/**
* Implementation of obd_ops::o_get_info.
*
* This function is not called from request handler, it is only used by
* direct call from nrs_orr_range_fill_physical() in ptlrpc, see LU-3239.
*
* \see ofd_get_info_hdl() for request handler function.
*
* \param[in] env execution environment
* \param[in] exp OBD export of OFD device
* \param[in] keylen length of \a key
* \param[in] key key name
* \param[out] vallen length of key value
* \param[out] val the key value to return
* \param[in] lsm not used in OFD
*
* \retval 0 if successful
* \retval negative value on error
*/
static int ofd_get_info(const struct lu_env *env, struct obd_export *exp,
__u32 keylen, void *key, __u32 *vallen, void *val,
struct lov_stripe_md *lsm)
{
struct ofd_thread_info *info;
struct ofd_device *ofd;
struct ll_fiemap_info_key *fm_key = key;
struct ll_user_fiemap *fiemap = val;
int rc = 0;
ENTRY;
if (exp->exp_obd == NULL) {
CDEBUG(D_IOCTL, "invalid client export %p\n", exp);
RETURN(-EINVAL);
}
ofd = ofd_exp(exp);
if (KEY_IS(KEY_FIEMAP)) {
info = ofd_info_init(env, exp);
rc = ostid_to_fid(&info->fti_fid, &fm_key->oa.o_oi,
ofd->ofd_lut.lut_lsd.lsd_osd_index);
if (rc != 0)
RETURN(rc);
rc = ofd_fiemap_get(env, ofd, &info->fti_fid, fiemap);
} else {
CERROR("%s: not supported key %s\n", ofd_name(ofd), (char*)key);
rc = -EOPNOTSUPP;
}
RETURN(rc);
}
/**
* Implementation of obd_ops::o_postrecov.
*
* This function is called from target_finish_recovery() upon recovery
* completion.
*
* \param[in] obd OBD device of OFD
*
* \retval 0 if successful
* \retval negative value on error
*/
static int ofd_obd_postrecov(struct obd_device *obd)
{
struct lu_env env;
struct lu_device *ldev = obd->obd_lu_dev;
int rc;
ENTRY;
rc = lu_env_init(&env, LCT_DT_THREAD);
if (rc)
RETURN(rc);
ofd_info_init(&env, obd->obd_self_export);
rc = ofd_postrecov(&env, ofd_dev(ldev));
lu_env_fini(&env);
RETURN(rc);
}
/**
* Implementation of ldlm_valblock_ops::lvbo_init for OFD.
*
* This function allocates and initializes new LVB data for the given
* LDLM resource if it is not allocated yet. New LVB is filled with attributes
* of the object associated with that resource. Function does nothing if LVB
* for the given LDLM resource is allocated already.
*
* Called with res->lr_lvb_sem held.
*
* \param[in] res LDLM resource
*
* \retval 0 on successful setup
* \retval negative value on error
*/
static int ofd_lvbo_init(struct ldlm_resource *res)
{
struct ost_lvb *lvb;
struct ofd_device *ofd;
struct ofd_object *fo;
struct ofd_thread_info *info;
struct lu_env env;
int rc = 0;
ENTRY;
LASSERT(res);
LASSERT(mutex_is_locked(&res->lr_lvb_mutex));
if (res->lr_lvb_data != NULL)
RETURN(0);
ofd = ldlm_res_to_ns(res)->ns_lvbp;
LASSERT(ofd != NULL);
if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_OST_LVB))
RETURN(-ENOMEM);
rc = lu_env_init(&env, LCT_DT_THREAD);
if (rc)
RETURN(rc);
OBD_ALLOC_PTR(lvb);
if (lvb == NULL)
GOTO(out_env, rc = -ENOMEM);
res->lr_lvb_data = lvb;
res->lr_lvb_len = sizeof(*lvb);
info = ofd_info_init(&env, NULL);
ost_fid_from_resid(&info->fti_fid, &res->lr_name,
ofd->ofd_lut.lut_lsd.lsd_osd_index);
fo = ofd_object_find(&env, ofd, &info->fti_fid);
if (IS_ERR(fo))
GOTO(out_lvb, rc = PTR_ERR(fo));
rc = ofd_attr_get(&env, fo, &info->fti_attr);
if (rc)
GOTO(out_obj, rc);
lvb->lvb_size = info->fti_attr.la_size;
lvb->lvb_blocks = info->fti_attr.la_blocks;
lvb->lvb_mtime = info->fti_attr.la_mtime;
lvb->lvb_atime = info->fti_attr.la_atime;
lvb->lvb_ctime = info->fti_attr.la_ctime;
CDEBUG(D_DLMTRACE, "res: "DFID" initial lvb size: "LPU64", "
"mtime: "LPX64", blocks: "LPX64"\n",
PFID(&info->fti_fid), lvb->lvb_size,
lvb->lvb_mtime, lvb->lvb_blocks);
EXIT;
out_obj:
ofd_object_put(&env, fo);
out_lvb:
if (rc != 0)
OST_LVB_SET_ERR(lvb->lvb_blocks, rc);
out_env:
lu_env_fini(&env);
/* Don't free lvb data on lookup error */
return rc;
}
/**
* Implementation of ldlm_valblock_ops::lvbo_update for OFD.
