* Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2011, 2012, Whamcloud, Inc.
+ * Copyright (c) 2011, 2012, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
{ MDS_WRITEPAGE, "mds_writepage" },
{ MDS_IS_SUBDIR, "mds_is_subdir" },
{ MDS_GET_INFO, "mds_get_info" },
+ { MDS_HSM_STATE_GET, "mds_hsm_state_get" },
+ { MDS_HSM_STATE_SET, "mds_hsm_state_set" },
+ { MDS_HSM_ACTION, "mds_hsm_action" },
+ { MDS_HSM_PROGRESS, "mds_hsm_progress" },
+ { MDS_HSM_REQUEST, "mds_hsm_request" },
+ { MDS_HSM_CT_REGISTER, "mds_hsm_ct_register" },
+ { MDS_HSM_CT_UNREGISTER, "mds_hsm_ct_unregister" },
+ { MDS_SWAP_LAYOUTS, "mds_swap_layouts" },
{ LDLM_ENQUEUE, "ldlm_enqueue" },
{ LDLM_CONVERT, "ldlm_convert" },
{ LDLM_CANCEL, "ldlm_cancel" },
{ OBD_PING, "obd_ping" },
{ OBD_LOG_CANCEL, "llog_origin_handle_cancel" },
{ OBD_QC_CALLBACK, "obd_quota_callback" },
+ { OBD_IDX_READ, "dt_index_read" },
{ LLOG_ORIGIN_HANDLE_CREATE, "llog_origin_handle_create" },
{ LLOG_ORIGIN_HANDLE_NEXT_BLOCK, "llog_origin_handle_next_block" },
{ LLOG_ORIGIN_HANDLE_READ_HEADER,"llog_origin_handle_read_header" },
{ SEC_CTX_INIT, "sec_ctx_init" },
{ SEC_CTX_INIT_CONT,"sec_ctx_init_cont" },
{ SEC_CTX_FINI, "sec_ctx_fini" },
- { FLD_QUERY, "fld_query" }
+ { FLD_QUERY, "fld_query" },
+ { UPDATE_OBJ, "update_obj" },
};
struct ll_eopcode {
if (val > cfs_num_physpages/(2 * bufpages))
return -ERANGE;
- cfs_spin_lock(&svc->srv_lock);
+ spin_lock(&svc->srv_lock);
if (val == 0)
svc->srv_hist_nrqbds_cpt_max = 0;
else
svc->srv_hist_nrqbds_cpt_max = max(1, (val / svc->srv_ncpts));
- cfs_spin_unlock(&svc->srv_lock);
+ spin_unlock(&svc->srv_lock);
return count;
}
if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
return -ERANGE;
- cfs_spin_lock(&svc->srv_lock);
+ spin_lock(&svc->srv_lock);
if (val > svc->srv_nthrs_cpt_limit * svc->srv_ncpts) {
- cfs_spin_unlock(&svc->srv_lock);
+ spin_unlock(&svc->srv_lock);
return -ERANGE;
}
svc->srv_nthrs_cpt_init = val / svc->srv_ncpts;
- cfs_spin_unlock(&svc->srv_lock);
+ spin_unlock(&svc->srv_lock);
return count;
}
int total = 0;
int i;
- LASSERT(svc->srv_parts != NULL);
ptlrpc_service_for_each_part(svcpt, i, svc)
total += svcpt->scp_nthrs_running;
if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
return -ERANGE;
- cfs_spin_lock(&svc->srv_lock);
+ spin_lock(&svc->srv_lock);
if (val < svc->srv_nthrs_cpt_init * svc->srv_ncpts) {
- cfs_spin_unlock(&svc->srv_lock);
+ spin_unlock(&svc->srv_lock);
return -ERANGE;
}
svc->srv_nthrs_cpt_limit = val / svc->srv_ncpts;
- cfs_spin_unlock(&svc->srv_lock);
+ spin_unlock(&svc->srv_lock);
return count;
}
+/**
+ * \addtogoup nrs
+ * @{
+ */
+extern struct nrs_core nrs_core;
+
+/**
+ * Translates \e ptlrpc_nrs_pol_state values to human-readable strings.
+ *
+ * \param[in] state The policy state
+ */
+static const char *
+nrs_state2str(enum ptlrpc_nrs_pol_state state)
+{
+ switch (state) {
+ default:
+ LBUG();
+ case NRS_POL_STATE_INVALID:
+ return "invalid";
+ case NRS_POL_STATE_UNAVAIL:
+ return "unavail";
+ case NRS_POL_STATE_STOPPED:
+ return "stopped";
+ case NRS_POL_STATE_STOPPING:
+ return "stopping";
+ case NRS_POL_STATE_STARTING:
+ return "starting";
+ case NRS_POL_STATE_STARTED:
+ return "started";
+ }
+}
+
+/**
+ * Obtains status information for \a policy.
