Whamcloud - gitweb
LU-5691 hsm: remove a request from the index if not present in the store
[fs/lustre-release.git] / lustre / mdt / mdt_coordinator.c
index 5c07bf0..ac5518f 100644 (file)
 #include <lustre_net.h>
 #include <lustre_export.h>
 #include <obd.h>
-#include <obd_lov.h>
 #include <lprocfs_status.h>
 #include <lustre_log.h>
 #include "mdt_internal.h"
 
-static struct lprocfs_vars lprocfs_mdt_hsm_vars[];
+static struct lprocfs_seq_vars lprocfs_mdt_hsm_vars[];
 
 /**
  * get obj and HSM attributes on a fid
@@ -133,7 +132,7 @@ struct hsm_scan_data {
        char                             fs_name[MTI_NAME_MAXLEN+1];
        /* request to be send to agents */
        int                              request_sz;    /** allocated size */
-       int                              max_request  /** vector size */
+       int                              max_requests;  /** vector size */
        int                              request_cnt;   /** used count */
        struct {
                int                      hal_sz;
@@ -182,14 +181,14 @@ static int mdt_coordinator_cb(const struct lu_env *env,
 
                /* Are agents full? */
                if (atomic_read(&cdt->cdt_request_count) ==
-                   cdt->cdt_max_request)
+                   cdt->cdt_max_requests)
                        break;
 
                /* first search if the request if known in the list we have
                 * build and if there is room in the request vector */
                empty_slot = -1;
                found = -1;
-               for (i = 0; i < hsd->max_request &&
+               for (i = 0; i < hsd->max_requests &&
                            (empty_slot == -1 || found == -1); i++) {
                        if (hsd->request[i].hal == NULL) {
                                empty_slot = i;
@@ -228,9 +227,8 @@ static int mdt_coordinator_cb(const struct lu_env *env,
                                RETURN(-ENOMEM);
                        }
                        hal->hal_version = HAL_VERSION;
-                       strncpy(hal->hal_fsname, hsd->fs_name,
-                               MTI_NAME_MAXLEN);
-                       hal->hal_fsname[MTI_NAME_MAXLEN] = '\0';
+                       strlcpy(hal->hal_fsname, hsd->fs_name,
+                               MTI_NAME_MAXLEN + 1);
                        hal->hal_compound_id = larr->arr_compound_id;
                        hal->hal_archive_id = larr->arr_archive_id;
                        hal->hal_flags = larr->arr_flags;
@@ -308,7 +306,8 @@ static int mdt_coordinator_cb(const struct lu_env *env,
 
                /* test if request too long, if yes cancel it
                 * the same way the copy tool acknowledge a cancel request */
-               if ((last + cdt->cdt_timeout) < cfs_time_current_sec()) {
+               if ((last + cdt->cdt_active_req_timeout)
+                    < cfs_time_current_sec()) {
                        struct hsm_progress_kernel pgs;
 
                        dump_llog_agent_req_rec("mdt_coordinator_cb(): "
@@ -337,6 +336,15 @@ static int mdt_coordinator_cb(const struct lu_env *env,
                                       hsm_copytool_action2name(
                                                     larr->arr_hai.hai_action));
 
+                       if (rc == -ENOENT) {
+                               /* The request no longer exists, forget
+                                * about it, and do not send a cancel request
+                                * to the client, for which an error will be
+                                * sent back, leading to an endless cycle of
+                                * cancellation. */
+                               RETURN(LLOG_DEL_RECORD);
+                       }
+
                        /* add the cookie to the list of record to be
                         * canceled by caller */
                        if (hsd->max_cookie == (hsd->cookie_cnt - 1)) {
@@ -371,7 +379,7 @@ static int mdt_coordinator_cb(const struct lu_env *env,
        case ARS_FAILED:
        case ARS_CANCELED:
        case ARS_SUCCEED:
-               if ((larr->arr_req_change + cdt->cdt_delay) <
+               if ((larr->arr_req_change + cdt->cdt_grace_delay) <
                    cfs_time_current_sec())
                        RETURN(LLOG_DEL_RECORD);
                break;
@@ -385,14 +393,14 @@ static int mdt_coordinator_cb(const struct lu_env *env,
  * \retval 0 success
  * \retval -ve failure
  */
-static int hsm_cdt_procfs_init(struct mdt_device *mdt)
+int hsm_cdt_procfs_init(struct mdt_device *mdt)
 {
        struct coordinator      *cdt = &mdt->mdt_coordinator;
        int                      rc = 0;
        ENTRY;
 
        /* init /proc entries, failure is not critical */
-       cdt->cdt_proc_dir = lprocfs_register("hsm",
+       cdt->cdt_proc_dir = lprocfs_seq_register("hsm",
                                             mdt2obd_dev(mdt)->obd_proc_entry,
                                             lprocfs_mdt_hsm_vars, mdt);
        if (IS_ERR(cdt->cdt_proc_dir)) {
@@ -407,6 +415,29 @@ static int hsm_cdt_procfs_init(struct mdt_device *mdt)
 }
 
