+ OBD_ALLOC_PTR(oinfo);
+ if (!oinfo)
+ RETURN(-ENOMEM);
+ oinfo->oi_oa = &body->oa;
+ oinfo->oi_policy.l_extent.start = oinfo->oi_oa->o_size;
+ oinfo->oi_policy.l_extent.end = oinfo->oi_oa->o_blocks;
+ oinfo->oi_capa = capa;
+
+ req->rq_status = obd_punch(exp, oinfo, oti, NULL);
+ OBD_FREE_PTR(oinfo);
+ ost_lock_put(exp, &lh, LCK_PW);
}
- repbody->oa = *oinfo.oi_oa;
+
+ repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
+ repbody->oa = body->oa;
ost_drop_id(exp, &repbody->oa);
RETURN(rc);
}
static int ost_sync(struct obd_export *exp, struct ptlrpc_request *req)
{
struct ost_body *body, *repbody;
+ struct obd_info *oinfo;
struct lustre_capa *capa = NULL;
int rc;
ENTRY;
if (body == NULL)
RETURN(-EFAULT);
+ rc = ost_validate_obdo(exp, &body->oa, NULL);
+ if (rc)
+ RETURN(rc);
+
if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
if (capa == NULL) {
if (rc)
RETURN(rc);
+ OBD_ALLOC_PTR(oinfo);
+ if (!oinfo)
+ RETURN(-ENOMEM);
+
+ oinfo->oi_oa = &body->oa;
+ oinfo->oi_capa = capa;
+ req->rq_status = obd_sync(exp, oinfo, body->oa.o_size,
+ body->oa.o_blocks, NULL);
+ OBD_FREE_PTR(oinfo);
+
repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
- memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
- req->rq_status = obd_sync(exp, &repbody->oa, NULL, repbody->oa.o_size,
- repbody->oa.o_blocks, capa);
+ repbody->oa = body->oa;
ost_drop_id(exp, &repbody->oa);
RETURN(0);
}
struct obd_trans_info *oti)
{
struct ost_body *body, *repbody;
+ struct obd_info *oinfo;
+ struct lustre_capa *capa = NULL;
int rc;
- struct obd_info oinfo = { { { 0 } } };
ENTRY;
body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
if (body == NULL)
RETURN(-EFAULT);
- rc = req_capsule_server_pack(&req->rq_pill);
+ rc = ost_validate_obdo(req->rq_export, &body->oa, NULL);
if (rc)
RETURN(rc);
- repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
- repbody->oa = body->oa;
+ rc = req_capsule_server_pack(&req->rq_pill);
+ if (rc)
+ RETURN(rc);
- oinfo.oi_oa = &repbody->oa;
- if (oinfo.oi_oa->o_valid & OBD_MD_FLOSSCAPA) {
- oinfo.oi_capa = req_capsule_client_get(&req->rq_pill,
- &RMF_CAPA1);
- if (oinfo.oi_capa == NULL) {
+ if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
+ capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
+ if (capa == NULL) {
CERROR("Missing capability for OST SETATTR");
RETURN (-EFAULT);
}
}
- req->rq_status = obd_setattr(exp, &oinfo, oti);
+
+ OBD_ALLOC_PTR(oinfo);
+ if (!oinfo)
+ RETURN(-ENOMEM);
+ oinfo->oi_oa = &body->oa;
+ oinfo->oi_capa = capa;
+
+ req->rq_status = obd_setattr(exp, oinfo, oti);
+
+ OBD_FREE_PTR(oinfo);
+
+ repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
+ repbody->oa = body->oa;
ost_drop_id(exp, &repbody->oa);
RETURN(0);
}
int i;
ENTRY;
- osc_build_res_name(obj->ioo_id, obj->ioo_gr, &res_id);
+ osc_build_res_name(obj->ioo_id, obj->ioo_seq, &res_id);
LASSERT(mode == LCK_PR || mode == LCK_PW);
LASSERT(!lustre_handle_is_used(lh));
struct ost_prolong_data opd = { 0 };
ENTRY;
- osc_build_res_name(obj->ioo_id, obj->ioo_gr, &res_id);
+ osc_build_res_name(obj->ioo_id, obj->ioo_seq, &res_id);
opd.opd_mode = mode;
opd.opd_exp = req->rq_export;
RETURN(opd.opd_lock_match);
}
+/* Allocate thread local buffers if needed */
+static struct ost_thread_local_cache *ost_tls_get(struct ptlrpc_request *r)
+{
+ struct ost_thread_local_cache *tls =
+ (struct ost_thread_local_cache *)(r->rq_svc_thread->t_data);
+
+ /* In normal mode of operation an I/O request is serviced only
+ * by ll_ost_io threads each of them has own tls buffers allocated by
+ * ost_thread_init().
