*/
/*
* This file is part of Lustre, http://www.lustre.org/
- * Lustre is a trademark of Sun Microsystems, Inc.
*
* lustre/mdt/mdt_open.c
*
GOTO(out, rc = -EPROTO);
}
- /* XXX: only exclusive open is supported. */
- lm = LCK_EX;
+ /* should conflict with new opens for write/execute */
+ lm = LCK_PW;
*ibits = MDS_INODELOCK_OPEN;
/* never grant LCK_EX layout lock to client */
PFID(mdt_object_fid(obj)),
atomic_read(&obj->mot_open_count), open_count);
- if (atomic_read(&obj->mot_open_count) > open_count)
- GOTO(out, rc = -EBUSY);
+ if (atomic_read(&obj->mot_open_count) > open_count) {
+ /* fail if anyone *else* has opened file for write */
+ if (mdt_write_read(obj) > 1)
+ GOTO(out, rc = -EBUSY);
+ }
}
GOTO(out, rc);
struct ptlrpc_request *req = mdt_info_req(info);
struct mdt_object *parent;
struct mdt_object *child;
- struct mdt_lock_handle *lh;
+ struct mdt_lock_handle *lh = NULL;
struct ldlm_reply *ldlm_rep;
struct mdt_body *repbody;
struct lu_fid *child_fid = &info->mti_tmp_fid1;
OBD_RACE(OBD_FAIL_MDS_REINT_OPEN);
again_pw:
- lh = &info->mti_lh[MDT_LH_PARENT];
- mdt_lock_pdo_init(lh, lock_mode, &rr->rr_name);
-
- result = mdt_object_lock(info, parent, lh, MDS_INODELOCK_UPDATE);
- if (result != 0) {
- mdt_object_put(info->mti_env, parent);
- GOTO(out, result);
- }
fid_zero(child_fid);
- result = -ENOENT;
- if ((open_flags & MDS_OPEN_VOLATILE) == 0)
+ if (open_flags & MDS_OPEN_VOLATILE) {
+ lh = NULL;
+ result = -ENOENT;
+ } else {
+ lh = &info->mti_lh[MDT_LH_PARENT];
+ mdt_lock_pdo_init(lh, lock_mode, &rr->rr_name);
+ result = mdt_object_lock(info, parent, lh, MDS_INODELOCK_UPDATE);
+ if (result != 0) {
+ mdt_object_put(info->mti_env, parent);
+ GOTO(out, result);
+ }
+
result = mdo_lookup(info->mti_env, mdt_object_child(parent),
&rr->rr_name, child_fid, &info->mti_spec);
+ }
LASSERTF(ergo(result == 0, fid_is_sane(child_fid)),
"looking for "DFID"/"DNAME", found FID = "DFID"\n",
if (mdt_rdonly(req->rq_export))
GOTO(out_parent, result = -EROFS);
- if (lock_mode == LCK_PR) {
+ LASSERT(equi(lh == NULL, open_flags & MDS_OPEN_VOLATILE));
+
+ if (lh != NULL && lock_mode == LCK_PR) {
/* first pass: get write lock and restart */
mdt_object_unlock(info, parent, lh, 1);
mdt_clear_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
/* Not found and with MDS_OPEN_CREAT: let's create it. */
mdt_set_disposition(info, ldlm_rep, DISP_OPEN_CREATE);
- /* Let lower layers know what is lock mode on directory. */
- info->mti_spec.sp_cr_mode =
- mdt_dlm_mode2mdl_mode(lh->mlh_pdo_mode);
-
/* Don't do lookup sanity check. We know name doesn't exist. */
info->mti_spec.sp_cr_lookup = 0;
info->mti_spec.sp_feat = &dt_directory_features;
if (result == 0)
mdt_pack_size2body(info, child_fid, &lhc->mlh_reg_lh);
out_parent:
- mdt_object_unlock_put(info, parent, lh, result || !created);
+ if (lh != NULL)
+ mdt_object_unlock(info, parent, lh, result || !created);
+
+ mdt_object_put(info->mti_env, parent);
out:
if (result)
lustre_msg_set_transno(req->rq_repmsg, 0);
.ln_namelen = sizeof("i_am_nobody") - 1,
};
struct lu_ucred *uc;