*
* When a client generates a glimpse enqueue, it wants to get the current
* file size and updated attributes for a stat() type operation, but these
* attributes may be writeback cached on another client. The client with
* the DLM extent lock at the highest offset is asked for its current
* attributes via a glimpse callback on its extent lock, on the assumption
* that it has the highest file size and the newest timestamps. The timestamps
* are guaranteed to be correct if there is only a single writer on the file,
* but may be slightly inaccurate if there are multiple concurrent writers on
* the same object. In order to avoid race conditions between the glimpse AST
* and the client cancelling the lock, ofd_lvbo_update() also updates
* the attributes from the local object. If the last client hasn't done any
* writes yet, or has already written its data and cancelled its lock before
* it processed the glimpse, then the local inode will have more uptodate
* information.
*
* This is called in two ways:
* \a req != NULL : called by the DLM itself after a glimpse callback
* \a req == NULL : called by the OFD after a disk write
*
* \param[in] res LDLM resource
* \param[in] req PTLRPC request
* \param[in] increase_only don't allow LVB values to decrease
*
* \retval 0 on successful setup
* \retval negative value on error
*/
static int ofd_lvbo_update(struct ldlm_resource *res,
struct ptlrpc_request *req, int increase_only)
{
struct ofd_device *ofd;
struct ofd_object *fo;
struct ofd_thread_info *info;
struct ost_lvb *lvb;
struct lu_env env;
int rc = 0;
ENTRY;
LASSERT(res != NULL);
ofd = ldlm_res_to_ns(res)->ns_lvbp;
LASSERT(ofd != NULL);
rc = lu_env_init(&env, LCT_DT_THREAD);
if (rc)
RETURN(rc);
info = ofd_info_init(&env, NULL);
fid_extract_from_res_name(&info->fti_fid, &res->lr_name);
lvb = res->lr_lvb_data;
if (lvb == NULL) {
CERROR("%s: no LVB data for "DFID"\n",
ofd_name(ofd), PFID(&info->fti_fid));
GOTO(out_env, rc = 0);
}
/* Update the LVB from the network message */
if (req != NULL) {
struct ost_lvb *rpc_lvb;
bool lvb_type;
if (req->rq_import != NULL)
lvb_type = imp_connect_lvb_type(req->rq_import);
else
lvb_type = exp_connect_lvb_type(req->rq_export);
if (!lvb_type) {
struct ost_lvb_v1 *lvb_v1;
lvb_v1 = req_capsule_server_swab_get(&req->rq_pill,
&RMF_DLM_LVB, lustre_swab_ost_lvb_v1);
if (lvb_v1 == NULL)
goto disk_update;
rpc_lvb = &info->fti_lvb;
memcpy(rpc_lvb, lvb_v1, sizeof *lvb_v1);
rpc_lvb->lvb_mtime_ns = 0;
rpc_lvb->lvb_atime_ns = 0;
rpc_lvb->lvb_ctime_ns = 0;
} else {
rpc_lvb = req_capsule_server_swab_get(&req->rq_pill,
&RMF_DLM_LVB,
lustre_swab_ost_lvb);
if (rpc_lvb == NULL)
goto disk_update;
}
lock_res(res);
if (rpc_lvb->lvb_size > lvb->lvb_size || !increase_only) {
CDEBUG(D_DLMTRACE, "res: "DFID" updating lvb size: "
LPU64" -> "LPU64"\n", PFID(&info->fti_fid),