 /**
+ * remove /proc entries for coordinator
+ * \param mdt [IN]
+ */
+void  hsm_cdt_procfs_fini(struct mdt_device *mdt)
+{
+       struct coordinator      *cdt = &mdt->mdt_coordinator;
+
+       LASSERT(cdt->cdt_state == CDT_STOPPED);
+       if (cdt->cdt_proc_dir != NULL)
+               lprocfs_remove(&cdt->cdt_proc_dir);
+}
+
+/**
+ * get vector of hsm cdt /proc vars
+ * \param none
+ * \retval var vector
+ */
+struct lprocfs_seq_vars *hsm_cdt_get_proc_vars(void)
+{
+       return lprocfs_mdt_hsm_vars;
+}
+
+/**
  * coordinator thread
  * \param data [IN] obd device
  * \retval 0 success
@@ -422,25 +453,21 @@ static int mdt_coordinator(void *data)
        ENTRY;
 
        cdt->cdt_thread.t_flags = SVC_RUNNING;
-       cfs_waitq_signal(&cdt->cdt_thread.t_ctl_waitq);
+       wake_up(&cdt->cdt_thread.t_ctl_waitq);
 
        CDEBUG(D_HSM, "%s: coordinator thread starting, pid=%d\n",
               mdt_obd_name(mdt), current_pid());
 
-       /*
-        * create /proc entries for coordinator
-        */
-       hsm_cdt_procfs_init(mdt);
        /* timeouted cookie vector initialization */
        hsd.max_cookie = 0;
        hsd.cookie_cnt = 0;
        hsd.cookies = NULL;
-       /* we use a copy of cdt_max_request in the cb, so if cdt_max_request
+       /* we use a copy of cdt_max_requests in the cb, so if cdt_max_requests
         * increases due to a change from /proc we do not overflow the
         * hsd.request[] vector
         */
-       hsd.max_request = cdt->cdt_max_request;
-       hsd.request_sz = hsd.max_request * sizeof(*hsd.request);
+       hsd.max_requests = cdt->cdt_max_requests;
+       hsd.request_sz = hsd.max_requests * sizeof(*hsd.request);
        OBD_ALLOC(hsd.request, hsd.request_sz);
        if (!hsd.request)
                GOTO(out, rc = -ENOMEM);
@@ -480,14 +507,14 @@ static int mdt_coordinator(void *data)
 
                CDEBUG(D_HSM, "coordinator starts reading llog\n");
 
-               if (hsd.max_request != cdt->cdt_max_request) {
-                       /* cdt_max_request has changed,
+               if (hsd.max_requests != cdt->cdt_max_requests) {
+                       /* cdt_max_requests has changed,
                         * we need to allocate a new buffer
                         */
                        OBD_FREE(hsd.request, hsd.request_sz);
-                       hsd.max_request = cdt->cdt_max_request;
+                       hsd.max_requests = cdt->cdt_max_requests;
                        hsd.request_sz =
-                                  hsd.max_request * sizeof(*hsd.request);
+                                  hsd.max_requests * sizeof(*hsd.request);
                        OBD_ALLOC(hsd.request, hsd.request_sz);
                        if (!hsd.request) {
                                rc = -ENOMEM;
@@ -537,7 +564,7 @@ static int mdt_coordinator(void *data)
                }
 
                /* here hsd contains a list of requests to be started */
-               for (i = 0; i < hsd.max_request; i++) {
+               for (i = 0; i < hsd.max_requests; i++) {
                        struct hsm_action_list  *hal;
                        struct hsm_action_item  *hai;
                        __u64                   *cookies;
@@ -546,7 +573,7 @@ static int mdt_coordinator(void *data)
 
                        /* still room for work ? */
                        if (atomic_read(&cdt->cdt_request_count) ==
-                           cdt->cdt_max_request)
+                           cdt->cdt_max_requests)
                                break;
 
                        if (hsd.request[i].hal == NULL)
@@ -619,7 +646,7 @@ clean_cb_alloc:
                }
 
                /* free hal allocated by callback */
-               for (i = 0; i < hsd.max_request; i++) {
+               for (i = 0; i < hsd.max_requests; i++) {
                        if (hsd.request[i].hal) {
                                OBD_FREE(hsd.request[i].hal,
                                         hsd.request[i].hal_sz);
@@ -652,7 +679,7 @@ out:
                 * and cdt cleaning will be done by event sender
                 */
                cdt->cdt_thread.t_flags = SVC_STOPPED;
-               cfs_waitq_signal(&cdt->cdt_thread.t_ctl_waitq);
+               wake_up(&cdt->cdt_thread.t_ctl_waitq);
        }
 
        if (rc != 0)
@@ -725,6 +752,10 @@ static int hsm_restore_cb(const struct lu_env *env,
 
        larr = (struct llog_agent_req_rec *)hdr;
        hai = &larr->arr_hai;
+       if (hai->hai_cookie > cdt->cdt_last_cookie)
+               /* update the cookie to avoid collision */
+               cdt->cdt_last_cookie = hai->hai_cookie + 1;
+
        if (hai->hai_action != HSMA_RESTORE ||
            agent_req_in_final_state(larr->arr_status))
                RETURN(0);
@@ -741,7 +772,7 @@ static int hsm_restore_cb(const struct lu_env *env,
        crh->extent.end = hai->hai_extent.offset + hai->hai_extent.length;
        */
        crh->crh_extent.start = 0;
-       crh->crh_extent.end = OBD_OBJECT_EOF;
+       crh->crh_extent.end = hai->hai_extent.length;
        /* get the layout lock */
        mdt_lock_reg_init(&crh->crh_lh, LCK_EX);
        child = mdt_object_find_lock(mti, &crh->crh_fid, &crh->crh_lh,
@@ -796,7 +827,7 @@ static int hsm_init_ucred(struct lu_ucred *uc)
        uc->uc_fsgid = 0;
        uc->uc_suppgids[0] = -1;
        uc->uc_suppgids[1] = -1;
-       uc->uc_cap = 0;
+       uc->uc_cap = CFS_CAP_FS_MASK;
        uc->uc_umask = 0777;
        uc->uc_ginfo = NULL;
        uc->uc_identity = NULL;
@@ -820,7 +851,7 @@ int mdt_hsm_cdt_wakeup(struct mdt_device *mdt)
 
