static int ofd_preprw_read(const struct lu_env *env, struct obd_export *exp,
struct ofd_device *ofd, struct lu_fid *fid,
struct lu_attr *la, int niocount,
struct niobuf_remote *rnb, int *nr_local,
struct niobuf_local *lnb, char *jobid)
{
struct ofd_object *fo;
int i, j, rc, tot_bytes = 0;
ENTRY;
LASSERT(env != NULL);
fo = ofd_object_find(env, ofd, fid);
if (IS_ERR(fo))
RETURN(PTR_ERR(fo));
LASSERT(fo != NULL);
ofd_read_lock(env, fo);
if (!ofd_object_exists(fo))
GOTO(unlock, rc = -ENOENT);
/* parse remote buffers to local buffers and prepare the latter */
*nr_local = 0;
for (i = 0, j = 0; i < niocount; i++) {
rc = dt_bufs_get(env, ofd_object_child(fo), rnb + i,
lnb + j, 0, ofd_object_capa(env, fo));
if (unlikely(rc < 0))
GOTO(buf_put, rc);
LASSERT(rc <= PTLRPC_MAX_BRW_PAGES);
/* correct index for local buffers to continue with */
j += rc;
*nr_local += rc;
LASSERT(j <= PTLRPC_MAX_BRW_PAGES);
tot_bytes += rnb[i].rnb_len;
}
LASSERT(*nr_local > 0 && *nr_local <= PTLRPC_MAX_BRW_PAGES);
rc = dt_attr_get(env, ofd_object_child(fo), la,
ofd_object_capa(env, fo));
if (unlikely(rc))
GOTO(buf_put, rc);
rc = dt_read_prep(env, ofd_object_child(fo), lnb, *nr_local);
if (unlikely(rc))
GOTO(buf_put, rc);
ofd_counter_incr(exp, LPROC_OFD_STATS_READ, jobid, tot_bytes);
RETURN(0);
buf_put:
dt_bufs_put(env, ofd_object_child(fo), lnb, *nr_local);
unlock:
ofd_read_unlock(env, fo);
ofd_object_put(env, fo);
return rc;
}
int ofd_attr_get(const struct lu_env *env, struct ofd_object *fo,
struct lu_attr *la)
{
int rc = 0;
ENTRY;
if (ofd_object_exists(fo)) {
rc = dt_attr_get(env, ofd_object_child(fo), la,
ofd_object_capa(env, fo));
#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0)
/* Try to correct for a bug in 2.1.0 (LU-221) that caused
* negative timestamps to appear to be in the far future,
* due old timestamp being stored on disk as an unsigned value.
* This fixes up any bad values stored on disk before
* returning them to the client, and ensures any timestamp
* updates are correct. LU-1042 */
if (unlikely(la->la_atime == LU221_BAD_TIME))
la->la_atime = 0;
if (unlikely(la->la_mtime == LU221_BAD_TIME))
la->la_mtime = 0;
if (unlikely(la->la_ctime == LU221_BAD_TIME))
la->la_ctime = 0;
#else
#warning "remove old LU-221/LU-1042 workaround code"
#endif
} else {
rc = -ENOENT;
}
RETURN(rc);
}
int ofd_object_punch(const struct lu_env *env, struct ofd_object *fo,
__u64 start, __u64 end, struct lu_attr *la,
struct filter_fid *ff, struct obdo *oa)
{
struct ofd_thread_info *info = ofd_info(env);
struct ofd_device *ofd = ofd_obj2dev(fo);
struct ofd_mod_data *fmd;
struct dt_object *dob = ofd_object_child(fo);
struct thandle *th;
int ff_needed = 0;
int rc;
ENTRY;
/* we support truncate, not punch yet */
LASSERT(end == OBD_OBJECT_EOF);
fmd = ofd_fmd_get(info->fti_exp, &fo->ofo_header.loh_fid);
if (fmd && fmd->fmd_mactime_xid < info->fti_xid)
fmd->fmd_mactime_xid = info->fti_xid;
ofd_fmd_put(info->fti_exp, fmd);
ofd_write_lock(env, fo);
if (!ofd_object_exists(fo))
GOTO(unlock, rc = -ENOENT);
