/* lu2dt_dev() */
#include <dt_object.h>
#include <lustre_mds.h>
-#include <lustre_mdt.h>
#include <lustre_log.h>
#include "mdt_internal.h"
#include <lustre_acl.h>
#include <lustre_param.h>
#include <lustre_quota.h>
#include <lustre_linkea.h>
+#include <lustre_lfsck.h>
mdl_mode_t mdt_mdl_lock_modes[] = {
[LCK_MINMODE] = MDL_MINMODE,
static const struct lu_object_operations mdt_obj_ops;
/* Slab for MDT object allocation */
-static cfs_mem_cache_t *mdt_object_kmem;
+static struct kmem_cache *mdt_object_kmem;
+
+/* For HSM restore handles */
+struct kmem_cache *mdt_hsm_cdt_kmem;
+
+/* For HSM request handles */
+struct kmem_cache *mdt_hsm_car_kmem;
static struct lu_kmem_descr mdt_caches[] = {
{
.ckd_size = sizeof(struct mdt_object)
},
{
+ .ckd_cache = &mdt_hsm_cdt_kmem,
+ .ckd_name = "mdt_cdt_restore_handle",
+ .ckd_size = sizeof(struct cdt_restore_handle)
+ },
+ {
+ .ckd_cache = &mdt_hsm_car_kmem,
+ .ckd_name = "mdt_cdt_agent_req",
+ .ckd_size = sizeof(struct cdt_agent_req)
+ },
+ {
.ckd_cache = NULL
}
};
b->blocks = 0;
/* if no object is allocated on osts, the size on mds is valid. b=22272 */
b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
- }
+ } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm &&
+ (ma->ma_lmm->lmm_pattern & LOV_PATTERN_F_RELEASED)) {
+ /* A released file stores its size on MDS. */
+ b->blocks = 0;
+ b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
+ }
if (fid) {
b->fid1 = *fid;
RETURN(rc);
}
+ /* if file is released, check if a restore is running */
+ if ((ma->ma_valid & MA_HSM) && (ma->ma_hsm.mh_flags & HS_RELEASED) &&
+ mdt_hsm_restore_is_running(info, mdt_object_fid(o))) {
+ repbody->t_state = MS_RESTORE;
+ repbody->valid |= OBD_MD_TSTATE;
+ }
+
is_root = lu_fid_eq(mdt_object_fid(o), &info->mti_mdt->mdt_md_root_fid);
/* the Lustre protocol supposes to return default striping
PFID(mdt_object_fid(o)), rc);
rc = -EFAULT;
} else {
- int print_limit = min_t(int, CFS_PAGE_SIZE - 128, rc);
+ int print_limit = min_t(int, PAGE_CACHE_SIZE - 128, rc);
if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
rc -= 2;
}
LASSERT(obj != NULL);
- LASSERT(lu_object_assert_exists(&obj->mot_obj.mo_lu));
+ LASSERT(lu_object_assert_exists(&obj->mot_obj));
- mode = lu_object_attr(&obj->mot_obj.mo_lu);
+ mode = lu_object_attr(&obj->mot_obj);
/* old clients may not report needed easize, use max value then */
req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
if (unlikely(!mdt_object_exists(parent))) {
LU_OBJECT_DEBUG(D_INODE, info->mti_env,
- &parent->mot_obj.mo_lu,
+ &parent->mot_obj,
"Parent doesn't exist!\n");
RETURN(-ESTALE);
} else if (!info->mti_cross_ref) {
if (lname) {
/* step 1: lock parent only if parent is a directory */
- if (S_ISDIR(lu_object_attr(&parent->mot_obj.mo_lu))) {
+ if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
lhp = &info->mti_lh[MDT_LH_PARENT];
mdt_lock_pdo_init(lhp, LCK_PR, name, namelen);
rc = mdt_object_lock(info, parent, lhp,
*/
child = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
- if (unlikely(IS_ERR(child)))
- GOTO(out_parent, rc = PTR_ERR(child));
- if (is_resent) {
- /* Do not take lock for resent case. */
- lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
- LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
- lhc->mlh_reg_lh.cookie);
-
- res_id = &lock->l_resource->lr_name;
- if (!fid_res_name_eq(mdt_object_fid(child),
- &lock->l_resource->lr_name)) {
- LASSERTF(fid_res_name_eq(mdt_object_fid(parent),
- &lock->l_resource->lr_name),
- "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
- (unsigned long)res_id->name[0],
- (unsigned long)res_id->name[1],
- (unsigned long)res_id->name[2],
- PFID(mdt_object_fid(parent)));
- CWARN("Although resent, but still not get child lock"
- "parent:"DFID" child:"DFID"\n",