        /* wake up coordinator */
        cdt->cdt_thread.t_flags = SVC_EVENT;
-       cfs_waitq_signal(&cdt->cdt_thread.t_ctl_waitq);
+       wake_up(&cdt->cdt_thread.t_ctl_waitq);
 
        RETURN(0);
 }
@@ -840,22 +871,22 @@ int mdt_hsm_cdt_init(struct mdt_device *mdt)
 
        cdt->cdt_state = CDT_STOPPED;
 
-       cfs_waitq_init(&cdt->cdt_thread.t_ctl_waitq);
+       init_waitqueue_head(&cdt->cdt_thread.t_ctl_waitq);
        mutex_init(&cdt->cdt_llog_lock);
        init_rwsem(&cdt->cdt_agent_lock);
        init_rwsem(&cdt->cdt_request_lock);
        mutex_init(&cdt->cdt_restore_lock);
 
-       CFS_INIT_LIST_HEAD(&cdt->cdt_requests);
-       CFS_INIT_LIST_HEAD(&cdt->cdt_agents);
-       CFS_INIT_LIST_HEAD(&cdt->cdt_restore_hdl);
+       INIT_LIST_HEAD(&cdt->cdt_requests);
+       INIT_LIST_HEAD(&cdt->cdt_agents);
+       INIT_LIST_HEAD(&cdt->cdt_restore_hdl);
 
        rc = lu_env_init(&cdt->cdt_env, LCT_MD_THREAD);
        if (rc < 0)
                RETURN(rc);
 
        /* for mdt_ucred(), lu_ucred stored in lu_ucred_key */
-       rc = lu_context_init(&cdt->cdt_session, LCT_SESSION);
+       rc = lu_context_init(&cdt->cdt_session, LCT_SERVER_SESSION);
        if (rc == 0) {
                lu_context_enter(&cdt->cdt_session);
                cdt->cdt_env.le_ses = &cdt->cdt_session;
@@ -872,6 +903,15 @@ int mdt_hsm_cdt_init(struct mdt_device *mdt)
 
        hsm_init_ucred(mdt_ucred(cdt_mti));
 
+       /* default values for /proc tunnables
+        * can be override by MGS conf */
+       cdt->cdt_default_archive_id = 1;
+       cdt->cdt_grace_delay = 60;
+       cdt->cdt_loop_period = 10;
+       cdt->cdt_max_requests = 3;
+       cdt->cdt_policy = CDT_DEFAULT_POLICY;
+       cdt->cdt_active_req_timeout = 3600;
+
        RETURN(0);
 }
 
@@ -918,18 +958,19 @@ int mdt_hsm_cdt_start(struct mdt_device *mdt)
                RETURN(-EALREADY);
        }
 
+       CLASSERT(1 << (CDT_POLICY_SHIFT_COUNT - 1) == CDT_POLICY_LAST);
        cdt->cdt_policy = CDT_DEFAULT_POLICY;
+
        cdt->cdt_state = CDT_INIT;
 
        atomic_set(&cdt->cdt_compound_id, cfs_time_current_sec());
        /* just need to be larger than previous one */
        /* cdt_last_cookie is protected by cdt_llog_lock */
        cdt->cdt_last_cookie = cfs_time_current_sec();
-       cdt->cdt_loop_period = 10;
-       cdt->cdt_delay = 60;
-       cdt->cdt_timeout = 3600;
-       cdt->cdt_max_request = 3;
        atomic_set(&cdt->cdt_request_count, 0);
+       cdt->cdt_user_request_mask = (1UL << HSMA_RESTORE);
+       cdt->cdt_group_request_mask = (1UL << HSMA_RESTORE);
+       cdt->cdt_other_request_mask = (1UL << HSMA_RESTORE);
 
        /* to avoid deadlock when start is made through /proc
         * /proc entries are created by the coordinator thread */
@@ -939,7 +980,7 @@ int mdt_hsm_cdt_start(struct mdt_device *mdt)
        rc = mdt_hsm_pending_restore(cdt_mti);
        if (rc)
                CERROR("%s: cannot take the layout locks needed"
-                      " for registered restore: %d",
+                      " for registered restore: %d\n",
                       mdt_obd_name(mdt), rc);
 
        task = kthread_run(mdt_coordinator, cdt_mti, "hsm_cdtr");
@@ -955,7 +996,7 @@ int mdt_hsm_cdt_start(struct mdt_device *mdt)
                rc = 0;
        }
 
-       cfs_wait_event(cdt->cdt_thread.t_ctl_waitq,
+       wait_event(cdt->cdt_thread.t_ctl_waitq,
                       (cdt->cdt_thread.t_flags & SVC_RUNNING));
 
        cdt->cdt_state = CDT_RUNNING;
@@ -982,16 +1023,12 @@ int mdt_hsm_cdt_stop(struct mdt_device *mdt)
                RETURN(-EALREADY);
        }
 
-       /* remove proc entries */
-       if (cdt->cdt_proc_dir != NULL)
-               lprocfs_remove(&cdt->cdt_proc_dir);
-
        if (cdt->cdt_state != CDT_STOPPING) {
                /* stop coordinator thread before cleaning */
                cdt->cdt_thread.t_flags = SVC_STOPPING;
-               cfs_waitq_signal(&cdt->cdt_thread.t_ctl_waitq);
-               cfs_wait_event(cdt->cdt_thread.t_ctl_waitq,
-                              cdt->cdt_thread.t_flags & SVC_STOPPED);
+               wake_up(&cdt->cdt_thread.t_ctl_waitq);
+               wait_event(cdt->cdt_thread.t_ctl_waitq,
+                          cdt->cdt_thread.t_flags & SVC_STOPPED);
        }
        cdt->cdt_state = CDT_STOPPED;
 