if (ofd->ofd_lfsck_verify_pfid && oa->o_valid & OBD_MD_FLFID) {
rc = ofd_verify_ff(env, fo, oa);
if (rc != 0)
GOTO(unlock, rc);
}
/* VBR: version recovery check */
rc = ofd_version_get_check(info, fo);
if (rc)
GOTO(unlock, rc);
rc = ofd_attr_handle_ugid(env, fo, la, 0 /* !is_setattr */);
if (rc != 0)
GOTO(unlock, rc);
if (ff != NULL) {
rc = ofd_object_ff_load(env, fo);
if (rc == -ENODATA)
ff_needed = 1;
else if (rc < 0)
GOTO(unlock, rc);
}
th = ofd_trans_create(env, ofd);
if (IS_ERR(th))
GOTO(unlock, rc = PTR_ERR(th));
rc = dt_declare_attr_set(env, dob, la, th);
if (rc)
GOTO(stop, rc);
rc = dt_declare_punch(env, dob, start, OBD_OBJECT_EOF, th);
if (rc)
GOTO(stop, rc);
if (ff_needed) {
info->fti_buf.lb_buf = ff;
info->fti_buf.lb_len = sizeof(*ff);
rc = dt_declare_xattr_set(env, ofd_object_child(fo),
&info->fti_buf, XATTR_NAME_FID, 0,
th);
if (rc)
GOTO(stop, rc);
}
rc = ofd_trans_start(env, ofd, fo, th);
if (rc)
GOTO(stop, rc);
rc = dt_punch(env, dob, start, OBD_OBJECT_EOF, th,
ofd_object_capa(env, fo));
if (rc)
GOTO(stop, rc);
rc = dt_attr_set(env, dob, la, th, ofd_object_capa(env, fo));
if (rc)
GOTO(stop, rc);
if (ff_needed) {
rc = dt_xattr_set(env, ofd_object_child(fo), &info->fti_buf,
XATTR_NAME_FID, 0, th, BYPASS_CAPA);
if (rc == 0) {
fo->ofo_pfid.f_seq = le64_to_cpu(ff->ff_parent.f_seq);
fo->ofo_pfid.f_oid = le32_to_cpu(ff->ff_parent.f_oid);
/* Currently, the filter_fid::ff_parent::f_ver is not
* the real parent MDT-object's FID::f_ver, instead it
* is the OST-object index in its parent MDT-object's
* layout EA. */
fo->ofo_pfid.f_stripe_idx =
le32_to_cpu(ff->ff_parent.f_stripe_idx);
}
}
GOTO(stop, rc);
stop:
ofd_trans_stop(env, ofd, th, rc);
unlock:
ofd_write_unlock(env, fo);
return rc;
}
int ofd_attr_set(const struct lu_env *env, struct ofd_object *fo,
struct lu_attr *la, struct filter_fid *ff)
{
struct ofd_thread_info *info = ofd_info(env);
struct ofd_device *ofd = ofd_obj2dev(fo);
struct thandle *th;
struct ofd_mod_data *fmd;
int ff_needed = 0;
int rc;
ENTRY;
ofd_write_lock(env, fo);
if (!ofd_object_exists(fo))
GOTO(unlock, rc = -ENOENT);
if (la->la_valid & (LA_ATIME | LA_MTIME | LA_CTIME)) {
fmd = ofd_fmd_get(info->fti_exp, &fo->ofo_header.loh_fid);
if (fmd && fmd->fmd_mactime_xid < info->fti_xid)
fmd->fmd_mactime_xid = info->fti_xid;
ofd_fmd_put(info->fti_exp, fmd);
}
/* VBR: version recovery check */
rc = ofd_version_get_check(info, fo);
if (rc)
GOTO(unlock, rc);
rc = ofd_attr_handle_ugid(env, fo, la, 1 /* is_setattr */);
if (rc != 0)
GOTO(unlock, rc);
if (ff != NULL) {
rc = ofd_object_ff_load(env, fo);
if (rc == -ENODATA)
ff_needed = 1;
else if (rc < 0)
GOTO(unlock, rc);
}
th = ofd_trans_create(env, ofd);
if (IS_ERR(th))
GOTO(unlock, rc = PTR_ERR(th));
rc = dt_declare_attr_set(env, ofd_object_child(fo), la, th);
if (rc)
GOTO(stop, rc);
if (ff_needed) {
info->fti_buf.lb_buf = ff;
info->fti_buf.lb_len = sizeof(*ff);
rc = dt_declare_xattr_set(env, ofd_object_child(fo),
&info->fti_buf, XATTR_NAME_FID, 0,
th);
if (rc)
GOTO(stop, rc);
}
