#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[] = {
}
static int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
- char *name)
+ const char *name)
{
const struct lu_env *env = info->mti_env;
int rc;
RETURN(rc);
}
-int mdt_attr_get_lov(struct mdt_thread_info *info,
- struct mdt_object *o, struct md_attr *ma)
+int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
+ struct md_attr *ma, const char *name)
{
struct md_object *next = mdt_object_child(o);
struct lu_buf *buf = &info->mti_buf;
int rc;
- buf->lb_buf = ma->ma_lmm;
- buf->lb_len = ma->ma_lmm_size;
- rc = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_LOV);
+ if (strcmp(name, XATTR_NAME_LOV) == 0) {
+ buf->lb_buf = ma->ma_lmm;
+ buf->lb_len = ma->ma_lmm_size;
+ LASSERT(!(ma->ma_valid & MA_LOV));
+ } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
+ buf->lb_buf = ma->ma_lmv;
+ buf->lb_len = ma->ma_lmv_size;
+ LASSERT(!(ma->ma_valid & MA_LMV));
+ } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
+ buf->lb_buf = ma->ma_lmv;
+ buf->lb_len = ma->ma_lmv_size;
+ LASSERT(!(ma->ma_valid & MA_LMV_DEF));
+ } else {
+ return -EINVAL;
+ }
+
+ rc = mo_xattr_get(info->mti_env, next, buf, name);
if (rc > 0) {
- ma->ma_lmm_size = rc;
- ma->ma_valid |= MA_LOV;
+ if (strcmp(name, XATTR_NAME_LOV) == 0) {
+ ma->ma_lmm_size = rc;
+ ma->ma_valid |= MA_LOV;
+ } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
+ ma->ma_lmv_size = rc;
+ ma->ma_valid |= MA_LMV;
+ } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
+ ma->ma_lmv_size = rc;
+ ma->ma_valid |= MA_LMV_DEF;
+ }
+
rc = 0;
} else if (rc == -ENODATA) {
/* no LOV EA */
rc = 0;
} else if (rc == -ERANGE) {
- rc = mdt_big_xattr_get(info, o, XATTR_NAME_LOV);
+ /* Default LMV has fixed size, so it must be able to fit
+ * in the original buffer */
+ if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
+ return rc;
+ rc = mdt_big_xattr_get(info, o, name);
if (rc > 0) {
info->mti_big_lmm_used = 1;
- ma->ma_valid |= MA_LOV;
- ma->ma_lmm = info->mti_big_lmm;
- ma->ma_lmm_size = rc;
+ if (!strcmp(name, XATTR_NAME_LOV)) {
+ ma->ma_valid |= MA_LOV;
+ ma->ma_lmm = info->mti_big_lmm;
+ ma->ma_lmm_size = rc;
+ } else if (!strcmp(name, XATTR_NAME_LMV)) {
+ ma->ma_valid |= MA_LMV;
+ ma->ma_lmv = info->mti_big_lmm;
+ ma->ma_lmv_size = rc;
+ } else {
+ return -EINVAL;
+ }
+
/* update mdt_max_mdsize so all clients
* will be aware about that */
if (info->mti_mdt->mdt_max_mdsize < rc)
}
if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
- rc = mdt_attr_get_lov(info, o, ma);
+ rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
if (rc)
GOTO(out, rc);
}
if (need & MA_LMV && S_ISDIR(mode)) {
- buf->lb_buf = ma->ma_lmv;
- buf->lb_len = ma->ma_lmv_size;
- rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_LMV);
- if (rc2 > 0) {
- ma->ma_lmv_size = rc2;
- ma->ma_valid |= MA_LMV;
- } else if (rc2 == -ENODATA) {
- /* no LMV EA */
- ma->ma_lmv_size = 0;
- } else
- GOTO(out, rc = rc2);
+ rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
+ if (rc != 0)
+ GOTO(out, rc);
+ }
+
+ if (need & MA_LMV_DEF && S_ISDIR(mode)) {
+ rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
+ if (rc != 0)
+ GOTO(out, rc);
}
if (need & MA_SOM && S_ISREG(mode)) {
/* If it is dir object and client require MEA, then we got MEA */
if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
- reqbody->valid & OBD_MD_MEA) {
+ (reqbody->valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
/* Assumption: MDT_MD size is enough for lmv size. */
ma->ma_lmv = buffer->lb_buf;
