/* Add log records for each OSC that this object is striped over, and return
* cookies for each one. We _would_ have nice abstraction here, except that
* we need to keep cookies in stripe order, even if some are NULL, so that
* the right cookies are passed back to the right OSTs at the client side.
* Unset cookies should be all-zero (which will never occur naturally). */
static int lov_llog_origin_add(const struct lu_env *env,
struct llog_ctxt *ctxt,
struct llog_rec_hdr *rec,
struct lov_stripe_md *lsm,
struct llog_cookie *logcookies, int numcookies)
{
struct obd_device *obd = ctxt->loc_obd;
struct lov_obd *lov = &obd->u.lov;
int i, rc = 0, cookies = 0;
ENTRY;
LASSERTF(logcookies && numcookies >= lsm->lsm_stripe_count,
"logcookies %p, numcookies %d lsm->lsm_stripe_count %d \n",
logcookies, numcookies, lsm->lsm_stripe_count);
for (i = 0; i < lsm->lsm_stripe_count; i++) {
struct lov_oinfo *loi = lsm->lsm_oinfo[i];
struct obd_device *child =
lov->lov_tgts[loi->loi_ost_idx]->ltd_exp->exp_obd;
struct llog_ctxt *cctxt = llog_get_context(child, ctxt->loc_idx);
/* fill mds unlink/setattr log record */
switch (rec->lrh_type) {
case MDS_UNLINK_REC: {
struct llog_unlink_rec *lur = (struct llog_unlink_rec *)rec;
lur->lur_oid = ostid_id(&loi->loi_oi);
lur->lur_oseq = (__u32)ostid_seq(&loi->loi_oi);
break;
}
case MDS_SETATTR64_REC: {
struct llog_setattr64_rec *lsr = (struct llog_setattr64_rec *)rec;
lsr->lsr_oi = loi->loi_oi;
break;
}
default:
break;
}
/* inject error in llog_obd_add() below */
if (OBD_FAIL_CHECK(OBD_FAIL_MDS_FAIL_LOV_LOG_ADD)) {
llog_ctxt_put(cctxt);
cctxt = NULL;
}
rc = llog_obd_add(env, cctxt, rec, NULL, logcookies + cookies,
numcookies - cookies);
llog_ctxt_put(cctxt);
if (rc < 0) {
CERROR("Can't add llog (rc = %d) for stripe %d\n",
rc, cookies);
memset(logcookies + cookies, 0,
sizeof(struct llog_cookie));
rc = 1; /* skip this cookie */
}
/* Note that rc is always 1 if llog_obd_add was successful */
cookies += rc;
}
RETURN(cookies);
}
/**
* Prepare bulk IO requests for processing.
*
* This function does initial checks of IO and calls corresponding
* functions for read/write processing.
*
* \param[in] env execution environment
* \param[in] cmd IO type (read/write)
* \param[in] exp OBD export of client
* \param[in] oa OBDO structure from request
* \param[in] objcount always 1
* \param[in] obj object data
* \param[in] rnb remote buffers
* \param[in] nr_local number of local buffers
* \param[in] lnb local buffers
*
* \retval 0 on successful prepare
* \retval negative value on error
*/
int ofd_preprw(const struct lu_env *env, int cmd, struct obd_export *exp,
struct obdo *oa, int objcount, struct obd_ioobj *obj,
struct niobuf_remote *rnb, int *nr_local,
struct niobuf_local *lnb)
{
struct tgt_session_info *tsi = tgt_ses_info(env);
struct ofd_device *ofd = ofd_exp(exp);
struct ofd_thread_info *info;
char *jobid;
const struct lu_fid *fid = &oa->o_oi.oi_fid;
int rc = 0;
if (*nr_local > PTLRPC_MAX_BRW_PAGES) {
CERROR("%s: bulk has too many pages %d, which exceeds the"
"maximum pages per RPC of %d\n",
exp->exp_obd->obd_name, *nr_local, PTLRPC_MAX_BRW_PAGES);
RETURN(-EPROTO);
}
if (tgt_ses_req(tsi) == NULL) { /* echo client case */