- PFID(mdt_object_fid(parent)),
- PFID(mdt_object_fid(child)));
- lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
- LDLM_LOCK_PUT(lock);
- GOTO(relock, 0);
- }
- LDLM_LOCK_PUT(lock);
- rc = 0;
- } else {
+ if (unlikely(IS_ERR(child)))
+ GOTO(out_parent, rc = PTR_ERR(child));
+ if (is_resent) {
+ /* Do not take lock for resent case. */
+ lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
+ LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
+ lhc->mlh_reg_lh.cookie);
+
+ res_id = &lock->l_resource->lr_name;
+ if (!fid_res_name_eq(mdt_object_fid(child),
+ &lock->l_resource->lr_name)) {
+ LASSERTF(fid_res_name_eq(mdt_object_fid(parent),
+ &lock->l_resource->lr_name),
+ "Lock res_id: "DLDLMRES", fid: "DFID"\n",
+ PLDLMRES(lock->l_resource),
+ PFID(mdt_object_fid(parent)));
+ CWARN("Although resent, but still not get child lock"
+ "parent:"DFID" child:"DFID"\n",
+ PFID(mdt_object_fid(parent)),
+ PFID(mdt_object_fid(child)));
+ lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
+ LDLM_LOCK_PUT(lock);
+ GOTO(relock, 0);
+ }
+ LDLM_LOCK_PUT(lock);
+ rc = 0;
+ } else {
bool try_layout = false;
relock:
if (!mdt_object_exists(child)) {
LU_OBJECT_DEBUG(D_INODE, info->mti_env,
- &child->mot_obj.mo_lu,
+ &child->mot_obj,
"Object doesn't exist!\n");
GOTO(out_child, rc = -ENOENT);
}
LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
exp_connect_layout(info->mti_exp) &&
- S_ISREG(lu_object_attr(&child->mot_obj.mo_lu)) &&
+ S_ISREG(lu_object_attr(&child->mot_obj)) &&
ldlm_rep != NULL) {
/* try to grant layout lock for regular file. */
try_layout = true;
rc = mdt_getattr_internal(info, child, ma_need);
if (unlikely(rc != 0)) {
mdt_object_unlock(info, child, lhc, 1);
- } else if (lock) {
- /* Debugging code. */
- res_id = &lock->l_resource->lr_name;
- LDLM_DEBUG(lock, "Returning lock to client");
- LASSERTF(fid_res_name_eq(mdt_object_fid(child),
- &lock->l_resource->lr_name),
- "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
- (unsigned long)res_id->name[0],
- (unsigned long)res_id->name[1],
- (unsigned long)res_id->name[2],
- PFID(mdt_object_fid(child)));
+ } else if (lock) {
+ /* Debugging code. */
+ res_id = &lock->l_resource->lr_name;
+ LDLM_DEBUG(lock, "Returning lock to client");
+ LASSERTF(fid_res_name_eq(mdt_object_fid(child),
+ &lock->l_resource->lr_name),
+ "Lock res_id: "DLDLMRES", fid: "DFID"\n",
+ PLDLMRES(lock->l_resource),
+ PFID(mdt_object_fid(child)));
if (mdt_object_exists(child) && !mdt_object_remote(child))
mdt_pack_size2body(info, child);
}
RETURN(0);
}
+int mdt_connect_check_sptlrpc(struct mdt_device *mdt, struct obd_export *exp,
+ struct ptlrpc_request *req);
+
/**
* Top-level handler for MDT connection requests.
*/
struct obd_export *exp;
struct ptlrpc_request *req = mdt_info_req(info);
+ ENTRY;
+
rc = target_handle_connect(req);
if (rc != 0)
- return err_serious(rc);
+ RETURN(err_serious(rc));
LASSERT(req->rq_export != NULL);
- info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
+ exp = req->rq_export;
+ info->mti_exp = exp;
+ info->mti_mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
rc = mdt_init_sec_level(info);
- if (rc != 0) {
- obd_disconnect(class_export_get(req->rq_export));
- return rc;
- }
+ if (rc != 0)
+ GOTO(err, rc);
+
+ rc = mdt_connect_check_sptlrpc(info->mti_mdt, exp, req);
+ if (rc)
+ GOTO(err, rc);
/* To avoid exposing partially initialized connection flags, changes up
* to this point have been staged in reply->ocd_connect_flags. Now that
* connection handling has completed successfully, atomically update
* the connect flags in the shared export data structure. LU-1623 */
reply = req_capsule_server_get(info->mti_pill, &RMF_CONNECT_DATA);
- exp = req->rq_export;
spin_lock(&exp->exp_lock);
*exp_connect_flags_ptr(exp) = reply->ocd_connect_flags;
