#define DEBUG_SUBSYSTEM S_MDS
#include <obd_support.h>
-#include <lustre_net.h>
#include <lustre_export.h>
#include <obd.h>
#include <lprocfs_status.h>
hcdcb.cdt = &mdt->mdt_coordinator;
hcdcb.hal = hal;
- rc = cdt_llog_process(env, mdt, hsm_find_compatible_cb, &hcdcb);
+ rc = cdt_llog_process(env, mdt, hsm_find_compatible_cb, &hcdcb, 0, 0,
+ READ);
RETURN(rc);
}
int rc;
ENTRY;
- if (hsma != HSMA_RESTORE &&
- exp_connect_flags(mti->mti_exp) & OBD_CONNECT_RDONLY)
+ if (hsma != HSMA_RESTORE && mdt_rdonly(mti->mti_exp))
RETURN(-EROFS);
if (md_capable(uc, CFS_CAP_SYS_ADMIN))
mdt_object_put(mti->mti_env, obj);
mutex_lock(&cdt->cdt_restore_lock);
+ if (unlikely((cdt->cdt_state == CDT_STOPPED) ||
+ (cdt->cdt_state == CDT_STOPPING))) {
+ mutex_unlock(&cdt->cdt_restore_lock);
+ mdt_object_unlock(mti, NULL, &crh->crh_lh, 1);
+ OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
+ GOTO(out, rc = -EAGAIN);
+ }
list_add_tail(&crh->crh_list, &cdt->cdt_restore_hdl);
mutex_unlock(&cdt->cdt_restore_lock);
}
record:
+ /*
+ * Wait here to catch the 2nd RESTORE request to the same FID.
+ * Normally layout lock protects against adding such request.
+ * But when cdt is stopping it cancel all locks via
+ * ldlm_resource_clean and protections may not work.
+ * See LU-9266 and sanity-hsm_407 for details.
+ */
+ OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_HSM_CDT_DELAY, cfs_fail_val);
/* record request */
rc = mdt_agent_record_add(mti->mti_env, mdt, compound_id,
archive_id, flags, hai);
GOTO(out, rc);
out:
- /* if work has been added, wake up coordinator */
+ /* if work has been added, signal the coordinator */
if (rc == 0 || rc == -ENODATA)
- mdt_hsm_cdt_wakeup(mdt);
+ mdt_hsm_cdt_event(cdt);
return rc;
}
/**
- * get running action on a FID list or from cookie
- * \param mti [IN]
- * \param hal [IN/OUT] requests
- * \retval 0 success
- * \retval -ve failure
- */
-int mdt_hsm_get_running(struct mdt_thread_info *mti,
- struct hsm_action_list *hal)
-{
- struct mdt_device *mdt = mti->mti_mdt;
- struct coordinator *cdt = &mdt->mdt_coordinator;
- struct hsm_action_item *hai;
- int i;
- ENTRY;
-
- hai = hai_first(hal);
- for (i = 0; i < hal->hal_count; i++, hai = hai_next(hai)) {
- struct cdt_agent_req *car;
-
- if (!fid_is_sane(&hai->hai_fid))
- RETURN(-EINVAL);
-
- car = mdt_cdt_find_request(cdt, 0, &hai->hai_fid);
- if (car == NULL) {
- hai->hai_cookie = 0;
- hai->hai_action = HSMA_NONE;
- } else {
- *hai = *car->car_hai;
- mdt_cdt_put_request(car);
- }
- }
- RETURN(0);
-}
-
-/**
* check if a restore is running on a FID
* this is redundant with mdt_hsm_coordinator_get_running()
* but as it can be called frequently when getting attr
for (i = 0; i < hal->hal_count; i++, hai = hai_next(hai)) {
struct cdt_agent_req *car;
- car = mdt_cdt_find_request(cdt, hai->hai_cookie, NULL);
+ car = mdt_cdt_find_request(cdt, hai->hai_cookie);
if (car == NULL) {
hai->hai_cookie = 0;
} else {