+ * During recovery, an I/O request may be queued until any of the ost
+ * service threads process it. Not necessary it should be one of
+ * ll_ost_io threads. In that case we dynamically allocating tls
+ * buffers for the request service time. */
+ if (unlikely(tls == NULL)) {
+ LASSERT(r->rq_export->exp_in_recovery);
+ OBD_ALLOC_PTR(tls);
+ if (tls != NULL) {
+ tls->temporary = 1;
+ r->rq_svc_thread->t_data = tls;
+ }
+ }
+ return tls;
+}
+
+/* Free thread local buffers if they were allocated only for servicing
+ * this one request */
+static void ost_tls_put(struct ptlrpc_request *r)
+{
+ struct ost_thread_local_cache *tls =
+ (struct ost_thread_local_cache *)(r->rq_svc_thread->t_data);
+
+ if (unlikely(tls->temporary)) {
+ OBD_FREE_PTR(tls);
+ r->rq_svc_thread->t_data = NULL;
+ }
+}
+
static int ost_brw_read(struct ptlrpc_request *req, struct obd_trans_info *oti)
{
struct ptlrpc_bulk_desc *desc = NULL;
struct lustre_handle lockh = { 0 };
int niocount, npages, nob = 0, rc, i;
int no_reply = 0;
+ struct ost_thread_local_cache *tls;
ENTRY;
req->rq_bulk_read = 1;
/* Check if there is eviction in progress, and if so, wait for it to
* finish */
- if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
+ if (unlikely(cfs_atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
lwi = LWI_INTR(NULL, NULL); // We do not care how long it takes
rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
- !atomic_read(&exp->exp_obd->obd_evict_inprogress),
+ !cfs_atomic_read(&exp->exp_obd->obd_evict_inprogress),
&lwi);
}
if (exp->exp_failed)
if (ioo == NULL)
GOTO(out, rc = -EFAULT);
+ rc = ost_validate_obdo(exp, &body->oa, ioo);
+ if (rc)
+ RETURN(rc);
+
niocount = ioo->ioo_bufcnt;
remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
if (remote_nb == NULL)
}
}
- req_capsule_set_size(&req->rq_pill, &RMF_RCS, RCL_SERVER, 0);
rc = req_capsule_server_pack(&req->rq_pill);
if (rc)
GOTO(out, rc);
- /*
- * Per-thread array of struct niobuf_{local,remote}'s was allocated by
- * ost_thread_init().
- */
- local_nb = ost_tls(req)->local;
+ tls = ost_tls_get(req);
+ if (tls == NULL)
+ GOTO(out_bulk, rc = -ENOMEM);
+ local_nb = tls->local;
rc = ost_brw_lock_get(LCK_PR, exp, ioo, remote_nb, &lockh);
if (rc != 0)
- GOTO(out_bulk, rc);
+ GOTO(out_tls, rc);
/*
* If getting the lock took more time than
desc = ptlrpc_prep_bulk_exp(req, npages,
BULK_PUT_SOURCE, OST_BULK_PORTAL);
if (desc == NULL)
- GOTO(out_lock, rc = -ENOMEM);
+ GOTO(out_commitrw, rc = -ENOMEM);
if (!lustre_handle_is_used(&lockh))
/* no needs to try to prolong lock if server is asked
if (rc == 0) {
/* Check if there is eviction in progress, and if so, wait for
* it to finish */
- if (unlikely(atomic_read(&exp->exp_obd->
- obd_evict_inprogress))) {
+ if (unlikely(cfs_atomic_read(&exp->exp_obd->
+ obd_evict_inprogress))) {
lwi = LWI_INTR(NULL, NULL);
rc = l_wait_event(exp->exp_obd->
- obd_evict_inprogress_waitq,
- !atomic_read(&exp->exp_obd->
- obd_evict_inprogress),
+ obd_evict_inprogress_waitq,
+ !cfs_atomic_read(&exp->exp_obd->