lvb->lvb_size, rpc_lvb->lvb_size);
lvb->lvb_size = rpc_lvb->lvb_size;
}
if (rpc_lvb->lvb_mtime > lvb->lvb_mtime || !increase_only) {
CDEBUG(D_DLMTRACE, "res: "DFID" updating lvb mtime: "
LPU64" -> "LPU64"\n", PFID(&info->fti_fid),
lvb->lvb_mtime, rpc_lvb->lvb_mtime);
lvb->lvb_mtime = rpc_lvb->lvb_mtime;
}
if (rpc_lvb->lvb_atime > lvb->lvb_atime || !increase_only) {
CDEBUG(D_DLMTRACE, "res: "DFID" updating lvb atime: "
LPU64" -> "LPU64"\n", PFID(&info->fti_fid),
lvb->lvb_atime, rpc_lvb->lvb_atime);
lvb->lvb_atime = rpc_lvb->lvb_atime;
}
if (rpc_lvb->lvb_ctime > lvb->lvb_ctime || !increase_only) {
CDEBUG(D_DLMTRACE, "res: "DFID" updating lvb ctime: "
LPU64" -> "LPU64"\n", PFID(&info->fti_fid),
lvb->lvb_ctime, rpc_lvb->lvb_ctime);
lvb->lvb_ctime = rpc_lvb->lvb_ctime;
}
if (rpc_lvb->lvb_blocks > lvb->lvb_blocks || !increase_only) {
CDEBUG(D_DLMTRACE, "res: "DFID" updating lvb blocks: "
LPU64" -> "LPU64"\n", PFID(&info->fti_fid),
lvb->lvb_blocks, rpc_lvb->lvb_blocks);
lvb->lvb_blocks = rpc_lvb->lvb_blocks;
}
unlock_res(res);
}
disk_update:
/* Update the LVB from the disk inode */
ost_fid_from_resid(&info->fti_fid, &res->lr_name,
ofd->ofd_lut.lut_lsd.lsd_osd_index);
fo = ofd_object_find(&env, ofd, &info->fti_fid);
if (IS_ERR(fo))
GOTO(out_env, rc = PTR_ERR(fo));
rc = ofd_attr_get(&env, fo, &info->fti_attr);
if (rc)
GOTO(out_obj, rc);
lock_res(res);
if (info->fti_attr.la_size > lvb->lvb_size || !increase_only) {
CDEBUG(D_DLMTRACE, "res: "DFID" updating lvb size from disk: "
LPU64" -> %llu\n", PFID(&info->fti_fid),
lvb->lvb_size, info->fti_attr.la_size);
lvb->lvb_size = info->fti_attr.la_size;
}
if (info->fti_attr.la_mtime >lvb->lvb_mtime || !increase_only) {
CDEBUG(D_DLMTRACE, "res: "DFID" updating lvb mtime from disk: "
LPU64" -> "LPU64"\n", PFID(&info->fti_fid),
lvb->lvb_mtime, info->fti_attr.la_mtime);
lvb->lvb_mtime = info->fti_attr.la_mtime;
}
if (info->fti_attr.la_atime >lvb->lvb_atime || !increase_only) {
CDEBUG(D_DLMTRACE, "res: "DFID" updating lvb atime from disk: "
LPU64" -> "LPU64"\n", PFID(&info->fti_fid),
lvb->lvb_atime, info->fti_attr.la_atime);
lvb->lvb_atime = info->fti_attr.la_atime;
}
if (info->fti_attr.la_ctime >lvb->lvb_ctime || !increase_only) {
CDEBUG(D_DLMTRACE, "res: "DFID" updating lvb ctime from disk: "
LPU64" -> "LPU64"\n", PFID(&info->fti_fid),
lvb->lvb_ctime, info->fti_attr.la_ctime);
lvb->lvb_ctime = info->fti_attr.la_ctime;
}
if (info->fti_attr.la_blocks > lvb->lvb_blocks || !increase_only) {
CDEBUG(D_DLMTRACE, "res: "DFID" updating lvb blocks from disk: "
LPU64" -> %llu\n", PFID(&info->fti_fid), lvb->lvb_blocks,
(unsigned long long)info->fti_attr.la_blocks);
lvb->lvb_blocks = info->fti_attr.la_blocks;
}
unlock_res(res);
out_obj:
ofd_object_put(&env, fo);
out_env:
lu_env_fini(&env);
return rc;
}
/**
* 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);
}