- cfs_cap_t uc_cap_save;
+ kernel_cap_t uc_cap_save;
int rc;
ENTRY;
spec->sp_cr_lookup = 0;
spec->sp_feat = &dt_directory_features;
- spec->sp_cr_mode = MDL_MINMODE; /* no lock */
spec->sp_cr_flags = MDS_OPEN_VOLATILE | fmode;
if (attr->ma_valid & MA_LOV) {
spec->u.sp_ea.eadata = attr->ma_lmm;
uc = lu_ucred(env);
uc_cap_save = uc->uc_cap;
- uc->uc_cap |= BIT(CAP_DAC_OVERRIDE);
+ cap_raise(uc->uc_cap, CAP_DAC_OVERRIDE);
rc = mdo_create(env, mdt_object_child(local_root), &lname,
mdt_object_child(obj), spec, attr);
uc->uc_cap = uc_cap_save;
struct mdt_object *orphan;
struct md_attr *orp_ma;
struct lu_buf *buf;
- cfs_cap_t cap;
+ kernel_cap_t cap;
bool lease_broken;
int rc;
int rc2;
/* The orphan has root ownership so we need to raise
* CAP_FOWNER to set the HSM attributes. */
cap = uc->uc_cap;
- uc->uc_cap |= MD_CAP_TO_MASK(CAP_FOWNER);
+ cap_raise(uc->uc_cap, CAP_FOWNER);
rc = mo_xattr_set(info->mti_env, mdt_object_child(orphan), buf,
XATTR_NAME_HSM, 0);
uc->uc_cap = cap;
}
out_reprocess:
- ldlm_reprocess_all(lease->l_resource, lease);
+ ldlm_reprocess_all(lease->l_resource,
+ lease->l_policy_data.l_inodebits.bits);
LDLM_LOCK_PUT(lease);
ma->ma_valid = 0;
struct mdt_lock_handle *lh2 = &info->mti_lh[MDT_LH_OLD];
struct close_data *data;
struct ldlm_lock *lease;
- struct mdt_object *o1 = o, *o2;
+ struct mdt_object *o1 = o, *o2 = NULL;
bool lease_broken;
- bool swap_objects;
+ bool swap_objects = false;
int rc;
ENTRY;
RETURN(-EINVAL);
rc = lu_fid_cmp(&data->cd_fid, mdt_object_fid(o));
- if (unlikely(rc == 0))
- RETURN(-EINVAL);
+ if (rc == 0) {
+ /**
+ * only MDS_CLOSE_LAYOUT_SPLIT use the same fid to indicate
+ * mirror deletion, so we'd zero cd_fid, and keeps o2 be NULL.
+ */
+ if (!(ma->ma_attr_flags & MDS_CLOSE_LAYOUT_SPLIT))
+ RETURN(-EINVAL);
- /* Exchange o1 and o2, to enforce locking order */
- swap_objects = (rc < 0);
+ /* zero cd_fid to keeps o2 be NULL */
+ fid_zero(&data->cd_fid);
+ } else if (rc < 0) {
+ /* Exchange o1 and o2, to enforce locking order */
+ swap_objects = true;
+ }
lease = ldlm_handle2lock(&data->cd_handle);
if (lease == NULL)
RETURN(-ESTALE);
- o2 = mdt_object_find(info->mti_env, info->mti_mdt, &data->cd_fid);
- if (IS_ERR(o2))
- GOTO(out_lease, rc = PTR_ERR(o2));
+ if (!fid_is_zero(&data->cd_fid)) {
+ o2 = mdt_object_find(info->mti_env, info->mti_mdt,
+ &data->cd_fid);
+ if (IS_ERR(o2))
+ GOTO(out_lease, rc = PTR_ERR(o2));
- if (!S_ISREG(lu_object_attr(&o2->mot_obj))) {
- swap_objects = false; /* not swapped yet */
- GOTO(out_obj, rc = -EINVAL);
- }
+ if (!mdt_object_exists(o2))
+ GOTO(out_obj, rc = -ENOENT);
+
+ if (!S_ISREG(lu_object_attr(&o2->mot_obj)))
+ GOTO(out_obj, rc = -EINVAL);
- if (swap_objects)
- swap(o1, o2);
+ if (swap_objects)
+ swap(o1, o2);
+ }
rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
MAY_WRITE);
if (rc < 0)
GOTO(out_obj, rc);
- rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
- MAY_WRITE);
- if (rc < 0)
- GOTO(out_obj, rc);
+ if (o2) {
+ rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2),
+ NULL, MAY_WRITE);
+ if (rc < 0)
