/**
* Process a granting attempt for plain lock.
* Must be called with ns lock held.
*
* This function looks for any conflicts for \a lock in the granted or
* waiting queues. The lock is granted if no conflicts are found in
* either queue.
*
* If \a first_enq is 0 (ie, called from ldlm_reprocess_queue):
* - blocking ASTs have already been sent
*
* If \a first_enq is 1 (ie, called from ldlm_lock_enqueue):
* - blocking ASTs have not been sent yet, so list of conflicting locks
* would be collected and ASTs sent.
*/
int ldlm_process_plain_lock(struct ldlm_lock *lock, __u64 *flags,
int first_enq, ldlm_error_t *err,
struct list_head *work_list)
{
struct ldlm_resource *res = lock->l_resource;
struct list_head rpc_list;
int rc;
ENTRY;
LASSERT(lock->l_granted_mode != lock->l_req_mode);
check_res_locked(res);
LASSERT(list_empty(&res->lr_converting));
INIT_LIST_HEAD(&rpc_list);
if (!first_enq) {
LASSERT(work_list != NULL);
rc = ldlm_plain_compat_queue(&res->lr_granted, lock, NULL);
if (!rc)
RETURN(LDLM_ITER_STOP);
rc = ldlm_plain_compat_queue(&res->lr_waiting, lock, NULL);
if (!rc)
RETURN(LDLM_ITER_STOP);
ldlm_resource_unlink_lock(lock);
ldlm_grant_lock(lock, work_list);
RETURN(LDLM_ITER_CONTINUE);
}
restart:
rc = ldlm_plain_compat_queue(&res->lr_granted, lock, &rpc_list);
rc += ldlm_plain_compat_queue(&res->lr_waiting, lock, &rpc_list);
if (rc != 2) {
/* If either of the compat_queue()s returned 0, then we
* have ASTs to send and must go onto the waiting list.
*
* bug 2322: we used to unlink and re-add here, which was a
* terrible folly -- if we goto restart, we could get
* re-ordered! Causes deadlock, because ASTs aren't sent! */
if (list_empty(&lock->l_res_link))
ldlm_resource_add_lock(res, &res->lr_waiting, lock);
unlock_res(res);
rc = ldlm_run_ast_work(ldlm_res_to_ns(res), &rpc_list,
LDLM_WORK_BL_AST);
lock_res(res);
if (rc == -ERESTART)
GOTO(restart, rc);
*flags |= LDLM_FL_BLOCK_GRANTED;
} else {
ldlm_resource_unlink_lock(lock);
ldlm_grant_lock(lock, NULL);
}
RETURN(0);
}
/* Called with res->lr_lvb_sem held */
static int mdt_lvbo_init(struct ldlm_resource *res)
{
if (IS_LQUOTA_RES(res)) {
struct mdt_device *mdt;
mdt = ldlm_res_to_ns(res)->ns_lvbp;
if (mdt->mdt_qmt_dev == NULL)
return 0;
/* call lvbo init function of quota master */
return qmt_hdls.qmth_lvbo_init(mdt->mdt_qmt_dev, res);
}
return 0;
}
static int mdt_lvbo_update(struct ldlm_resource *res,
struct ptlrpc_request *req,
int increase_only)
{
if (IS_LQUOTA_RES(res)) {
struct mdt_device *mdt;
mdt = ldlm_res_to_ns(res)->ns_lvbp;
if (mdt->mdt_qmt_dev == NULL)
return 0;
/* call lvbo update function of quota master */
return qmt_hdls.qmth_lvbo_update(mdt->mdt_qmt_dev, res, req,
increase_only);
}
return 0;
}
static int mdt_lvbo_size(struct ldlm_lock *lock)
{
struct mdt_device *mdt;
/* resource on server side never changes. */
mdt = ldlm_res_to_ns(lock->l_resource)->ns_lvbp;
LASSERT(mdt != NULL);
if (IS_LQUOTA_RES(lock->l_resource)) {
if (mdt->mdt_qmt_dev == NULL)
return 0;
/* call lvbo size function of quota master */
return qmt_hdls.qmth_lvbo_size(mdt->mdt_qmt_dev, lock);
}
if (ldlm_has_layout(lock))
return mdt->mdt_max_mdsize;
return 0;
}
/**
* 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;
}
/**
* 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] lock LDLM lock on resource
*
* \retval 0 on successful setup
* \retval negative value on error
*/
static int ofd_lvbo_init(const struct lu_env *env, 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);
if (!env) {
rc = lu_env_init(&_env, LCT_DT_THREAD);
if (rc)
RETURN(rc);
env = &_env;
}
OBD_ALLOC_PTR(lvb);
if (lvb == NULL)
GOTO(out, rc = -ENOMEM);
info = ofd_info(env);
res->lr_lvb_data = lvb;
res->lr_lvb_len = sizeof(*lvb);
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) {
struct ofd_seq *oseq;
__u64 seq;
/* Object could be recreated during the first
* CLEANUP_ORPHAN request. */
if (rc == -ENOENT) {
seq = fid_seq(&info->fti_fid);
oseq = ofd_seq_load(env, ofd, fid_seq_is_idif(seq) ?
FID_SEQ_OST_MDT0 : seq);
if (!IS_ERR_OR_NULL(oseq)) {
if (!oseq->os_last_id_synced)
rc = -EAGAIN;
ofd_seq_put(env, oseq);
}
}
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: %llu, "
"mtime: %#llx, blocks: %#llx\n",
PFID(&info->fti_fid), lvb->lvb_size,
lvb->lvb_mtime, lvb->lvb_blocks);
info->fti_attr.la_valid = 0;
EXIT;
out_obj:
ofd_object_put(env, fo);
out_lvb:
if (rc != 0)
OST_LVB_SET_ERR(lvb->lvb_blocks, rc);
out:
/* Don't free lvb data on lookup error */
if (env && env == &_env)
lu_env_fini(&_env);
return rc;
}