info = ofd_info_init(env, exp);
jobid = NULL;
} else {
info = tsi2ofd_info(tsi);
jobid = tsi->tsi_jobid;
}
LASSERT(oa != NULL);
if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT)) {
struct ofd_seq *oseq;
oseq = ofd_seq_load(env, ofd, ostid_seq(&oa->o_oi));
if (IS_ERR(oseq)) {
CERROR("%s: Can not find seq for "DOSTID
": rc = %ld\n", ofd_name(ofd), POSTID(&oa->o_oi),
PTR_ERR(oseq));
RETURN(-EINVAL);
}
if (oseq->os_destroys_in_progress == 0) {
/* don't fail lookups for orphan recovery, it causes
* later LBUGs when objects still exist during
* precreate */
ofd_seq_put(env, oseq);
RETURN(-ENOENT);
}
ofd_seq_put(env, oseq);
}
LASSERT(objcount == 1);
LASSERT(obj->ioo_bufcnt > 0);
if (cmd == OBD_BRW_WRITE) {
la_from_obdo(&info->fti_attr, oa, OBD_MD_FLGETATTR);
rc = ofd_preprw_write(env, exp, ofd, fid, &info->fti_attr, oa,
objcount, obj, rnb, nr_local, lnb, jobid);
} else if (cmd == OBD_BRW_READ) {
ofd_grant_prepare_read(env, exp, oa);
rc = ofd_preprw_read(env, exp, ofd, fid, &info->fti_attr, oa,
obj->ioo_bufcnt, rnb, nr_local, lnb,
jobid);
obdo_from_la(oa, &info->fti_attr, LA_ATIME);
} else {
CERROR("%s: wrong cmd %d received!\n",
exp->exp_obd->obd_name, cmd);
rc = -EPROTO;
}
RETURN(rc);
}
/**
* Implementation of struct cl_object_operations::coo_req_attr_set() for osc
* layer. osc is responsible for struct obdo::o_id and struct obdo::o_seq
* fields.
*/
static void osc_req_attr_set(const struct lu_env *env, struct cl_object *obj,
struct cl_req_attr *attr)
{
struct lov_oinfo *oinfo;
struct obdo *oa;
struct ost_lvb *lvb;
u64 flags = attr->cra_flags;
oinfo = cl2osc(obj)->oo_oinfo;
lvb = &oinfo->loi_lvb;
oa = attr->cra_oa;
if ((flags & OBD_MD_FLMTIME) != 0) {
oa->o_mtime = lvb->lvb_mtime;
oa->o_valid |= OBD_MD_FLMTIME;
}
if ((flags & OBD_MD_FLATIME) != 0) {
oa->o_atime = lvb->lvb_atime;
oa->o_valid |= OBD_MD_FLATIME;
}
if ((flags & OBD_MD_FLCTIME) != 0) {
oa->o_ctime = lvb->lvb_ctime;
oa->o_valid |= OBD_MD_FLCTIME;
}
if (flags & OBD_MD_FLGROUP) {
ostid_set_seq(&oa->o_oi, ostid_seq(&oinfo->loi_oi));
oa->o_valid |= OBD_MD_FLGROUP;
}
if (flags & OBD_MD_FLID) {
ostid_set_id(&oa->o_oi, ostid_id(&oinfo->loi_oi));
oa->o_valid |= OBD_MD_FLID;
}
if (flags & OBD_MD_FLHANDLE) {
struct ldlm_lock *lock;
struct osc_page *opg;
opg = osc_cl_page_osc(attr->cra_page, cl2osc(obj));
lock = osc_dlmlock_at_pgoff(env, cl2osc(obj), osc_index(opg),
OSC_DAP_FL_TEST_LOCK | OSC_DAP_FL_CANCELING);
if (lock == NULL && !opg->ops_srvlock) {
struct ldlm_resource *res;
struct ldlm_res_id *resname;
CL_PAGE_DEBUG(D_ERROR, env, attr->cra_page,
"uncovered page!\n");
resname = &osc_env_info(env)->oti_resname;
ostid_build_res_name(&oinfo->loi_oi, resname);
res = ldlm_resource_get(
osc_export(cl2osc(obj))->exp_obd->obd_namespace,
NULL, resname, LDLM_EXTENT, 0);
ldlm_resource_dump(D_ERROR, res);
libcfs_debug_dumpstack(NULL);
LBUG();
}
/* check for lockless io. */
if (lock != NULL) {
oa->o_handle = lock->l_remote_handle;
oa->o_valid |= OBD_MD_FLHANDLE;
LDLM_LOCK_PUT(lock);
}
}
}
/**
* Implementation of struct cl_req_operations::cro_attr_set() for osc
* layer. osc is responsible for struct obdo::o_id and struct obdo::o_seq
* fields.