spin_unlock(&exp->exp_lock);
rc = mdt_init_idmap(info);
if (rc != 0)
- obd_disconnect(class_export_get(req->rq_export));
-
+ GOTO(err, rc);
+ RETURN(0);
+err:
+ obd_disconnect(class_export_get(req->rq_export));
return rc;
}
for (i = 0, tmpcount = nob; i < rdpg->rp_npages && tmpcount > 0;
i++, tmpcount -= tmpsize) {
- tmpsize = min_t(int, tmpcount, CFS_PAGE_SIZE);
+ tmpsize = min_t(int, tmpcount, PAGE_CACHE_SIZE);
ptlrpc_prep_bulk_page_pin(desc, rdpg->rp_pages[i], 0, tmpsize);
}
rdpg->rp_attrs |= LUDA_64BITHASH;
rdpg->rp_count = min_t(unsigned int, reqbody->nlink,
exp_max_brw_size(info->mti_exp));
- rdpg->rp_npages = (rdpg->rp_count + CFS_PAGE_SIZE - 1) >>
- CFS_PAGE_SHIFT;
+ rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE - 1) >>
+ PAGE_CACHE_SHIFT;
OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
if (rdpg->rp_pages == NULL)
RETURN(-ENOMEM);
for (i = 0; i < rdpg->rp_npages; ++i) {
- rdpg->rp_pages[i] = cfs_alloc_page(CFS_ALLOC_STD);
+ rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
if (rdpg->rp_pages[i] == NULL)
GOTO(free_rdpg, rc = -ENOMEM);
}
for (i = 0; i < rdpg->rp_npages; i++)
if (rdpg->rp_pages[i] != NULL)
- cfs_free_page(rdpg->rp_pages[i]);
+ __free_page(rdpg->rp_pages[i]);
OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
GOTO(out, rc = -EFAULT);
rdpg->rp_count = min_t(unsigned int, req_ii->ii_count << LU_PAGE_SHIFT,
exp_max_brw_size(info->mti_exp));
- rdpg->rp_npages = (rdpg->rp_count + CFS_PAGE_SIZE -1) >> CFS_PAGE_SHIFT;
+ rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE - 1) >>
+ PAGE_CACHE_SHIFT;
/* allocate pages to store the containers */
OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
if (rdpg->rp_pages == NULL)
GOTO(out, rc = -ENOMEM);
for (i = 0; i < rdpg->rp_npages; i++) {
- rdpg->rp_pages[i] = cfs_alloc_page(CFS_ALLOC_STD);
+ rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
if (rdpg->rp_pages[i] == NULL)
GOTO(out, rc = -ENOMEM);
}
if (rdpg->rp_pages) {
for (i = 0; i < rdpg->rp_npages; i++)
if (rdpg->rp_pages[i])
- cfs_free_page(rdpg->rp_pages[i]);
+ __free_page(rdpg->rp_pages[i]);
OBD_FREE(rdpg->rp_pages,
rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
}
static struct mdt_object *mdt_obj(struct lu_object *o)
{
- LASSERT(lu_device_is_mdt(o->lo_dev));
- return container_of0(o, struct mdt_object, mot_obj.mo_lu);
+ LASSERT(lu_device_is_mdt(o->lo_dev));
+ return container_of0(o, struct mdt_object, mot_obj);
}
struct mdt_object *mdt_object_new(const struct lu_env *env,
ENTRY;
CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
- o = lu_object_find(env, &d->mdt_md_dev.md_lu_dev, f, &conf);
+ o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
if (unlikely(IS_ERR(o)))
m = (struct mdt_object *)o;
else
}
struct mdt_object *mdt_object_find(const struct lu_env *env,
- struct mdt_device *d,
- const struct lu_fid *f)
+ struct mdt_device *d,
+ const struct lu_fid *f)
{
- struct lu_object *o;
- struct mdt_object *m;
- ENTRY;
+ struct lu_object *o;
+ struct mdt_object *m;
+ ENTRY;
- CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
- o = lu_object_find(env, &d->mdt_md_dev.md_lu_dev, f, NULL);
- if (unlikely(IS_ERR(o)))
- m = (struct mdt_object *)o;
- else
- m = mdt_obj(o);
- RETURN(m);
+ CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
+ o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
+ if (unlikely(IS_ERR(o)))
+ m = (struct mdt_object *)o;
+ else
+ m = mdt_obj(o);
+
+ RETURN(m);
}
/**
RETURN(-ESTALE);
} else {
/* Non-dir object shouldn't have PDO lock */
- if (!S_ISDIR(lu_object_attr(&o->mot_obj.mo_lu)))
- RETURN(-ENOTDIR);
+ if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
+ RETURN(-ENOTDIR);
}
}
* uninitialized state, because it's too expensive to zero out whole
* mdt_thread_info (> 1K) on each request arrival.