ma->ma_lmv_size = buffer->lb_len;
ma->ma_need = MA_INODE;
- if (ma->ma_lmm_size > 0)
- ma->ma_need |= MA_LMV;
+ if (ma->ma_lmv_size > 0) {
+ if (reqbody->valid & OBD_MD_MEA)
+ ma->ma_need |= MA_LMV;
+ else if (reqbody->valid & OBD_MD_DEFAULT_MEA)
+ ma->ma_need |= MA_LMV_DEF;
+ }
} else {
ma->ma_lmm = buffer->lb_buf;
ma->ma_lmm_size = buffer->lb_len;
root = mdt_object_find(env, mdt, &rootfid);
if (IS_ERR(root))
RETURN(PTR_ERR(root));
- rc = mdt_attr_get_lov(info, root, ma);
+ rc = mdt_stripe_get(info, root, ma, XATTR_NAME_LOV);
mdt_object_put(info->mti_env, root);
if (unlikely(rc)) {
CERROR("%s: getattr error for "DFID": rc = %d\n",
else
repbody->valid |= OBD_MD_FLEASIZE;
}
- if (ma->ma_valid & MA_LMV) {
- LASSERT(S_ISDIR(la->la_mode));
- repbody->eadatasize = ma->ma_lmv_size;
- repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
- }
+ if (ma->ma_valid & MA_LMV) {
+ LASSERT(S_ISDIR(la->la_mode));
+ mdt_dump_lmv(D_INFO, ma->ma_lmv);
+ repbody->eadatasize = ma->ma_lmv_size;
+ repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
+ }
+ if (ma->ma_valid & MA_LMV_DEF) {
+ LASSERT(S_ISDIR(la->la_mode));
+ repbody->eadatasize = ma->ma_lmv_size;
+ repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_DEFAULT_MEA);
+ }
} else if (S_ISLNK(la->la_mode) &&
reqbody->valid & OBD_MD_LINKNAME) {
buffer->lb_buf = ma->ma_lmm;
ma_need |= MA_LOV;
}
} else {
+ /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
+ * client will enqueue the lock to the remote MDT */
+ if (mdt_object_remote(child))
+ child_bits &= ~MDS_INODELOCK_UPDATE;
rc = mdt_object_lock(info, child, lhc, child_bits,
MDT_CROSS_LOCK);
}
GOTO(out_ucred, rc = err_serious(rc));
if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
- rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
+ DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
+ rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
GOTO(out_ucred, rc);
}
rc = mdt_reint_rec(info, lhc);
[REINT_RENAME] = &RQF_MDS_REINT_RENAME,
[REINT_OPEN] = &RQF_MDS_REINT_OPEN,
[REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
- [REINT_RMENTRY] = &RQF_MDS_REINT_UNLINK
+ [REINT_RMENTRY] = &RQF_MDS_REINT_UNLINK,
+ [REINT_MIGRATE] = &RQF_MDS_REINT_RENAME
};
ENTRY;
RETURN(rc);
}
-int mdt_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
- void *data, int flag)
+/* Used for cross-MDT lock */
+int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
+ void *data, int flag)
{
struct lustre_handle lockh;
int rc;
}
int mdt_remote_object_lock(struct mdt_thread_info *mti,
- struct mdt_object *o, struct lustre_handle *lh,
- ldlm_mode_t mode, __u64 ibits)
+ struct mdt_object *o, const struct lu_fid *fid,
+ struct lustre_handle *lh, ldlm_mode_t mode,
+ __u64 ibits)
{
struct ldlm_enqueue_info *einfo = &mti->mti_einfo;
ldlm_policy_data_t *policy = &mti->mti_policy;
+ struct ldlm_res_id *res_id = &mti->mti_res_id;
int rc = 0;
ENTRY;
LASSERT(mdt_object_remote(o));
- LASSERT(ibits & MDS_INODELOCK_UPDATE);
+ LASSERT(ibits == MDS_INODELOCK_UPDATE);
+
+ fid_build_reg_res_name(fid, res_id);
memset(einfo, 0, sizeof(*einfo));
einfo->ei_type = LDLM_IBITS;
einfo->ei_mode = mode;
- einfo->ei_cb_bl = mdt_md_blocking_ast;
+ einfo->ei_cb_bl = mdt_remote_blocking_ast;
einfo->ei_cb_cp = ldlm_completion_ast;
+ einfo->ei_enq_slave = 0;
+ einfo->ei_res_id = res_id;
memset(policy, 0, sizeof(*policy));
policy->l_inodebits.bits = ibits;
RETURN(rc);
}
-static int mdt_object_lock0(struct mdt_thread_info *info, struct mdt_object *o,
- struct mdt_lock_handle *lh, __u64 ibits,