+ obd_evict_inprogress),
&lwi);
}
/* Check if client was evicted or tried to reconnect already */
no_reply = rc != 0;
}
+out_commitrw:
/* Must commit after prep above in all cases */
rc = obd_commitrw(OBD_BRW_READ, exp, &body->oa, 1, ioo,
remote_nb, npages, local_nb, oti, rc);
out_lock:
ost_brw_lock_put(LCK_PR, ioo, remote_nb, &lockh);
+out_tls:
+ ost_tls_put(req);
out_bulk:
if (desc)
ptlrpc_free_bulk(desc);
int rc, i, j;
obd_count client_cksum = 0, server_cksum = 0;
cksum_type_t cksum_type = OBD_CKSUM_CRC32;
- int no_reply = 0;
+ int no_reply = 0, mmap = 0;
__u32 o_uid = 0, o_gid = 0;
+ struct ost_thread_local_cache *tls;
ENTRY;
req->rq_bulk_write = 1;
/* Check if there is eviction in progress, and if so, wait for it to
* finish */
- if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
+ if (unlikely(cfs_atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
lwi = LWI_INTR(NULL, NULL); // We do not care how long it takes
rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
- !atomic_read(&exp->exp_obd->obd_evict_inprogress),
+ !cfs_atomic_read(&exp->exp_obd->obd_evict_inprogress),
&lwi);
}
if (exp->exp_failed)
if (body == NULL)
GOTO(out, rc = -EFAULT);
- if ((body->oa.o_flags & OBD_BRW_MEMALLOC) &&
- (exp->exp_connection->c_peer.nid == exp->exp_connection->c_self))
- libcfs_memory_pressure_set();
-
objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
RCL_CLIENT) / sizeof(*ioo);
ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
if (ioo == NULL)
GOTO(out, rc = -EFAULT);
+
+ rc = ost_validate_obdo(exp, &body->oa, ioo);
+ if (rc)
+ RETURN(rc);
+
for (niocount = i = 0; i < objcount; i++)
niocount += ioo[i].ioo_bufcnt;
&RMF_NIOBUF_REMOTE, RCL_CLIENT) / sizeof(*remote_nb)))
GOTO(out, rc = -EFAULT);
+ if ((remote_nb[0].flags & OBD_BRW_MEMALLOC) &&
+ (exp->exp_connection->c_peer.nid == exp->exp_connection->c_self))
+ cfs_memory_pressure_set();
+
if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
if (capa == NULL) {
OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, obd_fail_val);
rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
- /*
- * Per-thread array of struct niobuf_{local,remote}'s was allocated by
- * ost_thread_init().
- */
- local_nb = ost_tls(req)->local;
+ tls = ost_tls_get(req);
+ if (tls == NULL)
+ GOTO(out_bulk, rc = -ENOMEM);
+ local_nb = tls->local;
rc = ost_brw_lock_get(LCK_PW, exp, ioo, remote_nb, &lockh);
if (rc != 0)
- GOTO(out_bulk, rc);
+ GOTO(out_tls, rc);
/*
* If getting the lock took more time than
if (body->oa.o_valid & OBD_MD_FLFLAGS)
cksum_type = cksum_type_unpack(body->oa.o_flags);
}
+ if (body->oa.o_valid & OBD_MD_FLFLAGS && body->oa.o_flags & OBD_FL_MMAP)
+ mmap = 1;
/* Because we already sync grant info with client when reconnect,
* grant info will be cleared for resent req, then fed_grant and
desc = ptlrpc_prep_bulk_exp(req, npages,
BULK_GET_SINK, OST_BULK_PORTAL);
if (desc == NULL)
- GOTO(out_lock, rc = -ENOMEM);
+ GOTO(skip_transfer, rc = -ENOMEM);
/* NB Having prepped, we must commit... */
}
no_reply = rc != 0;
+skip_transfer:
repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
memcpy(&repbody->oa, &body->oa, sizeof(repbody->oa));
repbody->oa.o_cksum = server_cksum;
cksum_counter++;
if (unlikely(client_cksum != server_cksum)) {
- CERROR("client csum %x, server csum %x\n",
- client_cksum, server_cksum);
+ CDEBUG_LIMIT(mmap ? D_INFO : D_ERROR,
+ "client csum %x, server csum %x\n",
+ client_cksum, server_cksum);
cksum_counter = 0;
} else if ((cksum_counter & (-cksum_counter)) == cksum_counter){
CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
repbody->oa.o_gid = o_gid;
}
- if (unlikely(client_cksum != server_cksum && rc == 0)) {
+ /*
+ * Disable sending mtime back to the client. If the client locked the
+ * whole object, then it has already updated the mtime on its side,
+ * otherwise it will have to glimpse anyway (see bug 21489, comment 32)
+ */
+ repbody->oa.o_valid &= ~(OBD_MD_FLMTIME | OBD_MD_FLATIME);
+
+ if (unlikely(client_cksum != server_cksum && rc == 0 && !mmap)) {
int new_cksum = ost_checksum_bulk(desc, OST_WRITE, cksum_type);
char *msg;
char *via;
}
LCONSOLE_ERROR_MSG(0x168, "%s: BAD WRITE CHECKSUM: %s from "
- "%s%s%s inum "LPU64"/"LPU64" object "
+ "%s%s%s inode "DFID" object "
LPU64"/"LPU64" extent ["LPU64"-"LPU64"]\n",
exp->exp_obd->obd_name, msg,
libcfs_id2str(req->rq_peer),
via, router,
body->oa.o_valid & OBD_MD_FLFID ?
- body->oa.o_fid : (__u64)0,
+ body->oa.o_parent_seq : (__u64)0,
body->oa.o_valid & OBD_MD_FLFID ?
- body->oa.o_generation :(__u64)0,
+ body->oa.o_parent_oid : 0,
+ body->oa.o_valid & OBD_MD_FLFID ?
+ body->oa.o_parent_ver : 0,
body->oa.o_id,
body->oa.o_valid & OBD_MD_FLGROUP ?
- body->oa.o_gr : (__u64)0,
+ body->oa.o_seq : (__u64)0,
local_nb[0].offset,
local_nb[npages-1].offset +
local_nb[npages-1].len - 1 );
out_lock:
ost_brw_lock_put(LCK_PW, ioo, remote_nb, &lockh);
+out_tls:
+ ost_tls_put(req);
out_bulk:
if (desc)
ptlrpc_free_bulk(desc);
exp->exp_connection->c_remote_uuid.uuid,
libcfs_id2str(req->rq_peer));
}
- libcfs_memory_pressure_clr();
+ cfs_memory_pressure_clr();
RETURN(rc);
}
}
keylen = req_capsule_get_size(pill, &RMF_SETINFO_KEY, RCL_CLIENT);
+ if (KEY_IS(KEY_FIEMAP)) {
+ struct ll_fiemap_info_key *fm_key = key;
+ int rc;
+
+ rc = ost_validate_obdo(exp, &fm_key->oa, NULL);
+ if (rc)
+ RETURN(rc);
+ }
+
rc = obd_get_info(exp, keylen, key, &replylen, NULL, NULL);
if (rc)
RETURN(rc);
GOTO(out, rc);
repoqa = req_capsule_server_get(&req->rq_pill, &RMF_QUOTA_ADJUST_QUNIT);
- req->rq_status = obd_quota_adjust_qunit(req->rq_export, oqaq, qctxt);
+ req->rq_status = obd_quota_adjust_qunit(req->rq_export, oqaq, qctxt, NULL);
*repoqa = *oqaq;
out:
reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT | \
OBD_CONNECT_RMT_CLIENT_FORCE | \
OBD_CONNECT_OSS_CAPA); \
- spin_lock(&exp->exp_lock); \
+ cfs_spin_lock(&exp->exp_lock); \
exp->exp_connect_flags = reply->ocd_connect_flags; \
- spin_unlock(&exp->exp_lock); \
+ cfs_spin_unlock(&exp->exp_lock); \
} while (0)
static int ost_init_sec_level(struct ptlrpc_request *req)
if (!filter->fo_fl_oss_capa)
reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
- spin_lock(&exp->exp_lock);
+ cfs_spin_lock(&exp->exp_lock);
exp->exp_connect_flags = reply->ocd_connect_flags;
- spin_unlock(&exp->exp_lock);
+ cfs_spin_unlock(&exp->exp_lock);
}
break;
default:
}
if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
- read_lock(&filter->fo_sptlrpc_lock);
+ cfs_read_lock(&filter->fo_sptlrpc_lock);
sptlrpc_target_choose_flavor(&filter->fo_sptlrpc_rset,
req->rq_sp_from,
req->rq_peer.nid,
&flvr);
- read_unlock(&filter->fo_sptlrpc_lock);
+ cfs_read_unlock(&filter->fo_sptlrpc_lock);
- spin_lock(&exp->exp_lock);
+ cfs_spin_lock(&exp->exp_lock);
exp->exp_sp_peer = req->rq_sp_from;
exp->exp_flvr = flvr;
rc = -EACCES;
}
- spin_unlock(&exp->exp_lock);
+ cfs_spin_unlock(&exp->exp_lock);
} else {
if (exp->exp_sp_peer != req->rq_sp_from) {
CERROR("RPC source %s doesn't match %s\n",
return rc;
}
+/* Ensure that data and metadata are synced to the disk when lock is cancelled
+ * (if requested) */
+int ost_blocking_ast(struct ldlm_lock *lock,
+ struct ldlm_lock_desc *desc,
+ void *data, int flag)
+{
+ __u32 sync_lock_cancel = 0;
+ __u32 len = sizeof(sync_lock_cancel);
+ int rc = 0;
+ ENTRY;
+
+ rc = obd_get_info(lock->l_export, sizeof(KEY_SYNC_LOCK_CANCEL),
+ KEY_SYNC_LOCK_CANCEL, &len, &sync_lock_cancel, NULL);
+
+ if (!rc && flag == LDLM_CB_CANCELING &&
+ (lock->l_granted_mode & (LCK_PW|LCK_GROUP)) &&
+ (sync_lock_cancel == ALWAYS_SYNC_ON_CANCEL ||
+ (sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL &&
+ lock->l_flags & LDLM_FL_CBPENDING))) {
+ struct obd_info *oinfo;
+ struct obdo *oa;
+ int rc;
+
+ OBD_ALLOC_PTR(oinfo);
+ if (!oinfo)
+ RETURN(-ENOMEM);
+ OBDO_ALLOC(oa);
+ if (!oa) {
+ OBD_FREE_PTR(oa);
+ RETURN(-ENOMEM);
+ }
+ oa->o_id = lock->l_resource->lr_name.name[0];
+ oa->o_seq = lock->l_resource->lr_name.name[1];
+ oa->o_valid = OBD_MD_FLID|OBD_MD_FLGROUP;
+ oinfo->oi_oa = oa;
+
+ rc = obd_sync(lock->l_export, oinfo,
+ lock->l_policy_data.l_extent.start,
+ lock->l_policy_data.l_extent.end, NULL);
+ if (rc)
+ CERROR("Error %d syncing data on lock cancel\n", rc);
+
+ OBDO_FREE(oa);
+ OBD_FREE_PTR(oinfo);
+ }
+
+ rc = ldlm_server_blocking_ast(lock, desc, data, flag);
+ RETURN(rc);
+}
+
static int ost_filter_recovery_request(struct ptlrpc_request *req,
struct obd_device *obd, int *process)
{
struct obd_ioobj *ioo;
struct ost_body *body;
int objcount, niocount;
- int mode, opc, i;
+ int mode, opc, i, rc;
__u64 start, end;
ENTRY;
if (ioo == NULL)
RETURN(0);
+ rc = ost_validate_obdo(req->rq_export, &body->oa, ioo);
+ if (rc)
+ RETURN(rc);
+
for (niocount = i = 0; i < objcount; i++)
niocount += ioo[i].ioo_bufcnt;
nb[ioo->ioo_bufcnt - 1].len - 1) | ~CFS_PAGE_MASK;
LASSERT(lock->l_resource != NULL);
- if (!osc_res_name_eq(ioo->ioo_id, ioo->ioo_gr,
+ if (!osc_res_name_eq(ioo->ioo_id, ioo->ioo_seq,
&lock->l_resource->lr_name))
RETURN(0);
struct obd_ioobj *ioo;
struct ost_body *body;
int objcount, niocount;
- int mode, opc, i;
+ int mode, opc, i, rc;
ENTRY;
opc = lustre_msg_get_opc(req->rq_reqmsg);