+ *
+ * Information is copied in \a info.
+ *
+ * \param[in] policy The policy
+ * \param[out] info Holds returned status information
+ */
+void
+nrs_policy_get_info_locked(struct ptlrpc_nrs_policy *policy,
+ struct ptlrpc_nrs_pol_info *info)
+{
+ LASSERT(policy != NULL);
+ LASSERT(info != NULL);
+ LASSERT(spin_is_locked(&policy->pol_nrs->nrs_lock));
+
+ memcpy(info->pi_name, policy->pol_name, NRS_POL_NAME_MAX);
+
+ info->pi_fallback = !!(policy->pol_flags & PTLRPC_NRS_FL_FALLBACK);
+ info->pi_state = policy->pol_state;
+ /**
+ * XXX: These are accessed without holding
+ * ptlrpc_service_part::scp_req_lock.
+ */
+ info->pi_req_queued = policy->pol_req_queued;
+ info->pi_req_started = policy->pol_req_started;
+}
+
+/**
+ * Reads and prints policy status information for all policies of a PTLRPC
+ * service.
+ */
+static int
+ptlrpc_lprocfs_rd_nrs(char *page, char **start, off_t off,
+ int count, int *eof, void *data)
+{
+ struct ptlrpc_service *svc = data;
+ struct ptlrpc_service_part *svcpt;
+ struct ptlrpc_nrs *nrs;
+ struct ptlrpc_nrs_policy *policy;
+ struct ptlrpc_nrs_pol_info *infos;
+ struct ptlrpc_nrs_pol_info tmp;
+ unsigned num_pols;
+ unsigned pol_idx = 0;
+ bool hp = false;
+ int i;
+ int rc = 0;
+ int rc2 = 0;
+ ENTRY;
+
+ /**
+ * Serialize NRS core lprocfs operations with policy registration/
+ * unregistration.
+ */
+ mutex_lock(&nrs_core.nrs_mutex);
+
+ /**
+ * Use the first service partition's regular NRS head in order to obtain
+ * the number of policies registered with NRS heads of this service. All
+ * service partitions will have the same number of policies.
+ */
+ nrs = nrs_svcpt2nrs(svc->srv_parts[0], false);
+
+ spin_lock(&nrs->nrs_lock);
+ num_pols = svc->srv_parts[0]->scp_nrs_reg.nrs_num_pols;
+ spin_unlock(&nrs->nrs_lock);
+
+ OBD_ALLOC(infos, num_pols * sizeof(*infos));
+ if (infos == NULL)
+ GOTO(out, rc = -ENOMEM);
+again:
+
+ ptlrpc_service_for_each_part(svcpt, i, svc) {
+ nrs = nrs_svcpt2nrs(svcpt, hp);
+ spin_lock(&nrs->nrs_lock);
+
+ pol_idx = 0;
+
+ cfs_list_for_each_entry(policy, &nrs->nrs_policy_list,
+ pol_list) {
+ LASSERT(pol_idx < num_pols);
+
+ nrs_policy_get_info_locked(policy, &tmp);
+ /**
+ * Copy values when handling the first service
+ * partition.
+ */
+ if (i == 0) {
+ memcpy(infos[pol_idx].pi_name, tmp.pi_name,
+ NRS_POL_NAME_MAX);
+ memcpy(&infos[pol_idx].pi_state, &tmp.pi_state,
+ sizeof(tmp.pi_state));
+ infos[pol_idx].pi_fallback = tmp.pi_fallback;
+ /**
+ * For the rest of the service partitions
+ * sanity-check the values we get.
+ */
+ } else {
+ LASSERT(strncmp(infos[pol_idx].pi_name,
+ tmp.pi_name,
+ NRS_POL_NAME_MAX) == 0);
+ /**
+ * Not asserting ptlrpc_nrs_pol_info::pi_state,
+ * because it may be different between
+ * instances of the same policy in different
+ * service partitions.
+ */
+ LASSERT(infos[pol_idx].pi_fallback ==
+ tmp.pi_fallback);
+ }
+
+ infos[pol_idx].pi_req_queued += tmp.pi_req_queued;
+ infos[pol_idx].pi_req_started += tmp.pi_req_started;
+
+ pol_idx++;
+ }
+ spin_unlock(&nrs->nrs_lock);
+ }
+
+ /**
+ * Policy status information output is in YAML format.