*/
static void osc_req_attr_set(const struct lu_env *env,
const struct cl_req_slice *slice,
const struct cl_object *obj,
struct cl_req_attr *attr, u64 flags)
{
struct lov_oinfo *oinfo;
struct cl_req *clerq;
struct cl_page *apage; /* _some_ page in @clerq */
struct cl_lock *lock; /* _some_ lock protecting @apage */
struct osc_lock *olck;
struct osc_page *opg;
struct obdo *oa;
struct ost_lvb *lvb;
oinfo = cl2osc(obj)->oo_oinfo;
lvb = &oinfo->loi_lvb;
oa = attr->cra_oa;
if ((flags & OBD_MD_FLMTIME) != 0) {
oa->o_mtime = lvb->lvb_mtime;
oa->o_valid |= OBD_MD_FLMTIME;
}
if ((flags & OBD_MD_FLATIME) != 0) {
oa->o_atime = lvb->lvb_atime;
oa->o_valid |= OBD_MD_FLATIME;
}
if ((flags & OBD_MD_FLCTIME) != 0) {
oa->o_ctime = lvb->lvb_ctime;
oa->o_valid |= OBD_MD_FLCTIME;
}
if (flags & OBD_MD_FLGROUP) {
ostid_set_seq(&oa->o_oi, ostid_seq(&oinfo->loi_oi));
oa->o_valid |= OBD_MD_FLGROUP;
}
if (flags & OBD_MD_FLID) {
ostid_set_id(&oa->o_oi, ostid_id(&oinfo->loi_oi));
oa->o_valid |= OBD_MD_FLID;
}
if (flags & OBD_MD_FLHANDLE) {
clerq = slice->crs_req;
LASSERT(!list_empty(&clerq->crq_pages));
apage = container_of(clerq->crq_pages.next,
struct cl_page, cp_flight);
opg = osc_cl_page_osc(apage);
apage = opg->ops_cl.cpl_page; /* now apage is a sub-page */
lock = cl_lock_at_page(env, apage->cp_obj, apage, NULL, 1, 1);
if (lock == NULL) {
struct cl_object_header *head;
struct cl_lock *scan;
head = cl_object_header(apage->cp_obj);
list_for_each_entry(scan, &head->coh_locks,
cll_linkage)
CL_LOCK_DEBUG(D_ERROR, env, scan,
"no cover page!\n");
CL_PAGE_DEBUG(D_ERROR, env, apage,
"dump uncover page!\n");
dump_stack();
LBUG();
}
olck = osc_lock_at(lock);
LASSERT(olck != NULL);
LASSERT(ergo(opg->ops_srvlock, olck->ols_lock == NULL));
/* check for lockless io. */
if (olck->ols_lock != NULL) {
oa->o_handle = olck->ols_lock->l_remote_handle;
oa->o_valid |= OBD_MD_FLHANDLE;
}
cl_lock_put(env, lock);
}
int ofd_precreate_objects(const struct lu_env *env, struct ofd_device *ofd,
obd_id id, struct ofd_seq *oseq, int nr, int sync)
{
struct ofd_thread_info *info = ofd_info(env);
struct ofd_object *fo = NULL;
struct dt_object *next;
struct thandle *th;
struct ofd_object **batch;
struct lu_fid *fid = &info->fti_fid;
obd_id tmp;
int rc;
int i;
int objects = 0;
int nr_saved = nr;
ENTRY;
/* Don't create objects beyond the valid range for this SEQ */
if (unlikely(fid_seq_is_mdt0(ostid_seq(&oseq->os_oi)) &&
(id + nr) >= IDIF_MAX_OID)) {
CERROR("%s:"DOSTID" hit the IDIF_MAX_OID (1<<48)!\n",
ofd_name(ofd), id, ostid_seq(&oseq->os_oi));
RETURN(rc = -ENOSPC);
} else if (unlikely(!fid_seq_is_mdt0(ostid_seq(&oseq->os_oi)) &&
(id + nr) >= OBIF_MAX_OID)) {
CERROR("%s:"DOSTID" hit the OBIF_MAX_OID (1<<32)!\n",
ofd_name(ofd), id, ostid_seq(&oseq->os_oi));
RETURN(rc = -ENOSPC);
}
OBD_ALLOC(batch, nr_saved * sizeof(struct ofd_object *));
if (batch == NULL)
RETURN(-ENOMEM);
info->fti_attr.la_valid = LA_TYPE | LA_MODE;
/*
* We mark object SUID+SGID to flag it for accepting UID+GID from
* client on first write. Currently the permission bits on the OST are
* never used, so this is OK.