rc = ofd_trans_start(env, ofd, la->la_valid & LA_SIZE ? fo : NULL, th);
if (rc)
GOTO(stop, rc);
rc = dt_attr_set(env, ofd_object_child(fo), la, th,
ofd_object_capa(env, fo));
if (rc)
GOTO(stop, rc);
if (ff_needed) {
rc = dt_xattr_set(env, ofd_object_child(fo), &info->fti_buf,
XATTR_NAME_FID, 0, th, BYPASS_CAPA);
if (rc == 0) {
fo->ofo_pfid.f_seq = le64_to_cpu(ff->ff_parent.f_seq);
fo->ofo_pfid.f_oid = le32_to_cpu(ff->ff_parent.f_oid);
/* Currently, the filter_fid::ff_parent::f_ver is not
* the real parent MDT-object's FID::f_ver, instead it
* is the OST-object index in its parent MDT-object's
* layout EA. */
fo->ofo_pfid.f_stripe_idx =
le32_to_cpu(ff->ff_parent.f_stripe_idx);
}
}
GOTO(stop, rc);
stop:
ofd_trans_stop(env, ofd, th, rc);
unlock:
ofd_write_unlock(env, fo);
return rc;
}
int ofd_object_punch(const struct lu_env *env, struct ofd_object *fo,
__u64 start, __u64 end, struct lu_attr *la,
struct filter_fid *ff)
{
struct ofd_thread_info *info = ofd_info(env);
struct ofd_device *ofd = ofd_obj2dev(fo);
struct ofd_mod_data *fmd;
struct dt_object *dob = ofd_object_child(fo);
struct thandle *th;
int ff_needed = 0;
int rc;
ENTRY;
/* we support truncate, not punch yet */
LASSERT(end == OBD_OBJECT_EOF);
fmd = ofd_fmd_get(info->fti_exp, &fo->ofo_header.loh_fid);
if (fmd && fmd->fmd_mactime_xid < info->fti_xid)
fmd->fmd_mactime_xid = info->fti_xid;
ofd_fmd_put(info->fti_exp, fmd);
ofd_write_lock(env, fo);
if (!ofd_object_exists(fo))
GOTO(unlock, rc = -ENOENT);
/* VBR: version recovery check */
rc = ofd_version_get_check(info, fo);
if (rc)
GOTO(unlock, rc);
rc = ofd_attr_handle_ugid(env, fo, la, 0 /* !is_setattr */);
if (rc != 0)
GOTO(unlock, rc);
if (ff != NULL) {
rc = ofd_object_ff_check(env, fo);
if (rc == -ENODATA)
ff_needed = 1;
else if (rc < 0)
GOTO(unlock, rc);
}
th = ofd_trans_create(env, ofd);
if (IS_ERR(th))
GOTO(unlock, rc = PTR_ERR(th));
rc = dt_declare_attr_set(env, dob, la, th);
if (rc)
GOTO(stop, rc);
rc = dt_declare_punch(env, dob, start, OBD_OBJECT_EOF, th);
if (rc)
GOTO(stop, rc);
if (ff_needed) {
info->fti_buf.lb_buf = ff;
info->fti_buf.lb_len = sizeof(*ff);
rc = dt_declare_xattr_set(env, ofd_object_child(fo),
&info->fti_buf, XATTR_NAME_FID, 0,
th);
if (rc)
GOTO(stop, rc);
}
rc = ofd_trans_start(env, ofd, fo, th);
if (rc)
GOTO(stop, rc);
rc = dt_punch(env, dob, start, OBD_OBJECT_EOF, th,
ofd_object_capa(env, fo));
if (rc)
GOTO(stop, rc);
rc = dt_attr_set(env, dob, la, th, ofd_object_capa(env, fo));
if (rc)
GOTO(stop, rc);
if (ff_needed)
rc = dt_xattr_set(env, ofd_object_child(fo), &info->fti_buf,
XATTR_NAME_FID, 0, th, BYPASS_CAPA);
stop:
ofd_trans_stop(env, ofd, th, rc);
unlock:
ofd_write_unlock(env, fo);
RETURN(rc);
}
int ofd_attr_set(const struct lu_env *env, struct ofd_object *fo,
struct lu_attr *la, struct filter_fid *ff)
{
struct ofd_thread_info *info = ofd_info(env);
struct ofd_device *ofd = ofd_obj2dev(fo);
struct thandle *th;
struct ofd_mod_data *fmd;
int ff_needed = 0;
int rc;
ENTRY;
ofd_write_lock(env, fo);
if (!ofd_object_exists(fo))
GOTO(unlock, rc = -ENOENT);