if (ioo == NULL)
RETURN(-EFAULT);
+ rc = ost_validate_obdo(req->rq_export, &body->oa, ioo);
+ if (rc)
+ RETURN(rc);
+
for (niocount = i = 0; i < objcount; i++)
niocount += ioo[i].ioo_bufcnt;
nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
struct ldlm_lock *lock)
{
struct ost_body *body;
+ int rc;
ENTRY;
body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
if (body == NULL)
RETURN(0); /* can't return -EFAULT here */
+ rc = ost_validate_obdo(req->rq_export, &body->oa, NULL);
+ if (rc)
+ RETURN(rc);
+
if (body->oa.o_valid & OBD_MD_FLHANDLE &&
body->oa.o_handle.cookie == lock->l_handle.h_cookie)
RETURN(1);
static int ost_punch_hpreq_check(struct ptlrpc_request *req)
{
struct ost_body *body;
+ int rc;
body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
if (body == NULL)
RETURN(-EFAULT);
+ rc = ost_validate_obdo(req->rq_export, &body->oa, NULL);
+ if (rc)
+ RETURN(rc);
+
LASSERT(!(body->oa.o_valid & OBD_MD_FLFLAGS) ||
- !(body->oa.o_flags & OBD_FL_TRUNCLOCK));
+ !(body->oa.o_flags & OBD_FL_SRVLOCK));
RETURN(ost_punch_prolong_locks(req, &body->oa));
}
* it doesn't change).
*/
req_capsule_init(&req->rq_pill, req, RCL_SERVER);
- req_capsule_set(&req->rq_pill, &RQF_OST_BRW);
+ if (opc == OST_READ)
+ req_capsule_set(&req->rq_pill,
+ &RQF_OST_BRW_READ);
+ else
+ req_capsule_set(&req->rq_pill,
+ &RQF_OST_BRW_WRITE);
body = req_capsule_client_get(&req->rq_pill,
&RMF_OST_BODY);
CERROR("Missing/short ost_body\n");
RETURN(-EFAULT);
}
+
objcount = req_capsule_get_size(&req->rq_pill,
&RMF_OBD_IOOBJ,
RCL_CLIENT) /
}
if (!(body->oa.o_valid & OBD_MD_FLFLAGS) ||
- !(body->oa.o_flags & OBD_FL_TRUNCLOCK))
+ !(body->oa.o_flags & OBD_FL_SRVLOCK))
req->rq_ops = &ost_hpreq_punch;
}
}
req_capsule_init(&req->rq_pill, req, RCL_SERVER);
- /* XXX identical to MDS */
if (lustre_msg_get_opc(req->rq_reqmsg) != OST_CONNECT) {
- int recovering;
-
- if (req->rq_export == NULL) {
+ if (!class_connected_export(req->rq_export)) {
CDEBUG(D_HA,"operation %d on unconnected OST from %s\n",
lustre_msg_get_opc(req->rq_reqmsg),
libcfs_id2str(req->rq_peer));
obd = req->rq_export->exp_obd;
/* Check for aborted recovery. */
- spin_lock_bh(&obd->obd_processing_task_lock);
- recovering = obd->obd_recovering;
- spin_unlock_bh(&obd->obd_processing_task_lock);
- if (recovering) {
+ if (obd->obd_recovering) {
rc = ost_filter_recovery_request(req, obd,
&should_process);
if (rc || !should_process)
req_capsule_set(&req->rq_pill, &RQF_OST_CREATE);
if (OBD_FAIL_CHECK(OBD_FAIL_OST_CREATE_NET))
RETURN(0);
- if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
- GOTO(out, rc = -ENOSPC);
if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
GOTO(out, rc = -EROFS);
rc = ost_create(req->rq_export, req, oti);
rc = ost_setattr(req->rq_export, req, oti);
break;
case OST_WRITE:
- req_capsule_set(&req->rq_pill, &RQF_OST_BRW);
+ req_capsule_set(&req->rq_pill, &RQF_OST_BRW_WRITE);
CDEBUG(D_INODE, "write\n");
/* req->rq_request_portal would be nice, if it was set */
if (req->rq_rqbd->rqbd_service->srv_req_portal !=OST_IO_PORTAL){