@@ -1127,9 +1164,11 @@ out:
  * \param mti [IN] context
  * \param fid1 [IN]
  * \param fid2 [IN]
+ * \param mh_common [IN] MD HSM
  */
 static int hsm_swap_layouts(struct mdt_thread_info *mti,
-                           const lustre_fid *fid, const lustre_fid *dfid)
+                           const lustre_fid *fid, const lustre_fid *dfid,
+                           struct md_hsm *mh_common)
 {
        struct mdt_device       *mdt = mti->mti_mdt;
        struct mdt_object       *child1, *child2;
@@ -1152,15 +1191,28 @@ static int hsm_swap_layouts(struct mdt_thread_info *mti,
        /* if copy tool closes the volatile before sending the final
         * progress through llapi_hsm_copy_end(), all the objects
         * are removed and mdd_swap_layout LBUG */
-       if (mdt_object_exists(child2)) {
-               rc = mo_swap_layouts(mti->mti_env, mdt_object_child(child1),
-                                    mdt_object_child(child2), 0);
-       } else {
+       if (!mdt_object_exists(child2)) {
                CERROR("%s: Copytool has closed volatile file "DFID"\n",
                       mdt_obd_name(mti->mti_mdt), PFID(dfid));
-               rc = -ENOENT;
+               GOTO(out_child2, rc = -ENOENT);
        }
+       /* Since we only handle restores here, unconditionally use
+        * SWAP_LAYOUTS_MDS_HSM flag to ensure original layout will
+        * be preserved in case of failure during swap_layout and not
+        * leave a file in an intermediate but incoherent state.
+        * But need to setup HSM xattr of data FID before, reuse
+        * mti and mh presets for FID in hsm_cdt_request_completed(),
+        * only need to clear RELEASED and DIRTY.
+        */
+       mh_common->mh_flags &= ~(HS_RELEASED | HS_DIRTY);
+       rc = mdt_hsm_attr_set(mti, child2, mh_common);
+       if (rc == 0)
+               rc = mo_swap_layouts(mti->mti_env,
+                                    mdt_object_child(child1),
+                                    mdt_object_child(child2),
+                                    SWAP_LAYOUTS_MDS_HSM);
 
+out_child2:
        mdt_object_unlock_put(mti, child2, lh2, 1);
 out_child1:
        mdt_object_put(mti->mti_env, child1);
@@ -1290,8 +1342,10 @@ static int hsm_cdt_request_completed(struct mdt_thread_info *mti,
                case HSMA_RESTORE:
                        hsm_set_cl_event(&cl_flags, HE_RESTORE);
 
-                       /* clear RELEASED and DIRTY */
-                       mh.mh_flags &= ~(HS_RELEASED | HS_DIRTY);
+                       /* do not clear RELEASED and DIRTY here
+                        * this will occur in hsm_swap_layouts()
+                        */
+
                        /* Restoring has changed the file version on
                         * disk. */
                        mh.mh_arch_ver = pgs->hpk_data_version;
@@ -1348,7 +1402,7 @@ unlock:
                 * only if restore is successfull */
                if (pgs->hpk_errval == 0) {
                        rc = hsm_swap_layouts(mti, &car->car_hai->hai_fid,
-                                             &car->car_hai->hai_dfid);
+                                             &car->car_hai->hai_dfid, &mh);
                        if (rc) {
                                if (cdt->cdt_policy & CDT_NORETRY_ACTION)
                                        *status = ARS_FAILED;
@@ -1711,8 +1765,8 @@ static const struct {
        char            *name;
        char            *nickname;
 } hsm_policy_names[] = {
-       { CDT_NONBLOCKING_RESTORE,      "non_blocking_restore", "nbr"},
-       { CDT_NORETRY_ACTION,           "no_retry_action",      "nra"},
+       { CDT_NONBLOCKING_RESTORE,      "NonBlockingRestore",   "NBR"},
+       { CDT_NORETRY_ACTION,           "NoRetryAction",        "NRA"},
        { 0 },
 };
 
@@ -1727,7 +1781,8 @@ static __u64 hsm_policy_str2bit(const char *name)
        int      i;
 
        for (i = 0; hsm_policy_names[i].bit != 0; i++)
-               if (strcmp(hsm_policy_names[i].nickname, name) == 0)
+               if (strcmp(hsm_policy_names[i].nickname, name) == 0 ||
+                   strcmp(hsm_policy_names[i].name, name) == 0)
                        return hsm_policy_names[i].bit;
        return 0;
 }
@@ -1735,149 +1790,152 @@ static __u64 hsm_policy_str2bit(const char *name)
 /**
  * convert a policy bit field to a string
  * \param mask [IN] policy bit field
+ * \param hexa [IN] print mask before bit names
  * \param buffer [OUT] string
  * \param count [IN] size of buffer
- * \retval size filled in buffer
  */
-static int hsm_policy_bit2str(const __u64 mask, char *buffer, int count)
+static void hsm_policy_bit2str(struct seq_file *m, const __u64 mask,
+                               const bool hexa)
 {
-       int      i, j, sz;
-       char    *ptr;
+       int      i, j;
        __u64    bit;
        ENTRY;
 