*/
-static void mdt_thread_info_init(struct ptlrpc_request *req,
- struct mdt_thread_info *info)
+void mdt_thread_info_init(struct ptlrpc_request *req,
+ struct mdt_thread_info *info)
{
int i;
- req_capsule_init(&req->rq_pill, req, RCL_SERVER);
info->mti_pill = &req->rq_pill;
/* lock handle */
info->mti_spec.sp_rm_entry = 0;
}
-static void mdt_thread_info_fini(struct mdt_thread_info *info)
+void mdt_thread_info_fini(struct mdt_thread_info *info)
{
int i;
- req_capsule_fini(info->mti_pill);
if (info->mti_object != NULL) {
mdt_object_put(info->mti_env, info->mti_object);
info->mti_object = NULL;
for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
mdt_lock_handle_fini(&info->mti_lh[i]);
info->mti_env = NULL;
+ info->mti_pill = NULL;
+ info->mti_exp = NULL;
if (unlikely(info->mti_big_buf.lb_buf != NULL))
lu_buf_free(&info->mti_big_buf);
}
+int mdt_tgt_connect(struct tgt_session_info *tsi)
+{
+ struct ptlrpc_request *req = tgt_ses_req(tsi);
+ struct mdt_thread_info *mti;
+ int rc;
+
+ ENTRY;
+
+ rc = tgt_connect(tsi);
+ if (rc != 0)
+ RETURN(rc);
+
+ /* XXX: switch mdt_init_idmap() to use tgt_session_info */
+ lu_env_refill((void *)tsi->tsi_env);
+ mti = lu_context_key_get(&tsi->tsi_env->le_ctx, &mdt_thread_key);
+ LASSERT(mti != NULL);
+
+ mdt_thread_info_init(req, mti);
+ rc = mdt_init_idmap(mti);
+ mdt_thread_info_fini(mti);
+ if (rc != 0)
+ GOTO(err, rc);
+ RETURN(0);
+err:
+ obd_disconnect(class_export_get(req->rq_export));
+ return rc;
+}
+
static int mdt_filter_recovery_request(struct ptlrpc_request *req,
struct obd_device *obd, int *process)
{
if (likely(h != NULL)) {
rc = mdt_req_handle(info, h, req);
} else {
- CERROR("The unsupported opc: 0x%x\n",
- lustre_msg_get_opc(msg) );
+ CERROR("%s: opc unsupported: 0x%x\n",
+ mdt_obd_name(info->mti_mdt),
+ lustre_msg_get_opc(msg));
req->rq_status = -ENOTSUPP;
rc = ptlrpc_error(req);
RETURN(rc);
}
}
- } else
- CERROR(LUSTRE_MDT_NAME" drops mal-formed request\n");
- RETURN(rc);
+ } else {
+ CDEBUG(D_INFO, "%s: drops mal-formed request: rc = %d\n",
+ mdt_obd_name(info->mti_mdt), rc);
+ req->rq_status = rc;
+ rc = ptlrpc_error(req);
+ }
+ RETURN(rc);
}
/*
info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
LASSERT(info != NULL);
+ req_capsule_init(&req->rq_pill, req, RCL_SERVER);
mdt_thread_info_init(req, info);
rc = mdt_handle0(req, info, supported);
mdt_thread_info_fini(info);
+ req_capsule_fini(&req->rq_pill);
RETURN(rc);
}
*/
int mdt_recovery_handle(struct ptlrpc_request *req)
{
- int rc;
- ENTRY;
+ int rc;
- switch (lustre_msg_get_opc(req->rq_reqmsg)) {
- case FLD_QUERY:
- rc = mdt_handle_common(req, mdt_fld_handlers);
- break;
- case SEQ_QUERY:
- rc = mdt_handle_common(req, mdt_seq_handlers);
- break;
- default:
- rc = mdt_handle_common(req, mdt_regular_handlers);
- break;
- }
+ ENTRY;
- RETURN(rc);
+ rc = mdt_handle_common(req, mdt_regular_handlers);
+
+ RETURN(rc);
}
enum mdt_it_code {
struct mdt_thread_info *info,
struct ldlm_lock **,
__u64);
+
+static int mdt_intent_getxattr(enum mdt_it_code opcode,
+ struct mdt_thread_info *info,
+ struct ldlm_lock **lockp,
+ __u64 flags);
+
static int mdt_intent_layout(enum mdt_it_code opcode,
struct mdt_thread_info *info,
struct ldlm_lock **,
.it_act = NULL
},
[MDT_IT_GETXATTR] = {
- .it_fmt = NULL,
+ .it_fmt = &RQF_LDLM_INTENT_GETXATTR,
.it_flags = 0,
- .it_act = NULL
+ .it_act = mdt_intent_getxattr
},
[MDT_IT_LAYOUT] = {
.it_fmt = &RQF_LDLM_INTENT_LAYOUT,
remote_hdl.cookie);
}
+static int mdt_intent_getxattr(enum mdt_it_code opcode,
+ struct mdt_thread_info *info,
+ struct ldlm_lock **lockp,
+ __u64 flags)
+{
+ struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
+ struct ldlm_reply *ldlm_rep = NULL;
+ int rc, grc;
+
+ /*
+ * Initialize lhc->mlh_reg_lh either from a previously granted lock
+ * (for the resend case) or a new lock. Below we will use it to
+ * replace the original lock.