- bool nonblock, int locality)
+static int mdt_object_local_lock(struct mdt_thread_info *info,
+ struct mdt_object *o,
+ struct mdt_lock_handle *lh, __u64 ibits,
+ bool nonblock, int locality)
{
struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
ldlm_policy_data_t *policy = &info->mti_policy;
LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
LASSERT(lh->mlh_type != MDT_NUL_LOCK);
- if (mdt_object_remote(o)) {
- if (locality == MDT_CROSS_LOCK) {
- ibits &= ~(MDS_INODELOCK_UPDATE | MDS_INODELOCK_PERM |
- MDS_INODELOCK_LAYOUT);
- ibits |= MDS_INODELOCK_LOOKUP;
- } else {
- LASSERTF(!(ibits &
- (MDS_INODELOCK_UPDATE | MDS_INODELOCK_PERM |
- MDS_INODELOCK_LAYOUT)),
- "%s: wrong bit "LPX64" for remote obj "DFID"\n",
- mdt_obd_name(info->mti_mdt), ibits,
- PFID(mdt_object_fid(o)));
- LASSERT(ibits & MDS_INODELOCK_LOOKUP);
- }
- /* No PDO lock on remote object */
- LASSERT(lh->mlh_type != MDT_PDO_LOCK);
- }
+ /* Only enqueue LOOKUP lock for remote object */
+ if (mdt_object_remote(o))
+ LASSERT(ibits == MDS_INODELOCK_LOOKUP);
if (lh->mlh_type == MDT_PDO_LOCK) {
/* check for exists after object is locked */
RETURN(rc);
}
+int mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
+ struct mdt_lock_handle *lh, __u64 ibits,
+ bool nonblock, int locality)
+{
+ int rc;
+ ENTRY;
+
+ if (!mdt_object_remote(o))
+ return mdt_object_local_lock(info, o, lh, ibits, nonblock,
+ locality);
+
+ if (locality == MDT_LOCAL_LOCK) {
+ CERROR("%s: try to get local lock for remote object"
+ DFID".\n", mdt_obd_name(info->mti_mdt),
+ PFID(mdt_object_fid(o)));
+ RETURN(-EPROTO);
+ }
+
+ /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
+ ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
+ MDS_INODELOCK_XATTR);
+ if (ibits & MDS_INODELOCK_UPDATE) {
+ /* Sigh, PDO needs to enqueue 2 locks right now, but
+ * enqueue RPC can only request 1 lock, to avoid extra
+ * RPC, so it will instead enqueue EX lock for remote
+ * object anyway XXX*/
+ if (lh->mlh_type == MDT_PDO_LOCK &&
+ lh->mlh_pdo_hash != 0) {
+ CDEBUG(D_INFO, "%s: "DFID" convert PDO lock to"
+ "EX lock.\n", mdt_obd_name(info->mti_mdt),
+ PFID(mdt_object_fid(o)));
+ lh->mlh_pdo_hash = 0;
+ lh->mlh_rreg_mode = LCK_EX;
+ lh->mlh_type = MDT_REG_LOCK;
+ }
+ rc = mdt_remote_object_lock(info, o, mdt_object_fid(o),
+ &lh->mlh_rreg_lh,
+ lh->mlh_rreg_mode,
+ MDS_INODELOCK_UPDATE);
+ if (rc != ELDLM_OK)
+ RETURN(rc);
+ }
+
+ /* Only enqueue LOOKUP lock for remote object */
+ if (ibits & MDS_INODELOCK_LOOKUP) {
+ rc = mdt_object_local_lock(info, o, lh,
+ MDS_INODELOCK_LOOKUP,
+ nonblock, locality);
+ if (rc != ELDLM_OK)
+ RETURN(rc);
+ }
+
+ RETURN(0);
+}
+
int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
struct mdt_lock_handle *lh, __u64 ibits, int locality)
{
- return mdt_object_lock0(info, o, lh, ibits, false, locality);
+ return mdt_object_lock_internal(info, o, lh, ibits, false, locality);
}
int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
struct mdt_lock_handle tmp = *lh;
int rc;
- rc = mdt_object_lock0(info, o, &tmp, ibits, true, locality);
+ rc = mdt_object_lock_internal(info, o, &tmp, ibits, true, locality);
if (rc == 0)
*lh = tmp;
info->mti_opdata = 0;
info->mti_big_lmm_used = 0;
- /* To not check for split by default. */
info->mti_spec.no_create = 0;
info->mti_spec.sp_rm_entry = 0;
+
+ info->mti_spec.u.sp_ea.eadata = NULL;
+ info->mti_spec.u.sp_ea.eadatalen = 0;
}
void mdt_thread_info_fini(struct mdt_thread_info *info)
site->ls_top_dev = &mdt->mdt_lu_dev;
mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