+ * For example:
+ *
+ * regular_requests:
+ * - name: fifo
+ * state: started
+ * fallback: yes
+ * queued: 0
+ * active: 0
+ *
+ * - name: crrn
+ * state: started
+ * fallback: no
+ * queued: 2015
+ * active: 384
+ *
+ * high_priority_requests:
+ * - name: fifo
+ * state: started
+ * fallback: yes
+ * queued: 0
+ * active: 2
+ *
+ * - name: crrn
+ * state: stopped
+ * fallback: no
+ * queued: 0
+ * active: 0
+ */
+ rc2 = snprintf(page + rc, count - rc,
+ "%s\n", !hp ?
+ "\nregular_requests:" :
+ "high_priority_requests:");
+
+ if (rc2 >= count - rc) {
+ /** Output was truncated */
+ GOTO(out, rc = -EFBIG);
+ }
+
+ rc += rc2;
+
+ for (pol_idx = 0; pol_idx < num_pols; pol_idx++) {
+ rc2 = snprintf(page + rc, count - rc,
+ " - name: %s\n"
+ " state: %s\n"
+ " fallback: %s\n"
+ " queued: %-20d\n"
+ " active: %-20d\n\n",
+ infos[pol_idx].pi_name,
+ nrs_state2str(infos[pol_idx].pi_state),
+ infos[pol_idx].pi_fallback ? "yes" : "no",
+ (int)infos[pol_idx].pi_req_queued,
+ (int)infos[pol_idx].pi_req_started);
+
+
+ if (rc2 >= count - rc) {
+ /** Output was truncated */
+ GOTO(out, rc = -EFBIG);
+ }
+
+ rc += rc2;
+ }
+
+ if (!hp && nrs_svc_has_hp(svc)) {
+ memset(infos, 0, num_pols * sizeof(*infos));
+
+ /**
+ * Redo the processing for the service's HP NRS heads' policies.
+ */
+ hp = true;
+ goto again;
+ }
+
+ *eof = 1;
+
+out:
+ if (infos)
+ OBD_FREE(infos, num_pols * sizeof(*infos));
+
+ mutex_unlock(&nrs_core.nrs_mutex);
+
+ RETURN(rc);
+}
+
+/**
+ * The longest valid command string is the maxium policy name size, plus the
+ * length of the " reg" substring
+ */
+#define LPROCFS_NRS_WR_MAX_CMD (NRS_POL_NAME_MAX + sizeof(" reg") - 1)
+
+/**
+ * Starts and stops a given policy on a PTLRPC service.
+ *
+ * Commands consist of the policy name, followed by an optional [reg|hp] token;
+ * if the optional token is omitted, the operation is performed on both the
+ * regular and high-priority (if the service has one) NRS head.
+ */
+static int
+ptlrpc_lprocfs_wr_nrs(struct file *file, const char *buffer,
+ unsigned long count, void *data)
+{
+ struct ptlrpc_service *svc = data;
+ enum ptlrpc_nrs_queue_type queue = PTLRPC_NRS_QUEUE_BOTH;
+ char *cmd;
+ char *cmd_copy = NULL;
+ char *token;
+ int rc = 0;
+ ENTRY;
+
+ if (count >= LPROCFS_NRS_WR_MAX_CMD)
+ GOTO(out, rc = -EINVAL);
+
+ OBD_ALLOC(cmd, LPROCFS_NRS_WR_MAX_CMD);
+ if (cmd == NULL)
+ GOTO(out, rc = -ENOMEM);
+ /**
+ * strsep() modifies its argument, so keep a copy
+ */
+ cmd_copy = cmd;
+
+ if (cfs_copy_from_user(cmd, buffer, count))
+ GOTO(out, rc = -EFAULT);
+
+ cmd[count] = '\0';
+
+ token = strsep(&cmd, " ");
+
+ if (strlen(token) > NRS_POL_NAME_MAX - 1)
+ GOTO(out, rc = -EINVAL);
+
+ /**
+ * No [reg|hp] token has been specified
+ */
+ if (cmd == NULL)
+ goto default_queue;
+
+ /**
+ * The second token is either NULL, or an optional [reg|hp] string
+ */
+ if (strcmp(cmd, "reg") == 0)
+ queue = PTLRPC_NRS_QUEUE_REG;
+ else if (strcmp(cmd, "hp") == 0)
+ queue = PTLRPC_NRS_QUEUE_HP;
+ else
+ GOTO(out, rc = -EINVAL);
+
+default_queue:
+
+ if (queue == PTLRPC_NRS_QUEUE_HP && !nrs_svc_has_hp(svc))
+ GOTO(out, rc = -ENODEV);
+ else if (queue == PTLRPC_NRS_QUEUE_BOTH && !nrs_svc_has_hp(svc))
+ queue = PTLRPC_NRS_QUEUE_REG;
+
+ /**
+ * Serialize NRS core lprocfs operations with policy registration/
+ * unregistration.