-       ptr = buffer;
-       sz = snprintf(buffer, count, "("LPX64") ", mask);
-       ptr += sz;
-       count -= sz;
-       for (i = 0; i < (sizeof(mask) * 8); i++) {
+       if (hexa)
+               seq_printf(m, "("LPX64") ", mask);
+
+       for (i = 0; i < CDT_POLICY_SHIFT_COUNT; i++) {
                bit = (1ULL << i);
-               if (!(bit  & mask))
-                       continue;
 
                for (j = 0; hsm_policy_names[j].bit != 0; j++) {
-                       if (hsm_policy_names[j].bit == bit) {
-                               sz = snprintf(ptr, count, "%s(%s) ",
-                                             hsm_policy_names[j].name,
-                                             hsm_policy_names[j].nickname);
-                               ptr += sz;
-                               count -= sz;
+                       if (hsm_policy_names[j].bit == bit)
                                break;
-                       }
                }
+               if (bit & mask)
+                       seq_printf(m, "[%s] ", hsm_policy_names[j].name);
+               else
+                       seq_printf(m, "%s ", hsm_policy_names[j].name);
        }
-       RETURN(ptr - buffer);
+       /* remove last ' ' */
+       m->count--;
+       seq_putc(m, '\0');
 }
 
 /* methods to read/write HSM policy flags */
-static int lprocfs_rd_hsm_policy(char *page, char **start, off_t off,
-                                int count, int *eof, void *data)
+static int mdt_hsm_policy_seq_show(struct seq_file *m, void *data)
 {
-       struct mdt_device       *mdt = data;
+       struct mdt_device       *mdt = m->private;
        struct coordinator      *cdt = &mdt->mdt_coordinator;
-       int                      sz;
        ENTRY;
 
-       sz = hsm_policy_bit2str(cdt->cdt_policy, page, count);
-       page[sz] = '\n';
-       sz++;
-       page[sz] = '\0';
-       *eof = 1;
-       RETURN(sz);
+       hsm_policy_bit2str(m, cdt->cdt_policy, false);
+       RETURN(0);
 }
 
-static int lprocfs_wr_hsm_policy(struct file *file, const char *buffer,
-                                unsigned long count, void *data)
+static ssize_t
+mdt_hsm_policy_seq_write(struct file *file, const char __user *buffer,
+                        size_t count, loff_t *off)
 {
-       struct mdt_device       *mdt = data;
+       struct seq_file         *m = file->private_data;
+       struct mdt_device       *mdt = m->private;
        struct coordinator      *cdt = &mdt->mdt_coordinator;
-       int                      sz;
-       char                    *start, *end;
-       __u64                    policy;
-       int                      set;
+       char                    *start, *token, sign;
        char                    *buf;
+       __u64                    policy;
+       __u64                    add_mask, remove_mask, set_mask;
+       int                      rc;
        ENTRY;
 
-       if (strncmp(buffer, "help", 4) == 0) {
-               sz = PAGE_SIZE;
-               OBD_ALLOC(buf, sz);
-               if (!buf)
-                       RETURN(-ENOMEM);
-
-               hsm_policy_bit2str(CDT_POLICY_MASK, buf, sz);
-               CWARN("Supported policies are: %s\n", buf);
-               OBD_FREE(buf, sz);
-               RETURN(count);
-       }
+       if (count + 1 > PAGE_SIZE)
+               RETURN(-EINVAL);
 
        OBD_ALLOC(buf, count + 1);
        if (buf == NULL)
                RETURN(-ENOMEM);
 
        if (copy_from_user(buf, buffer, count))
-               RETURN(-EFAULT);
+               GOTO(out, rc = -EFAULT);
 
        buf[count] = '\0';
+
        start = buf;
+       CDEBUG(D_HSM, "%s: receive new policy: '%s'\n", mdt_obd_name(mdt),
+              start);
 
-       policy = 0;
+       add_mask = remove_mask = set_mask = 0;
        do {
-               end = strchr(start, ' ');
-               if (end != NULL)
-                       *end = '\0';
-               switch (*start) {
+               token = strsep(&start, "\n ");
+               sign = *token;
+
+               if (sign == '\0')
+                       continue;
+
+               if (sign == '-' || sign == '+')
+                       token++;
+
+               policy = hsm_policy_str2bit(token);
+               if (policy == 0) {
+                       CWARN("%s: '%s' is unknown, "
+                             "supported policies are:\n", mdt_obd_name(mdt),
+                             token);
+                       hsm_policy_bit2str(m, 0, false);
+                       GOTO(out, rc = -EINVAL);
+               }
+               switch (sign) {
                case '-':
-                       start++;
-                       set = 0;
+                       remove_mask |= policy;
                        break;
                case '+':
-                       start++;
-                       set = 1;
+                       add_mask |= policy;
                        break;
                default:
-                       set = 2;
+                       set_mask |= policy;
                        break;
                }
-               policy = hsm_policy_str2bit(start);
-               if (!policy)
-                       break;
 
-               switch (set) {
-               case 0:
-                       cdt->cdt_policy &= ~policy;
-                       break;
-               case 1:
-                       cdt->cdt_policy |= policy;
-                       break;
-               case 2:
-                       cdt->cdt_policy = policy;
-                       break;
-               }
+       } while (start != NULL);
+
+       CDEBUG(D_HSM, "%s: new policy: rm="LPX64" add="LPX64" set="LPX64"\n",
+              mdt_obd_name(mdt), remove_mask, add_mask, set_mask);
 