/* ost_brw_write sends its own replies */
RETURN(rc);
case OST_READ:
- req_capsule_set(&req->rq_pill, &RQF_OST_BRW);
+ req_capsule_set(&req->rq_pill, &RQF_OST_BRW_READ);
CDEBUG(D_INODE, "read\n");
/* req->rq_request_portal would be nice, if it was set */
if (req->rq_rqbd->rqbd_service->srv_req_portal !=OST_IO_PORTAL){
if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_ENQUEUE))
RETURN(0);
rc = ldlm_handle_enqueue(req, ldlm_server_completion_ast,
- ldlm_server_blocking_ast,
+ ost_blocking_ast,
ldlm_server_glimpse_ast);
fail = OBD_FAIL_OST_LDLM_REPLY_NET;
break;
int rc;
ENTRY;
- rc = cleanup_group_info();
+ rc = cfs_cleanup_group_info();
if (rc)
RETURN(rc);
lprocfs_ost_init_vars(&lvars);
lprocfs_obd_setup(obd, lvars.obd_vars);
- sema_init(&ost->ost_health_sem, 1);
+ cfs_sema_init(&ost->ost_health_sem, 1);
if (oss_num_threads) {
/* If oss_num_threads is set, it is the min and the max. */
oss_max_threads = oss_min_threads = oss_num_threads;
} else {
/* Base min threads on memory and cpus */
- oss_min_threads = num_possible_cpus() * CFS_NUM_CACHEPAGES >>
+ oss_min_threads =
+ cfs_num_possible_cpus() * CFS_NUM_CACHEPAGES >>
(27 - CFS_PAGE_SHIFT);
if (oss_min_threads < OSS_THREADS_MIN)
oss_min_threads = OSS_THREADS_MIN;
GOTO(out_lprocfs, rc = -ENOMEM);
}
- rc = ptlrpc_start_threads(obd, ost->ost_service);
+ rc = ptlrpc_start_threads(ost->ost_service);
if (rc)
GOTO(out_service, rc = -EINVAL);
GOTO(out_service, rc = -ENOMEM);
}
- rc = ptlrpc_start_threads(obd, ost->ost_create_service);
+ rc = ptlrpc_start_threads(ost->ost_create_service);
if (rc)
GOTO(out_create, rc = -EINVAL);
ost->ost_io_service->srv_init = ost_thread_init;
ost->ost_io_service->srv_done = ost_thread_done;
ost->ost_io_service->srv_cpu_affinity = 1;
- rc = ptlrpc_start_threads(obd, ost->ost_io_service);
+ rc = ptlrpc_start_threads(ost->ost_io_service);
if (rc)
GOTO(out_io, rc = -EINVAL);
ping_evictor_stop();
- spin_lock_bh(&obd->obd_processing_task_lock);
- if (obd->obd_recovering) {
- target_cancel_recovery_timer(obd);
- obd->obd_recovering = 0;
- }
- spin_unlock_bh(&obd->obd_processing_task_lock);
-
- down(&ost->ost_health_sem);
+ /* there is no recovery for OST OBD, all recovery is controlled by
+ * obdfilter OBD */
+ LASSERT(obd->obd_recovering == 0);
+ cfs_down(&ost->ost_health_sem);
ptlrpc_unregister_service(ost->ost_service);
ptlrpc_unregister_service(ost->ost_create_service);
ptlrpc_unregister_service(ost->ost_io_service);
ost->ost_service = NULL;
ost->ost_create_service = NULL;
- up(&ost->ost_health_sem);
+ cfs_up(&ost->ost_health_sem);
lprocfs_obd_cleanup(obd);
struct ost_obd *ost = &obd->u.ost;
int rc = 0;
- down(&ost->ost_health_sem);
+ cfs_down(&ost->ost_health_sem);
rc |= ptlrpc_service_health_check(ost->ost_service);
rc |= ptlrpc_service_health_check(ost->ost_create_service);
rc |= ptlrpc_service_health_check(ost->ost_io_service);
- up(&ost->ost_health_sem);
+ cfs_up(&ost->ost_health_sem);
/*
* health_check to return 0 on healthy
git://git.whamcloud.com / fs / lustre-release.git / blobdiff