+ */
+ mdt_intent_fixup_resent(info, *lockp, NULL, lhc);
+ if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
+ mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
+ rc = mdt_object_lock(info, info->mti_object, lhc,
+ MDS_INODELOCK_XATTR,
+ MDT_LOCAL_LOCK);
+ if (rc)
+ return rc;
+ }
+
+ grc = mdt_getxattr(info);
+
+ rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
+
+ if (mdt_info_req(info)->rq_repmsg != NULL)
+ ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
+ if (ldlm_rep == NULL)
+ RETURN(err_serious(-EFAULT));
+
+ ldlm_rep->lock_policy_res2 = grc;
+
+ return rc;
+}
+
static int mdt_intent_getattr(enum mdt_it_code opcode,
struct mdt_thread_info *info,
struct ldlm_lock **lockp,
/* execute policy */
rc = flv->it_act(opc, info, lockp, flags);
- rep = req_capsule_server_get(pill, &RMF_DLM_REP);
- rep->lock_policy_res2 =
- ptlrpc_status_hton(rep->lock_policy_res2);
+ /* Check whether the reply has been packed successfully. */
+ if (mdt_info_req(info)->rq_repmsg != NULL) {
+ rep = req_capsule_server_get(info->mti_pill,
+ &RMF_DLM_REP);
+ rep->lock_policy_res2 =
+ ptlrpc_status_hton(rep->lock_policy_res2);
+ }
} else {
rc = -EOPNOTSUPP;
}
site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
LASSERT(site);
LASSERT(mdt_lu_site(mdt) == NULL);
- mdt->mdt_md_dev.md_lu_dev.ld_site = site;
- site->ls_top_dev = &mdt->mdt_md_dev.md_lu_dev;
+ mdt->mdt_lu_dev.ld_site = site;
+ site->ls_top_dev = &mdt->mdt_lu_dev;
mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
if (rc != 0)
CERROR("Failure to refill session: '%d'\n", rc);
- lu_dev_add_linkage(site, &mdt->mdt_md_dev.md_lu_dev);
+ lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
EXIT;
class_detach:
/* configure local quota objects */
rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
- &mdt->mdt_md_dev.md_lu_dev,
+ &mdt->mdt_lu_dev,
mdt->mdt_qmt_dev);
if (rc)
GOTO(class_cleanup, rc);
EXIT;
}
+static struct tgt_handler mdt_tgt_handlers[] = {
+TGT_RPC_HANDLER(MDS_FIRST_OPC,
+ 0, MDS_CONNECT, mdt_tgt_connect,
+ &RQF_CONNECT, LUSTRE_OBD_VERSION),
+TGT_RPC_HANDLER(MDS_FIRST_OPC,
+ 0, MDS_DISCONNECT, tgt_disconnect,
+ &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
+};
+
+static struct tgt_opc_slice mdt_common_slice[] = {
+ {
+ .tos_opc_start = MDS_FIRST_OPC,
+ .tos_opc_end = MDS_LAST_OPC,
+ .tos_hs = mdt_tgt_handlers
+ },
+ {
+ .tos_opc_start = OBD_FIRST_OPC,
+ .tos_opc_end = OBD_LAST_OPC,
+ .tos_hs = tgt_obd_handlers
+ },
+ {
+ .tos_opc_start = LDLM_FIRST_OPC,
+ .tos_opc_end = LDLM_LAST_OPC,
+ .tos_hs = tgt_dlm_handlers
+ },
+ {
+ .tos_opc_start = SEC_FIRST_OPC,
+ .tos_opc_end = SEC_LAST_OPC,
+ .tos_hs = tgt_sec_ctx_handlers
+ },
+ {
+ .tos_opc_start = UPDATE_OBJ,
+ .tos_opc_end = UPDATE_LAST_OPC,
+ .tos_hs = tgt_out_handlers
+ },
+ {
+ .tos_opc_start = FLD_FIRST_OPC,
+ .tos_opc_end = FLD_LAST_OPC,
+ .tos_hs = fld_handlers
+ },
+ {
+ .tos_opc_start = SEQ_FIRST_OPC,
+ .tos_opc_end = SEQ_LAST_OPC,
+ .tos_hs = seq_handlers
+ },
+ {
+ .tos_hs = NULL
+ }
+};
+
static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
{
- struct md_device *next = m->mdt_child;
- struct lu_device *d = &m->mdt_md_dev.md_lu_dev;
+ struct md_device *next = m->mdt_child;
+ struct lu_device *d = &m->mdt_lu_dev;
struct obd_device *obd = mdt2obd_dev(m);
ENTRY;
ping_evictor_stop();
-
mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
+
+ if (m->mdt_opts.mo_coordinator)
+ mdt_hsm_cdt_stop(m);
+
+ mdt_hsm_cdt_fini(m);
+
+ mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
obd_exports_barrier(obd);
obd_zombie_barrier();
m->mdt_nosquash_strlen = 0;
}
- next->md_ops->mdo_iocontrol(env, next, OBD_IOC_PAUSE_LFSCK,
- 0, NULL);
+ next->md_ops->mdo_iocontrol(env, next, OBD_IOC_PAUSE_LFSCK, 0, NULL);
+
mdt_seq_fini(env, m);
mdt_fld_fini(env, m);
sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