+ GOTO(out_obj, rc);
+ }
/* try to hold open_sem so that nobody else can open the file */
if (!down_write_trylock(&o->mot_open_sem)) {
if (rc < 0)
GOTO(out_unlock_sem, rc);
- mdt_lock_reg_init(lh2, LCK_EX);
- rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
- MDS_INODELOCK_XATTR);
- if (rc < 0)
- GOTO(out_unlock1, rc);
+ if (o2) {
+ mdt_lock_reg_init(lh2, LCK_EX);
+ rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
+ MDS_INODELOCK_XATTR);
+ if (rc < 0)
+ GOTO(out_unlock1, rc);
+ }
/* Swap layout with orphan object */
if (ma->ma_attr_flags & MDS_CLOSE_LAYOUT_SWAP) {
struct lu_buf *buf = &info->mti_buf;
struct md_rejig_data mrd;
- mrd.mrd_obj = mdt_object_child(o == o1 ? o2 : o1);
- if (ma->ma_attr_flags & MDS_CLOSE_LAYOUT_SPLIT)
+ if (o2) {
+ mrd.mrd_obj = mdt_object_child(o == o1 ? o2 : o1);
+ } else {
+ if (!(ma->ma_attr_flags & MDS_CLOSE_LAYOUT_SPLIT)) {
+ /* paranoid check again */
+ CERROR(DFID
+ ":only mirror split support NULL o2 object\n",
+ PFID(mdt_object_fid(o)));
+ GOTO(out_unlock1, rc = -EINVAL);
+ }
+
+ /* set NULL mrd_obj for deleting mirror objects */
+ mrd.mrd_obj = NULL;
+ }
+
+ if (ma->ma_attr_flags & MDS_CLOSE_LAYOUT_SPLIT) {
mrd.mrd_mirror_id = data->cd_mirror_id;
+ /* set a small enough blocks in the SoM */
+ ma->ma_attr.la_blocks >>= 1;
+ }
buf->lb_len = sizeof(mrd);
buf->lb_buf = &mrd;
XATTR_LUSTRE_LOV,
ma->ma_attr_flags & MDS_CLOSE_LAYOUT_SPLIT ?
LU_XATTR_SPLIT : LU_XATTR_MERGE);
- if (rc == 0 && ma->ma_attr.la_valid & (LA_SIZE | LA_BLOCKS)) {
+ if (rc == 0 && ma->ma_attr.la_valid & (LA_SIZE | LA_BLOCKS |
+ LA_LSIZE | LA_LBLOCKS)) {
int rc2;
+ enum lustre_som_flags lsf;
+
+ if (ma->ma_attr.la_valid & (LA_SIZE | LA_BLOCKS))
+ lsf = SOM_FL_STRICT;
+ else
+ lsf = SOM_FL_LAZY;
mutex_lock(&o->mot_som_mutex);
- rc2 = mdt_set_som(info, o, SOM_FL_STRICT,
+ rc2 = mdt_set_som(info, o, lsf,
ma->ma_attr.la_size,
ma->ma_attr.la_blocks);
mutex_unlock(&o->mot_som_mutex);
out_unlock2:
/* Release exclusive LL */
- mdt_object_unlock(info, o2, lh2, 1);
+ if (o2)
+ mdt_object_unlock(info, o2, lh2, 1);
out_unlock1:
mdt_object_unlock(info, o1, lh1, 1);
}
out_obj:
- mdt_object_put(info->mti_env, swap_objects ? o1 : o2);
+ if (o1 != o)
+ /* the 2nd object has been used, and swapped to o1 */
+ mdt_object_put(info->mti_env, o1);
+ else if (o2)
+ /* the 2nd object has been used, and not swapped */
+ mdt_object_put(info->mti_env, o2);
- ldlm_reprocess_all(lease->l_resource, lease);
+ ldlm_reprocess_all(lease->l_resource,
+ lease->l_policy_data.l_inodebits.bits);
out_lease:
LDLM_LOCK_PUT(lease);
if (data == NULL)
RETURN(-EPROTO);
- if (ptlrpc_req_need_swab(mdt_info_req(info)))
+ if (req_capsule_req_need_swab(info->mti_pill))
lustre_swab_close_data_resync_done(&data->cd_resync);
if (!fid_is_zero(&data->cd_fid))
OBD_FREE_PTR_ARRAY(resync_ids, resync_count);
out_reprocess:
- ldlm_reprocess_all(lease->l_resource, lease);
+ ldlm_reprocess_all(lease->l_resource,
+ lease->l_policy_data.l_inodebits.bits);
LDLM_LOCK_PUT(lease);
ma->ma_valid = 0;
git://git.whamcloud.com / fs / lustre-release.git / blobdiff