*/
info->fti_attr.la_mode = S_IFREG | S_ISUID | S_ISGID | 0666;
info->fti_dof.dof_type = dt_mode_to_dft(S_IFREG);
/* 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_valid |= LA_ATIME | LA_MTIME | LA_CTIME;
info->fti_attr.la_atime = 0;
info->fti_attr.la_mtime = 0;
info->fti_attr.la_ctime = 0;
LASSERT(id != 0);
/* prepare objects */
*fid = *lu_object_fid(&oseq->os_lastid_obj->do_lu);
for (i = 0; i < nr; i++) {
rc = fid_set_id(fid, id + i);
if (rc != 0) {
if (i == 0)
GOTO(out, rc);
nr = i;
break;
}
fo = ofd_object_find(env, ofd, fid);
if (IS_ERR(fo)) {
if (i == 0)
GOTO(out, rc = PTR_ERR(fo));
nr = i;
break;
}
ofd_write_lock(env, fo);
batch[i] = fo;
}
info->fti_buf.lb_buf = &tmp;
info->fti_buf.lb_len = sizeof(tmp);
info->fti_off = 0;
th = ofd_trans_create(env, ofd);
if (IS_ERR(th))
GOTO(out, rc = PTR_ERR(th));
th->th_sync |= sync;
rc = dt_declare_record_write(env, oseq->os_lastid_obj, &info->fti_buf,
info->fti_off, th);
if (rc)
GOTO(trans_stop, rc);
for (i = 0; i < nr; i++) {
fo = batch[i];
LASSERT(fo);
if (unlikely(ofd_object_exists(fo))) {
/* object may exist being re-created by write replay */
CDEBUG(D_INODE, "object "LPX64"/"LPX64" exists: "
DFID"\n", ostid_seq(&oseq->os_oi), id,
PFID(lu_object_fid(&fo->ofo_obj.do_lu)));
continue;
}
next = ofd_object_child(fo);
LASSERT(next != NULL);
rc = dt_declare_create(env, next, &info->fti_attr, NULL,
&info->fti_dof, th);
if (rc) {
nr = i;
break;
}
}
rc = dt_trans_start_local(env, ofd->ofd_osd, th);
if (rc)
GOTO(trans_stop, rc);
CDEBUG(D_OTHER, "%s: create new object "DFID" nr %d\n",
ofd_name(ofd), PFID(fid), nr);
LASSERT(nr > 0);
/* When the LFSCK scanning the whole device to verify the LAST_ID file
* consistency, it will load the last_id into RAM firstly, and compare
* the last_id with each OST-object's ID. If the later one is larger,
* then it will regard the LAST_ID file crashed. But during the LFSCK
* scanning, the OFD may continue to create new OST-objects. Those new
* created OST-objects will have larger IDs than the LFSCK known ones.
* So from the LFSCK view, it needs to re-load the last_id from disk
* file, and if the latest last_id is still smaller than the object's
* ID, then the LAST_ID file is real crashed.