mntopt_t mntopts;
ENTRY;
- md_device_init(&m->mdt_md_dev, ldt);
+ lu_device_init(&m->mdt_lu_dev, ldt);
/*
* Environment (env) might be missing mdt_thread_key values at that
* point, if device is allocated when mdt_thread_key is in QUIESCENT
m->mdt_som_conf = 0;
m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
+
+ /* default is coordinator off, it is started through conf_param
+ * or /proc */
+ m->mdt_opts.mo_coordinator = 0;
+
lmi = server_get_mount(dev);
if (lmi == NULL) {
CERROR("Cannot get mount info for %s!\n", dev);
}
rwlock_init(&m->mdt_sptlrpc_lock);
- sptlrpc_rule_set_init(&m->mdt_sptlrpc_rset);
+ sptlrpc_rule_set_init(&m->mdt_sptlrpc_rset);
spin_lock_init(&m->mdt_ioepoch_lock);
m->mdt_opts.mo_compat_resname = 0;
m->mdt_enable_remote_dir = 0;
m->mdt_enable_remote_dir_gid = 0;
- m->mdt_md_dev.md_lu_dev.ld_ops = &mdt_lu_ops;
- m->mdt_md_dev.md_lu_dev.ld_obd = obd;
- /* set this lu_device to obd, because error handling need it */
- obd->obd_lu_dev = &m->mdt_md_dev.md_lu_dev;
+ m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
+ m->mdt_lu_dev.ld_obd = obd;
+ /* Set this lu_device to obd for error handling purposes. */
+ obd->obd_lu_dev = &m->mdt_lu_dev;
/* init the stack */
rc = mdt_stack_init((struct lu_env *)env, m, cfg);
}
}
- rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom);
- if (rc)
- GOTO(err_fini_stack, rc);
-
rc = mdt_fld_init(env, mdt_obd_name(m), m);
if (rc)
- GOTO(err_lut, rc);
+ GOTO(err_fini_stack, rc);
rc = mdt_seq_init(env, mdt_obd_name(m), m);
if (rc)
cfs_timer_init(&m->mdt_ck_timer, mdt_ck_timer_callback, m);
- rc = mdt_ck_thread_start(m);
- if (rc)
+ rc = mdt_hsm_cdt_init(m);
+ if (rc != 0) {
+ CERROR("%s: error initializing coordinator, rc %d\n",
+ mdt_obd_name(m), rc);
GOTO(err_free_ns, rc);
+ }
- rc = mdt_fs_setup(env, m, obd, lsi);
+ rc = mdt_ck_thread_start(m);
if (rc)
- GOTO(err_capa, rc);
+ GOTO(err_free_hsm, rc);
+
+ rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
+ OBD_FAIL_MDS_ALL_REQUEST_NET,
+ OBD_FAIL_MDS_ALL_REPLY_NET);
+ if (rc)
+ GOTO(err_capa, rc);
+
+ rc = mdt_fs_setup(env, m, obd, lsi);
+ if (rc)
+ GOTO(err_tgt, rc);
mdt_adapt_sptlrpc_conf(obd, 1);
rc = next->md_ops->mdo_iocontrol(env, next, OBD_IOC_GET_MNTOPT, 0,
&mntopts);
if (rc)
- GOTO(err_llog_cleanup, rc);
+ GOTO(err_fs_cleanup, rc);
if (mntopts & MNTOPT_USERXATTR)
m->mdt_opts.mo_user_xattr = 1;
else
m->mdt_opts.mo_user_xattr = 0;
+ rc = next->md_ops->mdo_maxeasize_get(env, next, &m->mdt_max_ea_size);
+ if (rc)
+ GOTO(err_fs_cleanup, rc);
+
if (mntopts & MNTOPT_ACL)
m->mdt_opts.mo_acl = 1;
else
if (IS_ERR(m->mdt_identity_cache)) {
rc = PTR_ERR(m->mdt_identity_cache);
m->mdt_identity_cache = NULL;
- GOTO(err_llog_cleanup, rc);
+ GOTO(err_fs_cleanup, rc);
}
rc = mdt_procfs_init(m, dev);
ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
"mdt_ldlm_client", m->mdt_ldlm_client);
- ping_evictor_start();
+ ping_evictor_start();
/* recovery will be started upon mdt_prepare()
* when the whole stack is complete and ready
ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
RETURN(0);
-
err_procfs:
- mdt_procfs_fini(m);
+ mdt_procfs_fini(m);
err_recovery:
- target_recovery_fini(obd);
- upcall_cache_cleanup(m->mdt_identity_cache);
- m->mdt_identity_cache = NULL;
-err_llog_cleanup:
- mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
- mdt_fs_cleanup(env, m);
+ target_recovery_fini(obd);
+ upcall_cache_cleanup(m->mdt_identity_cache);
+ m->mdt_identity_cache = NULL;
+err_fs_cleanup:
+ mdt_fs_cleanup(env, m);
+err_tgt:
+ tgt_fini(env, &m->mdt_lut);
err_capa:
- cfs_timer_disarm(&m->mdt_ck_timer);
- mdt_ck_thread_stop(m);
+ cfs_timer_disarm(&m->mdt_ck_timer);
+ mdt_ck_thread_stop(m);
+err_free_hsm:
+ mdt_hsm_cdt_fini(m);
err_free_ns:
- ldlm_namespace_free(m->mdt_namespace, NULL, 0);
- obd->obd_namespace = m->mdt_namespace = NULL;