-
/* now connect to bottom OSD */
snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
mdt->mdt_bottom =
lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
-
rc = lu_env_refill((struct lu_env *)env);
if (rc != 0)
CERROR("Failure to refill session: '%d'\n", rc);
.tos_hs = mdt_sec_ctx_ops
},
{
- .tos_opc_start = UPDATE_OBJ,
- .tos_opc_end = UPDATE_LAST_OPC,
+ .tos_opc_start = OUT_UPDATE_FIRST_OPC,
+ .tos_opc_end = OUT_UPDATE_LAST_OPC,
.tos_hs = tgt_out_handlers
},
{
}
static struct lu_object *mdt_object_alloc(const struct lu_env *env,
- const struct lu_object_header *hdr,
- struct lu_device *d)
+ const struct lu_object_header *hdr,
+ struct lu_device *d)
{
- struct mdt_object *mo;
+ struct mdt_object *mo;
- ENTRY;
+ ENTRY;
- OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, __GFP_IO);
- if (mo != NULL) {
- struct lu_object *o;
- struct lu_object_header *h;
+ OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
+ if (mo != NULL) {
+ struct lu_object *o;
+ struct lu_object_header *h;
o = &mo->mot_obj;
- h = &mo->mot_header;
- lu_object_header_init(h);
- lu_object_init(o, h, d);
- lu_object_add_top(h, o);
- o->lo_ops = &mdt_obj_ops;
+ h = &mo->mot_header;
+ lu_object_header_init(h);
+ lu_object_init(o, h, d);
+ lu_object_add_top(h, o);
+ o->lo_ops = &mdt_obj_ops;
mutex_init(&mo->mot_ioepoch_mutex);
mutex_init(&mo->mot_lov_mutex);
init_rwsem(&mo->mot_open_sem);
if (rc)
RETURN(rc);
- lsp.lsp_namespace = mdt->mdt_namespace;
+ rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
+ /* The LFSCK instance is registered just now, so it must be there when
+ * register the namespace to such instance. */
+ LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
+
lsp.lsp_start = NULL;
lsp.lsp_index_valid = 0;
rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
rc = tgt_client_new(env, lexp);
if (rc == 0)
mdt_export_stats_init(obd, lexp, localdata);
+
+ /* For phase I, sync for cross-ref operation. */
+ lexp->exp_keep_sync = 1;
}
if (rc != 0) {
int pli_fidcount; /**< number of \a pli_fids */
};
-static int mdt_links_read(struct mdt_thread_info *info,
- struct mdt_object *mdt_obj, struct linkea_data *ldata)
+int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
+ struct linkea_data *ldata)
{
int rc;
break;
}
- lsp.lsp_namespace = mdt->mdt_namespace;
lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
lsp.lsp_index_valid = 0;
rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
case OBD_IOC_STOP_LFSCK: {
struct md_device *next = mdt->mdt_child;
struct obd_ioctl_data *data = karg;
- struct lfsck_stop *stop =
- (struct lfsck_stop *)(data->ioc_inlbuf1);
+ struct lfsck_stop stop;
+
+ stop.ls_status = LS_STOPPED;
+ /* Old lfsck utils may pass NULL @stop. */
+ if (data->ioc_inlbuf1 == NULL)
+ stop.ls_flags = 0;
+ else
+ stop.ls_flags =
+ ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
- stop->ls_status = LS_STOPPED;
- rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, stop);
+ rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
break;
}
case OBD_IOC_GET_OBJ_VERSION: {
rc = mdt_ioc_version_get(mti, karg);
break;
}
- case OBD_IOC_CATLOGLIST:
- rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg);
+ case OBD_IOC_CATLOGLIST: {
+ struct mdt_thread_info *mti;
+
+ mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
+ lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
+ rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
+ &mti->mti_tmp_fid1);
break;
+ }
default:
rc = -EOPNOTSUPP;
CERROR("%s: Not supported cmd = %d, rc = %d\n",