-               start = end + 1;
-       } while (end != NULL);
+       /* if no sign in all string, it is a clear and set
+        * if some sign found, all unsigned are converted
+        * to add
+        * P1 P2 = set to P1 and P2
+        * P1 -P2 = add P1 clear P2 same as +P1 -P2
+        */
+       if (remove_mask == 0 && add_mask == 0) {
+               cdt->cdt_policy = set_mask;
+       } else {
+               cdt->cdt_policy |= set_mask | add_mask;
+               cdt->cdt_policy &= ~remove_mask;
+       }
+
+       GOTO(out, rc = count);
+
+out:
        OBD_FREE(buf, count + 1);
-       RETURN(count);
+       RETURN(rc);
 }
+LPROC_SEQ_FOPS(mdt_hsm_policy);
 
 #define GENERATE_PROC_METHOD(VAR)                                      \
-static int lprocfs_rd_hsm_##VAR(char *page, char **start, off_t off,   \
-                               int count, int *eof, void *data)        \
+static int mdt_hsm_##VAR##_seq_show(struct seq_file *m, void *data)    \
 {                                                                      \
-       struct mdt_device       *mdt = data;                            \
+       struct mdt_device       *mdt = m->private;                      \
        struct coordinator      *cdt = &mdt->mdt_coordinator;           \
-       int                      sz;                                    \
        ENTRY;                                                          \
                                                                        \
-       sz = snprintf(page, count, LPU64"\n", (__u64)cdt->VAR);         \
-       *eof = 1;                                                       \
-       RETURN(sz);                                                     \
+       seq_printf(m, LPU64"\n", (__u64)cdt->VAR);                      \
+       RETURN(0);                                                      \
 }                                                                      \
-static int lprocfs_wr_hsm_##VAR(struct file *file, const char *buffer, \
-                               unsigned long count, void *data)        \
+static ssize_t                                                         \
+mdt_hsm_##VAR##_seq_write(struct file *file, const char __user *buffer,        \
+                         size_t count, loff_t *off)                    \
                                                                        \
 {                                                                      \
-       struct mdt_device       *mdt = data;                            \
+       struct seq_file         *m = file->private_data;                \
+       struct mdt_device       *mdt = m->private;                      \
        struct coordinator      *cdt = &mdt->mdt_coordinator;           \
        int                      val;                                   \
        int                      rc;                                    \
@@ -1891,12 +1949,13 @@ static int lprocfs_wr_hsm_##VAR(struct file *file, const char *buffer,  \
                RETURN(count);                                          \
        }                                                               \
        RETURN(-EINVAL);                                                \
-}
+}                                                                      \
 
 GENERATE_PROC_METHOD(cdt_loop_period)
-GENERATE_PROC_METHOD(cdt_delay)
-GENERATE_PROC_METHOD(cdt_timeout)
-GENERATE_PROC_METHOD(cdt_max_request)
+GENERATE_PROC_METHOD(cdt_grace_delay)
+GENERATE_PROC_METHOD(cdt_active_req_timeout)
+GENERATE_PROC_METHOD(cdt_max_requests)
+GENERATE_PROC_METHOD(cdt_default_archive_id)
 
 /*
  * procfs write method for MDT/hsm_control
@@ -1907,32 +1966,59 @@ GENERATE_PROC_METHOD(cdt_max_request)
 #define CDT_DISABLE_CMD  "disabled"
 #define CDT_PURGE_CMD    "purge"
 #define CDT_HELP_CMD     "help"
+#define CDT_MAX_CMD_LEN  10
 
-int lprocfs_wr_hsm_cdt_control(struct file *file, const char *buffer,
-                              unsigned long count, void *data)
+ssize_t
+mdt_hsm_cdt_control_seq_write(struct file *file, const char __user *buffer,
+                             size_t count, loff_t *off)
 {
-       struct obd_device       *obd = data;
+       struct seq_file         *m = file->private_data;
+       struct obd_device       *obd = m->private;
        struct mdt_device       *mdt = mdt_dev(obd->obd_lu_dev);
        struct coordinator      *cdt = &(mdt->mdt_coordinator);
        int                      rc, usage = 0;
+       char                     kernbuf[CDT_MAX_CMD_LEN];
        ENTRY;
 