*
* To make above mechanism to work, before OFD pre-create OST-objects,
* it needs to update the LAST_ID file firstly, otherwise, the LFSCK
* may cannot get latest last_id although new OST-object created. */
if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_SKIP_LASTID)) {
tmp = cpu_to_le64(id + nr - 1);
dt_write_lock(env, oseq->os_lastid_obj, 0);
rc = dt_record_write(env, oseq->os_lastid_obj,
&info->fti_buf, &info->fti_off, th);
dt_write_unlock(env, oseq->os_lastid_obj);
if (rc != 0)
GOTO(trans_stop, rc);
}
for (i = 0; i < nr; i++) {
fo = batch[i];
LASSERT(fo);
/* Only the new created objects need to be recorded. */
if (ofd->ofd_osd->dd_record_fid_accessed) {
lfsck_pack_rfa(&ofd_info(env)->fti_lr,
lu_object_fid(&fo->ofo_obj.do_lu));
lfsck_in_notify(env, ofd->ofd_osd,
&ofd_info(env)->fti_lr);
}
if (likely(!ofd_object_exists(fo) &&
!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING))) {
next = ofd_object_child(fo);
LASSERT(next != NULL);
rc = dt_create(env, next, &info->fti_attr, NULL,
&info->fti_dof, th);
if (rc)
break;
LASSERT(ofd_object_exists(fo));
}
ofd_seq_last_oid_set(oseq, id + i);
}
objects = i;
/* NOT all the wanted objects have been created,
* set the LAST_ID as the real created. */
if (unlikely(objects < nr)) {
int rc1;
info->fti_off = 0;
tmp = cpu_to_le64(ofd_seq_last_oid(oseq));
dt_write_lock(env, oseq->os_lastid_obj, 0);
rc1 = dt_record_write(env, oseq->os_lastid_obj,
&info->fti_buf, &info->fti_off, th);
dt_write_unlock(env, oseq->os_lastid_obj);
if (rc1 != 0)
CERROR("%s: fail to reset the LAST_ID for seq ("LPX64
") from "LPU64" to "LPU64"\n", ofd_name(ofd),
ostid_seq(&oseq->os_oi), id + nr - 1,
ofd_seq_last_oid(oseq));
}
trans_stop:
ofd_trans_stop(env, ofd, th, rc);
out:
for (i = 0; i < nr_saved; i++) {
fo = batch[i];
if (fo) {
ofd_write_unlock(env, fo);
ofd_object_put(env, fo);
}
}
OBD_FREE(batch, nr_saved * sizeof(struct ofd_object *));
CDEBUG((objects == 0 && rc == 0) ? D_ERROR : D_OTHER,
"created %d/%d objects: %d\n", objects, nr_saved, rc);
LASSERT(ergo(objects == 0, rc < 0));
RETURN(objects > 0 ? objects : rc);
}
int ofd_precreate_objects(const struct lu_env *env, struct ofd_device *ofd,
obd_id id, struct ofd_seq *oseq, int nr, int sync)
{
struct ofd_thread_info *info = ofd_info(env);
struct ofd_object *fo = NULL;
struct dt_object *next;
struct thandle *th;
struct ofd_object **batch;
obd_id tmp;
int rc;
int i;
int objects = 0;
int nr_saved = nr;
ENTRY;
/* Don't create objects beyond the valid range for this SEQ */
if (unlikely(fid_seq_is_mdt0(ostid_seq(&oseq->os_oi)) &&
(id + nr) >= IDIF_MAX_OID)) {
CERROR("%s:"DOSTID" hit the IDIF_MAX_OID (1<<48)!\n",
ofd_name(ofd), id, ostid_seq(&oseq->os_oi));
RETURN(rc = -ENOSPC);
} else if (unlikely(!fid_seq_is_mdt0(ostid_seq(&oseq->os_oi)) &&
(id + nr) >= OBIF_MAX_OID)) {
CERROR("%s:"DOSTID" hit the OBIF_MAX_OID (1<<32)!\n",
ofd_name(ofd), id, ostid_seq(&oseq->os_oi));
RETURN(rc = -ENOSPC);
}
OBD_ALLOC(batch, nr_saved * sizeof(struct ofd_object *));
if (batch == NULL)
RETURN(-ENOMEM);
info->fti_attr.la_valid = LA_TYPE | LA_MODE;
/*
* We mark object SUID+SGID to flag it for accepting UID+GID from
* client on first write. Currently the permission bits on the OST are
* never used, so this is OK.