if (la->la_valid & (LA_ATIME | LA_MTIME | LA_CTIME)) {
fmd = ofd_fmd_get(info->fti_exp, &fo->ofo_header.loh_fid);
if (fmd && fmd->fmd_mactime_xid < info->fti_xid)
fmd->fmd_mactime_xid = info->fti_xid;
ofd_fmd_put(info->fti_exp, fmd);
}
/* VBR: version recovery check */
rc = ofd_version_get_check(info, fo);
if (rc)
GOTO(unlock, rc);
rc = ofd_attr_handle_ugid(env, fo, la, 1 /* is_setattr */);
if (rc != 0)
GOTO(unlock, rc);
if (ff != NULL) {
rc = ofd_object_ff_check(env, fo);
if (rc == -ENODATA)
ff_needed = 1;
else if (rc < 0)
GOTO(unlock, rc);
}
th = ofd_trans_create(env, ofd);
if (IS_ERR(th))
GOTO(unlock, rc = PTR_ERR(th));
rc = dt_declare_attr_set(env, ofd_object_child(fo), la, th);
if (rc)
GOTO(stop, rc);
if (ff_needed) {
info->fti_buf.lb_buf = ff;
info->fti_buf.lb_len = sizeof(*ff);
rc = dt_declare_xattr_set(env, ofd_object_child(fo),
&info->fti_buf, XATTR_NAME_FID, 0,
th);
if (rc)
GOTO(stop, rc);
}
rc = ofd_trans_start(env, ofd, la->la_valid & LA_SIZE ? fo : NULL, th);
if (rc)
GOTO(stop, rc);
rc = dt_attr_set(env, ofd_object_child(fo), la, th,
ofd_object_capa(env, fo));
if (rc)
GOTO(stop, rc);
if (ff_needed)
rc = dt_xattr_set(env, ofd_object_child(fo), &info->fti_buf,
XATTR_NAME_FID, 0, th, BYPASS_CAPA);
stop:
ofd_trans_stop(env, ofd, th, rc);
unlock:
ofd_write_unlock(env, fo);
RETURN(rc);
}
static int
ofd_write_attr_set(const struct lu_env *env, struct ofd_device *ofd,
struct ofd_object *ofd_obj, struct lu_attr *la,
struct filter_fid *ff)
{
struct ofd_thread_info *info = ofd_info(env);
__u64 valid = la->la_valid;
int rc;
struct thandle *th;
struct dt_object *dt_obj;
int ff_needed = 0;
ENTRY;
LASSERT(la);
dt_obj = ofd_object_child(ofd_obj);
LASSERT(dt_obj != NULL);
la->la_valid &= LA_UID | LA_GID;
rc = ofd_attr_handle_ugid(env, ofd_obj, la, 0 /* !is_setattr */);
if (rc != 0)
GOTO(out, rc);
if (ff != NULL) {
rc = ofd_object_ff_check(env, ofd_obj);
if (rc == -ENODATA)
ff_needed = 1;
else if (rc < 0)
GOTO(out, rc);
}
if (!la->la_valid && !ff_needed)
/* no attributes to set */
GOTO(out, rc = 0);
th = ofd_trans_create(env, ofd);
if (IS_ERR(th))
GOTO(out, rc = PTR_ERR(th));
if (la->la_valid) {
rc = dt_declare_attr_set(env, dt_obj, la, th);
if (rc)
GOTO(out_tx, rc);
}
if (ff_needed) {
info->fti_buf.lb_buf = ff;
info->fti_buf.lb_len = sizeof(*ff);
rc = dt_declare_xattr_set(env, dt_obj, &info->fti_buf,
XATTR_NAME_FID, 0, th);
if (rc)
GOTO(out_tx, rc);
}
/* We don't need a transno for this operation which will be re-executed
* anyway when the OST_WRITE (with a transno assigned) is replayed */
rc = dt_trans_start_local(env, ofd->ofd_osd , th);
if (rc)
GOTO(out_tx, rc);
/* set uid/gid */
if (la->la_valid) {
rc = dt_attr_set(env, dt_obj, la, th,
ofd_object_capa(env, ofd_obj));
if (rc)
GOTO(out_tx, rc);
}
/* set filter fid EA */
if (ff_needed) {
rc = dt_xattr_set(env, dt_obj, &info->fti_buf, XATTR_NAME_FID,
0, th, BYPASS_CAPA);
if (rc)
GOTO(out_tx, rc);
}
EXIT;
out_tx:
dt_trans_stop(env, ofd->ofd_osd, th);
out:
la->la_valid = valid;
return rc;
}