+       if (count == 0 || count >= sizeof(kernbuf))
+               RETURN(-EINVAL);
+
+       if (copy_from_user(kernbuf, buffer, count))
+               RETURN(-EFAULT);
+       kernbuf[count] = 0;
+
+       if (kernbuf[count - 1] == '\n')
+               kernbuf[count - 1] = 0;
+
        rc = 0;
-       if (strncmp(buffer, CDT_ENABLE_CMD, strlen(CDT_ENABLE_CMD)) == 0) {
+       if (strcmp(kernbuf, CDT_ENABLE_CMD) == 0) {
                if (cdt->cdt_state == CDT_DISABLE) {
                        cdt->cdt_state = CDT_RUNNING;
                        mdt_hsm_cdt_wakeup(mdt);
                } else {
                        rc = mdt_hsm_cdt_start(mdt);
                }
-       } else if (strncmp(buffer, CDT_STOP_CMD, strlen(CDT_STOP_CMD)) == 0) {
-               cdt->cdt_state = CDT_STOPPING;
-       } else if (strncmp(buffer, CDT_DISABLE_CMD,
-                          strlen(CDT_DISABLE_CMD)) == 0) {
-               cdt->cdt_state = CDT_DISABLE;
-       } else if (strncmp(buffer, CDT_PURGE_CMD, strlen(CDT_PURGE_CMD)) == 0) {
+       } else if (strcmp(kernbuf, CDT_STOP_CMD) == 0) {
+               if ((cdt->cdt_state == CDT_STOPPING) ||
+                   (cdt->cdt_state == CDT_STOPPED)) {
+                       CERROR("%s: Coordinator already stopped\n",
+                              mdt_obd_name(mdt));
+                       rc = -EALREADY;
+               } else {
+                       cdt->cdt_state = CDT_STOPPING;
+               }
+       } else if (strcmp(kernbuf, CDT_DISABLE_CMD) == 0) {
+               if ((cdt->cdt_state == CDT_STOPPING) ||
+                   (cdt->cdt_state == CDT_STOPPED)) {
+                       CERROR("%s: Coordinator is stopped\n",
+                              mdt_obd_name(mdt));
+                       rc = -EINVAL;
+               } else {
+                       cdt->cdt_state = CDT_DISABLE;
+               }
+       } else if (strcmp(kernbuf, CDT_PURGE_CMD) == 0) {
                rc = hsm_cancel_all_actions(mdt);
-       } else if (strncmp(buffer, CDT_HELP_CMD, strlen(CDT_HELP_CMD)) == 0) {
+       } else if (strcmp(kernbuf, CDT_HELP_CMD) == 0) {
                usage = 1;
        } else {
                usage = 1;
@@ -1951,51 +2037,206 @@ int lprocfs_wr_hsm_cdt_control(struct file *file, const char *buffer,
        RETURN(count);
 }
 
-int lprocfs_rd_hsm_cdt_control(char *page, char **start, off_t off,
-                              int count, int *eof, void *data)
+int mdt_hsm_cdt_control_seq_show(struct seq_file *m, void *data)
 {
-       struct obd_device       *obd = data;
+       struct obd_device       *obd = m->private;
        struct coordinator      *cdt;
-       int                      sz;
        ENTRY;
 
        cdt = &(mdt_dev(obd->obd_lu_dev)->mdt_coordinator);
-       *eof = 1;
 
        if (cdt->cdt_state == CDT_INIT)
-               sz = snprintf(page, count, "init\n");
+               seq_printf(m, "init\n");
        else if (cdt->cdt_state == CDT_RUNNING)
-               sz = snprintf(page, count, "enabled\n");
+               seq_printf(m, "enabled\n");
        else if (cdt->cdt_state == CDT_STOPPING)
-               sz = snprintf(page, count, "stopping\n");
+               seq_printf(m, "stopping\n");
        else if (cdt->cdt_state == CDT_STOPPED)
-               sz = snprintf(page, count, "stopped\n");
+               seq_printf(m, "stopped\n");
        else if (cdt->cdt_state == CDT_DISABLE)
-               sz = snprintf(page, count, "disabled\n");
+               seq_printf(m, "disabled\n");
        else
-               sz = snprintf(page, count, "unknown\n");
+               seq_printf(m, "unknown\n");
+
+       RETURN(0);
+}
+
+static int
+mdt_hsm_request_mask_show(struct seq_file *m, __u64 mask)
+{
+       int i, rc = 0;
+       ENTRY;
+
+       for (i = 0; i < 8 * sizeof(mask); i++) {
+               if (mask & (1UL << i))
+                       rc += seq_printf(m, "%s%s", rc == 0 ? "" : " ",
+                                       hsm_copytool_action2name(i));
+       }
+       rc += seq_printf(m, "\n");
+
+       RETURN(rc);
+}
+
+static int
+mdt_hsm_user_request_mask_seq_show(struct seq_file *m, void *data)
+{
+       struct mdt_device *mdt = m->private;
+       struct coordinator *cdt = &mdt->mdt_coordinator;
+
+       return mdt_hsm_request_mask_show(m, cdt->cdt_user_request_mask);
+}
+
+static int
+mdt_hsm_group_request_mask_seq_show(struct seq_file *m, void *data)
+{
+       struct mdt_device *mdt = m->private;
+       struct coordinator *cdt = &mdt->mdt_coordinator;
+
+       return mdt_hsm_request_mask_show(m, cdt->cdt_group_request_mask);
+}
+
+static int
+mdt_hsm_other_request_mask_seq_show(struct seq_file *m, void *data)
+{
+       struct mdt_device *mdt = m->private;
+       struct coordinator *cdt = &mdt->mdt_coordinator;
+
+       return mdt_hsm_request_mask_show(m, cdt->cdt_other_request_mask);
+}
+
+static inline enum hsm_copytool_action
+hsm_copytool_name2action(const char *name)
+{
+       if (strcasecmp(name, "NOOP") == 0)
+               return HSMA_NONE;
+       else if (strcasecmp(name, "ARCHIVE") == 0)
+               return HSMA_ARCHIVE;
+       else if (strcasecmp(name, "RESTORE") == 0)
+               return HSMA_RESTORE;
+       else if (strcasecmp(name, "REMOVE") == 0)
+               return HSMA_REMOVE;
+       else if (strcasecmp(name, "CANCEL") == 0)
+               return HSMA_CANCEL;
+       else
+               return -1;
+}
+
+static ssize_t
+mdt_write_hsm_request_mask(struct file *file, const char __user *user_buf,
+                           size_t user_count, __u64 *mask)
+{
+       char *buf, *pos, *name;
+       size_t buf_size;
+       __u64 new_mask = 0;
+       int rc;
+       ENTRY;
+
+       if (!(user_count < 4096))
+               RETURN(-ENOMEM);
+
+       buf_size = user_count + 1;
+
+       OBD_ALLOC(buf, buf_size);
+       if (buf == NULL)
+               RETURN(-ENOMEM);
+
+       if (copy_from_user(buf, user_buf, buf_size - 1))
+               GOTO(out, rc = -EFAULT);
+
+       buf[buf_size - 1] = '\0';
+
+       pos = buf;
+       while ((name = strsep(&pos, " \t\v\n")) != NULL) {
+               int action;
+
+               if (*name == '\0')
+                       continue;
+
+               action = hsm_copytool_name2action(name);
+               if (action < 0)
+                       GOTO(out, rc = -EINVAL);
+
+               new_mask |= (1UL << action);
+       }
+
+       *mask = new_mask;
+       rc = user_count;
+out:
+       OBD_FREE(buf, buf_size);
+
+       RETURN(rc);
+}
+
+static ssize_t
+mdt_hsm_user_request_mask_seq_write(struct file *file, const char __user *buf,
+                                       size_t count, loff_t *off)
+{
+       struct seq_file         *m = file->private_data;
+       struct mdt_device       *mdt = m->private;
+       struct coordinator *cdt = &mdt->mdt_coordinator;
+
+       return mdt_write_hsm_request_mask(file, buf, count,
+                                          &cdt->cdt_user_request_mask);
+}
+
+static ssize_t
+mdt_hsm_group_request_mask_seq_write(struct file *file, const char __user *buf,
+                                       size_t count, loff_t *off)
+{
+       struct seq_file         *m = file->private_data;
+       struct mdt_device       *mdt = m->private;
+       struct coordinator      *cdt = &mdt->mdt_coordinator;
+
+       return mdt_write_hsm_request_mask(file, buf, count,
+                                          &cdt->cdt_group_request_mask);
+}
+
+static ssize_t
+mdt_hsm_other_request_mask_seq_write(struct file *file, const char __user *buf,
+                                       size_t count, loff_t *off)
+{
+       struct seq_file         *m = file->private_data;
+       struct mdt_device       *mdt = m->private;
+       struct coordinator      *cdt = &mdt->mdt_coordinator;
 