+ */
+ mutex_lock(&nrs_core.nrs_mutex);
+
+ rc = ptlrpc_nrs_policy_control(svc, queue, token, PTLRPC_NRS_CTL_START,
+ false, NULL);
+
+ mutex_unlock(&nrs_core.nrs_mutex);
+out:
+ if (cmd_copy)
+ OBD_FREE(cmd_copy, LPROCFS_NRS_WR_MAX_CMD);
+
+ RETURN(rc < 0 ? rc : count);
+}
+
+/** @} nrs */
+
struct ptlrpc_srh_iterator {
int srhi_idx;
__u64 srhi_seq;
* Since the service history is LRU (i.e. culled reqs will
* be near the head), we shouldn't have to do long
* re-scans */
- LASSERT (srhi->srhi_seq == srhi->srhi_req->rq_history_seq);
+ LASSERTF(srhi->srhi_seq == srhi->srhi_req->rq_history_seq,
+ "%s:%d: seek seq "LPU64", request seq "LPU64"\n",
+ svcpt->scp_service->srv_name, svcpt->scp_cpt,
+ srhi->srhi_seq, srhi->srhi_req->rq_history_seq);
LASSERTF(!cfs_list_empty(&svcpt->scp_hist_reqs),
"%s:%d: seek offset "LPU64", request seq "LPU64", "
"last culled "LPU64"\n",
return -ENOENT;
}
+/*
+ * ptlrpc history sequence is used as "position" of seq_file, in some case,
+ * seq_read() will increase "position" to indicate reading the next
+ * element, however, low bits of history sequence are reserved for CPT id
+ * (check the details from comments before ptlrpc_req_add_history), which
+ * means seq_read() might change CPT id of history sequence and never
+ * finish reading of requests on a CPT. To make it work, we have to shift
+ * CPT id to high bits and timestamp to low bits, so seq_read() will only
+ * increase timestamp which can correctly indicate the next position.
+ */
+
+/* convert seq_file pos to cpt */
+#define PTLRPC_REQ_POS2CPT(svc, pos) \
+ ((svc)->srv_cpt_bits == 0 ? 0 : \
+ (__u64)(pos) >> (64 - (svc)->srv_cpt_bits))
+
+/* make up seq_file pos from cpt */
+#define PTLRPC_REQ_CPT2POS(svc, cpt) \
+ ((svc)->srv_cpt_bits == 0 ? 0 : \
+ (cpt) << (64 - (svc)->srv_cpt_bits))
+
+/* convert sequence to position */
+#define PTLRPC_REQ_SEQ2POS(svc, seq) \
+ ((svc)->srv_cpt_bits == 0 ? (seq) : \
+ ((seq) >> (svc)->srv_cpt_bits) | \
+ ((seq) << (64 - (svc)->srv_cpt_bits)))
+
+/* convert position to sequence */
+#define PTLRPC_REQ_POS2SEQ(svc, pos) \
+ ((svc)->srv_cpt_bits == 0 ? (pos) : \
+ ((__u64)(pos) << (svc)->srv_cpt_bits) | \
+ ((__u64)(pos) >> (64 - (svc)->srv_cpt_bits)))
+
static void *
ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
{
struct ptlrpc_service *svc = s->private;
struct ptlrpc_service_part *svcpt;
struct ptlrpc_srh_iterator *srhi;
+ unsigned int cpt;
int rc;
int i;
+ if (sizeof(loff_t) != sizeof(__u64)) { /* can't support */
+ CWARN("Failed to read request history because size of loff_t "
+ "%d can't match size of u64\n", (int)sizeof(loff_t));
+ return NULL;
+ }
+
OBD_ALLOC(srhi, sizeof(*srhi));
if (srhi == NULL)
return NULL;
srhi->srhi_seq = 0;
srhi->srhi_req = NULL;
- ptlrpc_service_for_each_part(svcpt, i, svc) {
- srhi->srhi_idx = i;
+ cpt = PTLRPC_REQ_POS2CPT(svc, *pos);
- cfs_spin_lock(&svcpt->scp_lock);
- rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, *pos);
- cfs_spin_unlock(&svcpt->scp_lock);
+ ptlrpc_service_for_each_part(svcpt, i, svc) {
+ if (i < cpt) /* skip */
+ continue;
+ if (i > cpt) /* make up the lowest position for this CPT */
+ *pos = PTLRPC_REQ_CPT2POS(svc, i);
+
+ spin_lock(&svcpt->scp_lock);
+ rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi,
+ PTLRPC_REQ_POS2SEQ(svc, *pos));
+ spin_unlock(&svcpt->scp_lock);
if (rc == 0) {
- *pos = srhi->srhi_seq;
+ *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
+ srhi->srhi_idx = i;
return srhi;
}
}
struct ptlrpc_service *svc = s->private;
struct ptlrpc_srh_iterator *srhi = iter;
struct ptlrpc_service_part *svcpt;
- int rc = 0;
+ __u64 seq;
+ int rc;
int i;
for (i = srhi->srhi_idx; i < svc->srv_ncpts; i++) {
svcpt = svc->srv_parts[i];
- srhi->srhi_idx = i;
+ if (i > srhi->srhi_idx) { /* reset iterator for a new CPT */
+ srhi->srhi_req = NULL;
+ seq = srhi->srhi_seq = 0;
+ } else { /* the next sequence */
+ seq = srhi->srhi_seq + (1 << svc->srv_cpt_bits);
+ }
- cfs_spin_lock(&svcpt->scp_lock);
- rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, *pos + 1);
- cfs_spin_unlock(&svcpt->scp_lock);
- if (rc == 0)
- break;
+ spin_lock(&svcpt->scp_lock);
+ rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, seq);
+ spin_unlock(&svcpt->scp_lock);
+ if (rc == 0) {
+ *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
+ srhi->srhi_idx = i;
+ return srhi;
+ }
}
- if (rc != 0) {
- OBD_FREE(srhi, sizeof(*srhi));
- return NULL;
- }
-
- *pos = srhi->srhi_seq;
- return srhi;
+ OBD_FREE(srhi, sizeof(*srhi));
+ return NULL;
}
/* common ost/mdt so_req_printer */
svcpt = svc->srv_parts[srhi->srhi_idx];
- cfs_spin_lock(&svcpt->scp_lock);
+ spin_lock(&svcpt->scp_lock);
rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq);
svc->srv_ops.so_req_printer(s, srhi->srhi_req);
}
- cfs_spin_unlock(&svcpt->scp_lock);
+ spin_unlock(&svcpt->scp_lock);
return rc;
}
int rc = 0;
int i;
- LASSERT(svc->srv_parts != NULL);
-
if (AT_OFF) {
rc += snprintf(page + rc, count - rc,
"adaptive timeouts off, using obd_timeout %u\n",
if (val < 0)
return -ERANGE;
- cfs_spin_lock(&svc->srv_lock);
+ spin_lock(&svc->srv_lock);
svc->srv_hpreq_ratio = val;
- cfs_spin_unlock(&svc->srv_lock);
+ spin_unlock(&svc->srv_lock);
return count;
}
{.name = "timeouts",
.read_fptr = ptlrpc_lprocfs_rd_timeouts,
.data = svc},
- {NULL}
+ {.name = "nrs_policies",
+ .read_fptr = ptlrpc_lprocfs_rd_nrs,
+ .write_fptr = ptlrpc_lprocfs_wr_nrs,
+ .data = svc},
+ {NULL}
};
static struct file_operations req_history_fops = {
.owner = THIS_MODULE,
* the proc entries under the being destroyed export{}, so I have
* to drop the lock at first here.
* - jay, jxiong@clusterfs.com */
- class_incref(obd, __FUNCTION__, cfs_current());
LPROCFS_EXIT();
+ class_incref(obd, __FUNCTION__, cfs_current());
if (strncmp(tmpbuf, "nid:", 4) == 0)
obd_export_evict_by_nid(obd, tmpbuf + 4);
else
obd_export_evict_by_uuid(obd, tmpbuf);
+ class_decref(obd, __FUNCTION__, cfs_current());
LPROCFS_ENTRY();
- class_decref(obd, __FUNCTION__, cfs_current());
out:
OBD_FREE(kbuf, BUFLEN);
return -ERANGE;
LPROCFS_CLIMP_CHECK(obd);
- cfs_spin_lock(&imp->imp_lock);
- imp->imp_no_pinger_recover = !val;
- cfs_spin_unlock(&imp->imp_lock);
- LPROCFS_CLIMP_EXIT(obd);
+ spin_lock(&imp->imp_lock);
+ imp->imp_no_pinger_recover = !val;
+ spin_unlock(&imp->imp_lock);
+ LPROCFS_CLIMP_EXIT(obd);
- return count;
+ return count;
}
EXPORT_SYMBOL(lprocfs_wr_pinger_recov);