static int ofd_preprw_write(const struct lu_env *env, struct obd_export *exp,
struct ofd_device *ofd, struct lu_fid *fid,
struct lu_attr *la, struct obdo *oa,
int objcount, struct obd_ioobj *obj,
struct niobuf_remote *rnb, int *nr_local,
struct niobuf_local *lnb, char *jobid)
{
struct ofd_object *fo;
int i, j, k, rc = 0, tot_bytes = 0;
ENTRY;
LASSERT(env != NULL);
LASSERT(objcount == 1);
if (unlikely(exp->exp_obd->obd_recovering)) {
struct ofd_thread_info *info = ofd_info(env);
/* copied from ofd_precreate_object */
/* XXX this should be consolidated to use the same code
* instead of a copy, due to the ongoing risk of bugs. */
memset(&info->fti_attr, 0, sizeof(info->fti_attr));
info->fti_attr.la_valid = LA_TYPE | LA_MODE;
info->fti_attr.la_mode = S_IFREG | S_ISUID | S_ISGID | 0666;
info->fti_attr.la_valid |= LA_ATIME | LA_MTIME | LA_CTIME;
/* Initialize a/c/m time so any client timestamp will always
* be newer and update the inode. ctime = 0 is also handled
* specially in osd_inode_setattr(). See LU-221, LU-1042 */
info->fti_attr.la_atime = 0;
info->fti_attr.la_mtime = 0;
info->fti_attr.la_ctime = 0;
fo = ofd_object_find_or_create(env, ofd, fid, &info->fti_attr);
} else {
fo = ofd_object_find(env, ofd, fid);
}
if (IS_ERR(fo))
GOTO(out, rc = PTR_ERR(fo));
LASSERT(fo != NULL);
ofd_read_lock(env, fo);
if (!ofd_object_exists(fo)) {
CERROR("%s: BRW to missing obj "DOSTID"\n",
exp->exp_obd->obd_name, POSTID(&obj->ioo_oid));
ofd_read_unlock(env, fo);
ofd_object_put(env, fo);
GOTO(out, rc = -ENOENT);
}
/* Process incoming grant info, set OBD_BRW_GRANTED flag and grant some
* space back if possible */
ofd_grant_prepare_write(env, exp, oa, rnb, obj->ioo_bufcnt);
/* parse remote buffers to local buffers and prepare the latter */
*nr_local = 0;
for (i = 0, j = 0; i < obj->ioo_bufcnt; i++) {
rc = dt_bufs_get(env, ofd_object_child(fo),
rnb + i, lnb + j, 1,
ofd_object_capa(env, fo));
if (unlikely(rc < 0))
GOTO(err, rc);
LASSERT(rc <= PTLRPC_MAX_BRW_PAGES);
/* correct index for local buffers to continue with */
for (k = 0; k < rc; k++) {
lnb[j+k].lnb_flags = rnb[i].rnb_flags;
if (!(rnb[i].rnb_flags & OBD_BRW_GRANTED))
lnb[j+k].lnb_rc = -ENOSPC;
/* remote client can't break through quota */
if (exp_connect_rmtclient(exp))
lnb[j+k].lnb_flags &= ~OBD_BRW_NOQUOTA;
}
j += rc;
*nr_local += rc;
LASSERT(j <= PTLRPC_MAX_BRW_PAGES);
tot_bytes += rnb[i].rnb_len;
}
LASSERT(*nr_local > 0 && *nr_local <= PTLRPC_MAX_BRW_PAGES);
rc = dt_write_prep(env, ofd_object_child(fo), lnb, *nr_local);
if (unlikely(rc != 0))
GOTO(err, rc);
ofd_counter_incr(exp, LPROC_OFD_STATS_WRITE, jobid, tot_bytes);
RETURN(0);
err:
dt_bufs_put(env, ofd_object_child(fo), lnb, *nr_local);
ofd_read_unlock(env, fo);
/* ofd_grant_prepare_write() was called, so we must commit */
ofd_grant_commit(env, exp, rc);
out:
/* let's still process incoming grant information packed in the oa,
* but without enforcing grant since we won't proceed with the write.
* Just like a read request actually. */
ofd_grant_prepare_read(env, exp, oa);
return rc;
}