+ ldlm_namespace_free(m->mdt_namespace, NULL, 0);
+ obd->obd_namespace = m->mdt_namespace = NULL;
err_fini_seq:
mdt_seq_fini(env, m);
err_fini_fld:
mdt_fld_fini(env, m);
-err_lut:
- tgt_fini(env, &m->mdt_lut);
err_fini_stack:
- mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
+ mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
err_lmi:
if (lmi)
server_put_mount(dev, lmi->lmi_mnt);
- return (rc);
+ return(rc);
}
/* For interoperability, the left element is old parameter, the right one
ENTRY;
- OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, CFS_ALLOC_IO);
+ OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, __GFP_IO);
if (mo != NULL) {
struct lu_object *o;
struct lu_object_header *h;
- o = &mo->mot_obj.mo_lu;
+ o = &mo->mot_obj;
h = &mo->mot_header;
lu_object_header_init(h);
lu_object_init(o, h, d);
o->lo_ops = &mdt_obj_ops;
mutex_init(&mo->mot_ioepoch_mutex);
mutex_init(&mo->mot_lov_mutex);
- RETURN(o);
- } else
- RETURN(NULL);
+ init_rwsem(&mo->mot_open_sem);
+ init_rwsem(&mo->mot_xattr_sem);
+ RETURN(o);
+ }
+ RETURN(NULL);
}
static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
CDEBUG(D_INFO, "object free, fid = "DFID"\n",
PFID(lu_object_fid(o)));
- lu_object_fini(o);
- lu_object_header_fini(h);
+ LASSERT(atomic_read(&mo->mot_open_count) == 0);
+ LASSERT(atomic_read(&mo->mot_lease_count) == 0);
+
+ lu_object_fini(o);
+ lu_object_header_fini(h);
OBD_SLAB_FREE_PTR(mo, mdt_object_kmem);
- EXIT;
+ EXIT;
}
static int mdt_object_print(const struct lu_env *env, void *cookie,
struct mdt_device *mdt = mdt_dev(cdev);
struct lu_device *next = &mdt->mdt_child->md_lu_dev;
struct obd_device *obd = cdev->ld_obd;
+ struct lfsck_start_param lsp;
int rc;
ENTRY;
if (rc)
RETURN(rc);
+ rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
+ if (rc)
+ RETURN(rc);
+
+ lsp.lsp_start = NULL;
+ lsp.lsp_namespace = mdt->mdt_namespace;
rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
OBD_IOC_START_LFSCK,
- 0, NULL);
+ 0, &lsp);
if (rc != 0) {
CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
mdt_obd_name(mdt), rc);
if (rc)
RETURN(rc);
}
+
LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
target_recovery_init(&mdt->mdt_lut, mdt_recovery_handle);
set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
}
}
+ data->ocd_max_easize = mdt->mdt_max_ea_size;
+
return 0;
}
-static int mdt_connect_check_sptlrpc(struct mdt_device *mdt,
- struct obd_export *exp,
- struct ptlrpc_request *req)
+int mdt_connect_check_sptlrpc(struct mdt_device *mdt, struct obd_export *exp,
+ struct ptlrpc_request *req)
{
struct sptlrpc_flavor flvr;
int rc = 0;
struct obd_connect_data *data,
void *localdata)
{
- struct mdt_thread_info *info;
struct obd_export *lexp;
struct lustre_handle conn = { 0 };
struct mdt_device *mdt;
- struct ptlrpc_request *req;
int rc;
ENTRY;
if (!exp || !obd || !cluuid)
RETURN(-EINVAL);
- info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
- req = info->mti_pill->rc_req;
- mdt = mdt_dev(obd->obd_lu_dev);
+ mdt = mdt_dev(obd->obd_lu_dev);
/*
* first, check whether the stack is ready to handle requests
lexp = class_conn2export(&conn);
LASSERT(lexp != NULL);
- rc = mdt_connect_check_sptlrpc(mdt, lexp, req);
- if (rc)
- GOTO(out, rc);
-
- if (OBD_FAIL_CHECK(OBD_FAIL_TGT_RCVG_FLAG))
- lustre_msg_add_op_flags(req->rq_repmsg, MSG_CONNECT_RECOVERING);
-
rc = mdt_connect_internal(lexp, mdt, data);
if (rc == 0) {
struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
LASSERT(lcd);
- info->mti_exp = lexp;
memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
rc = tgt_client_new(env, lexp);
if (rc == 0)
mdt_export_stats_init(obd, lexp, localdata);
}
-out:
if (rc != 0) {
class_disconnect(lexp);
*exp = NULL;
struct obd_connect_data *data,
void *localdata)
{
- struct mdt_thread_info *info;