-       RETURN(sz);
+       return mdt_write_hsm_request_mask(file, buf, count,
+                                          &cdt->cdt_other_request_mask);
 }
 
-static struct lprocfs_vars lprocfs_mdt_hsm_vars[] = {
-       { "agents",             NULL, NULL, NULL, &mdt_hsm_agent_fops, 0 },
-       { "agent_actions",      NULL, NULL, NULL,
-                               &mdt_agent_actions_fops, 0444 },
-       { "grace_delay",        lprocfs_rd_hsm_cdt_delay,
-                               lprocfs_wr_hsm_cdt_delay,
-                               NULL, NULL, 0 },
-       { "loop_period",        lprocfs_rd_hsm_cdt_loop_period,
-                               lprocfs_wr_hsm_cdt_loop_period,
-                               NULL, NULL, 0 },
-       { "max_requests",       lprocfs_rd_hsm_cdt_max_request,
-                               lprocfs_wr_hsm_cdt_max_request,
-                               NULL, NULL, 0 },
-       { "policy",             lprocfs_rd_hsm_policy, lprocfs_wr_hsm_policy,
-                               NULL, NULL, 0 },
-       { "request_timeout",    lprocfs_rd_hsm_cdt_timeout,
-                               lprocfs_wr_hsm_cdt_timeout,
-                               NULL, NULL, 0 },
-       { "requests",           NULL, NULL, NULL, &mdt_hsm_request_fops, 0 },
+LPROC_SEQ_FOPS(mdt_hsm_cdt_loop_period);
+LPROC_SEQ_FOPS(mdt_hsm_cdt_grace_delay);
+LPROC_SEQ_FOPS(mdt_hsm_cdt_active_req_timeout);
+LPROC_SEQ_FOPS(mdt_hsm_cdt_max_requests);
+LPROC_SEQ_FOPS(mdt_hsm_cdt_default_archive_id);
+LPROC_SEQ_FOPS(mdt_hsm_user_request_mask);
+LPROC_SEQ_FOPS(mdt_hsm_group_request_mask);
+LPROC_SEQ_FOPS(mdt_hsm_other_request_mask);
+
+static struct lprocfs_seq_vars lprocfs_mdt_hsm_vars[] = {
+       { .name =       "agents",
+         .fops =       &mdt_hsm_agent_fops                     },
+       { .name =       "actions",
+         .fops =       &mdt_hsm_actions_fops,
+         .proc_mode =  0444                                    },
+       { .name =       "default_archive_id",
+         .fops =       &mdt_hsm_cdt_default_archive_id_fops    },
+       { .name =       "grace_delay",
+         .fops =       &mdt_hsm_cdt_grace_delay_fops           },
+       { .name =       "loop_period",
+         .fops =       &mdt_hsm_cdt_loop_period_fops           },
+       { .name =       "max_requests",
+         .fops =       &mdt_hsm_cdt_max_requests_fops          },
+       { .name =       "policy",
+         .fops =       &mdt_hsm_policy_fops                    },
+       { .name =       "active_request_timeout",
+         .fops =       &mdt_hsm_cdt_active_req_timeout_fops    },
+       { .name =       "active_requests",
+         .fops =       &mdt_hsm_active_requests_fops           },
+       { .name =       "user_request_mask",
+         .fops =       &mdt_hsm_user_request_mask_fops,        },
+       { .name =       "group_request_mask",
+         .fops =       &mdt_hsm_group_request_mask_fops,       },
+       { .name =       "other_request_mask",
+         .fops =       &mdt_hsm_other_request_mask_fops,       },
        { 0 }
 };