*/
info->fti_attr.la_mode = S_IFREG | S_ISUID | S_ISGID | 0666;
info->fti_dof.dof_type = dt_mode_to_dft(S_IFREG);
/* 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_valid |= LA_ATIME | LA_MTIME | LA_CTIME;
info->fti_attr.la_atime = 0;
info->fti_attr.la_mtime = 0;
info->fti_attr.la_ctime = 0;
/* prepare objects */
ostid_set_seq(&info->fti_ostid, ostid_seq(&oseq->os_oi));
for (i = 0; i < nr; i++) {
ostid_set_id(&info->fti_ostid, id + i);
rc = ostid_to_fid(&info->fti_fid, &info->fti_ostid, 0);
if (rc) {
if (i == 0)
GOTO(out, rc);
nr = i;
break;
}
fo = ofd_object_find(env, ofd, &info->fti_fid);
if (IS_ERR(fo)) {
if (i == 0)
GOTO(out, rc = PTR_ERR(fo));
nr = i;
break;
}
ofd_write_lock(env, fo);
batch[i] = fo;
}
info->fti_buf.lb_buf = &tmp;
info->fti_buf.lb_len = sizeof(tmp);
info->fti_off = 0;
th = ofd_trans_create(env, ofd);
if (IS_ERR(th))
GOTO(out, rc = PTR_ERR(th));
th->th_sync |= sync;
rc = dt_declare_record_write(env, oseq->os_lastid_obj, sizeof(tmp),
info->fti_off, th);
if (rc)
GOTO(trans_stop, rc);
for (i = 0; i < nr; i++) {
fo = batch[i];
LASSERT(fo);
if (unlikely(ofd_object_exists(fo))) {
/* object may exist being re-created by write replay */
CDEBUG(D_INODE, "object "LPX64"/"LPX64" exists: "
DFID"\n", ostid_seq(&oseq->os_oi), id,
PFID(&info->fti_fid));
continue;
}
next = ofd_object_child(fo);
LASSERT(next != NULL);
rc = dt_declare_create(env, next, &info->fti_attr, NULL,
&info->fti_dof, th);
if (rc) {
nr = i;
break;
}
}
rc = dt_trans_start_local(env, ofd->ofd_osd, th);
if (rc)
GOTO(trans_stop, rc);
CDEBUG(D_OTHER, "%s: create new object "DFID" nr %d\n",
ofd_name(ofd), PFID(&info->fti_fid), nr);
for (i = 0; i < nr; i++) {
fo = batch[i];
LASSERT(fo);
if (likely(!ofd_object_exists(fo))) {
next = ofd_object_child(fo);
LASSERT(next != NULL);
rc = dt_create(env, next, &info->fti_attr, NULL,
&info->fti_dof, th);
if (rc)
break;
LASSERT(ofd_object_exists(fo));
}
ofd_seq_last_oid_set(oseq, id + i);
}
objects = i;
if (objects > 0) {
tmp = cpu_to_le64(ofd_seq_last_oid(oseq));
rc = dt_record_write(env, oseq->os_lastid_obj,
&info->fti_buf, &info->fti_off, th);
}
trans_stop:
ofd_trans_stop(env, ofd, th, rc);
out:
for (i = 0; i < nr_saved; i++) {
fo = batch[i];
if (fo) {
ofd_write_unlock(env, fo);
ofd_object_put(env, fo);
}
}
OBD_FREE(batch, nr_saved * sizeof(struct ofd_object *));
CDEBUG((objects == 0 && rc == 0) ? D_ERROR : D_OTHER,
"created %d/%d objects: %d\n", objects, nr_saved, rc);
LASSERT(ergo(objects == 0, rc < 0));
RETURN(objects > 0 ? objects : rc);
}