- struct mdt_device *mdt;
- struct ptlrpc_request *req;
int rc;
ENTRY;
if (exp == NULL || obd == NULL || cluuid == NULL)
RETURN(-EINVAL);
- info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
- req = info->mti_pill->rc_req;
- mdt = mdt_dev(obd->obd_lu_dev);
-
- rc = mdt_connect_check_sptlrpc(mdt, exp, req);
- if (rc)
- RETURN(rc);
-
rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
if (rc == 0)
mdt_export_stats_init(obd, exp, localdata);
RETURN(rc);
}
+static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
+ struct mdt_thread_info *info,
+ struct mdt_file_data *mfd)
+{
+ struct lu_context ses;
+ int rc;
+ ENTRY;
+
+ rc = lu_context_init(&ses, LCT_SESSION);
+ if (rc)
+ RETURN(rc);
+
+ env->le_ses = &ses;
+ lu_context_enter(&ses);
+
+ mdt_ucred(info)->uc_valid = UCRED_OLD;
+ rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
+
+ lu_context_exit(&ses);
+ lu_context_fini(&ses);
+ env->le_ses = NULL;
+
+ RETURN(rc);
+}
+
static int mdt_export_cleanup(struct obd_export *exp)
{
struct mdt_export_data *med = &exp->exp_mdt_data;
cfs_list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
cfs_list_del_init(&mfd->mfd_list);
ma->ma_need = ma->ma_valid = 0;
+
+ /* This file is being closed due to an eviction, it
+ * could have been modified and now dirty regarding to
+ * HSM archive, check this!
+ * The logic here is to mark a file dirty if there's a
+ * chance it was dirtied before the client was evicted,
+ * so that we don't have to wait for a release attempt
+ * before finding out the file was actually dirty and
+ * fail the release. Aggressively marking it dirty here
+ * will cause the policy engine to attempt to
+ * re-archive it; when rearchiving, we can compare the
+ * current version to the HSM data_version and make the
+ * archive request into a noop if it's not actually
+ * dirty.
+ */
+ if (mfd->mfd_mode & (FMODE_WRITE|MDS_FMODE_TRUNC))
+ rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
+
/* Don't unlink orphan on failover umount, LU-184 */
if (exp->exp_flags & OBD_OPT_FAILOVER) {
ma->ma_valid = MA_FLAGS;
if (rc < 0)
return rc;
- linkea_init(ldata);
-
- return 0;
+ return linkea_init(ldata);
}
static int mdt_path_current(struct mdt_thread_info *info,
rc = mdt_links_read(info, mdt_obj, &ldata);
mdt_object_put(info->mti_env, mdt_obj);
if (rc != 0)
- GOTO(out, rc = PTR_ERR(buf));
+ GOTO(out, rc);
leh = buf->lb_buf;
lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
case OBD_IOC_CHANGELOG_CLEAR:
rc = mdt_ioc_child(&env, mdt, cmd, len, karg);
break;
- case OBD_IOC_START_LFSCK:
- case OBD_IOC_STOP_LFSCK: {
+ case OBD_IOC_START_LFSCK: {
struct md_device *next = mdt->mdt_child;
struct obd_ioctl_data *data = karg;
+ struct lfsck_start_param lsp;
if (unlikely(data == NULL)) {
rc = -EINVAL;
break;
}
- rc = next->md_ops->mdo_iocontrol(&env, next, cmd,
- data->ioc_inllen1,
- data->ioc_inlbuf1);
+ lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
+ lsp.lsp_namespace = mdt->mdt_namespace;
+ rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
+ break;
+ }
+ case OBD_IOC_STOP_LFSCK: {
+ struct md_device *next = mdt->mdt_child;
+
+ rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, NULL);
break;
}
case OBD_IOC_GET_OBJ_VERSION: {
static struct lu_device *mdt_device_free(const struct lu_env *env,
struct lu_device *d)
{
- struct mdt_device *m = mdt_dev(d);
- ENTRY;
+ struct mdt_device *m = mdt_dev(d);
+ ENTRY;
- md_device_fini(&m->mdt_md_dev);
- OBD_FREE_PTR(m);
- RETURN(NULL);
+ lu_device_fini(&m->mdt_lu_dev);
+ OBD_FREE_PTR(m);
+
+ RETURN(NULL);
}
static struct lu_device *mdt_device_alloc(const struct lu_env *env,
if (m != NULL) {
int rc;
- l = &m->mdt_md_dev.md_lu_dev;
+ l = &m->mdt_lu_dev;
rc = mdt_init0(env, m, t, cfg);
if (rc != 0) {
mdt_device_free(env, l);
git://git.whamcloud.com / fs / lustre-release.git / blobdiff